Your buyers are asking ChatGPT, Claude, and Perplexity how much your product costs.

VisibilityStack analyzed 13,867 pricing questions across 108 SaaS brands and six AI platforms, then compared every answer with the pricing published by the vendor. AI got the complete pricing right only 24% of the time.

But you can fix it.

Why does AI get your pricing wrong 76% of the time?

In 76% of answers, pricing wasn’t fully accurate. Even the starting price was only right 43% of the time, despite usually being the easiest number to find.

So where does AI source information to answer pricing questions from?

Only 29% of the sources AI cites are from the vendor’s own website. About one in five citations points to a competitor. The rest comes from review sites, software directories, comparison pages, Reddit, YouTube, and other sources.

AI is effectively building its own version of your pricing from multiple sources. If you don’t publish your pricing, someone else will.

Among 27 brands in the study that published no pricing at all, AI gave buyers a specific price 81% of the time. Brands that published pricing had 4.1x higher pricing accuracy.

And when AI gets a price wrong, a third of misquotes are too high. That can get you filtered out from a buyer’s shortlist before they even visit your website or speak with your sales team.

So what can you do?

The 8 actions to fix it

To get AI to quote your pricing correctly, do three things: audit what it says today, make your own pricing easier to retrieve, and fix or outweigh outdated information elsewhere.

1. Ask AI what you cost

This is probably the easiest place to start. Open a fresh conversation in ChatGPT, Claude, Perplexity, Gemini and Google (in AI mode) and ask what a prospective buyer would ask.

Copy and paste this prompt:

I’m evaluating [COMPANY]’s [PRODUCT] for a business with [NUMBER OF USERS / EXPECTED USAGE] in [COUNTRY].

How much should I expect to pay? Break down the available plans, monthly and annual costs, what’s included, minimum commitments, potential overage charges, and any additional fees I should budget for.

Then ask a follow-up:

What sources did you use for these prices? Give me the direct links for each price or estimate.

Compare the answers against your actual pricing. Record the incorrect answers and, critically, the sources AI cites. Repeat the same test across multiple AI platforms. That tells you which errors you actually need to fix.

VisibilityStack recommends repeating the same audit every month. In this article, Stefan walks through how you can automate pricing tasks like this with the AI of your choice. Once automated, you can even run it weekly or daily.

2. Make your pricing machine-readable

Your pricing needs to work for two audiences: humans and machines.

Humans get the polished experience. Product cards, pricing calculators, comparison tables, and interactive selectors.

AI should get a machine-readable, crawlable version in plain text: plan names, pricing metrics, included usage, overage rates, and billing rules.

Cloudflare is a good example. Its pricing page is designed for humans, and AI agents can request a text-only index.md.

Give humans the polished pricing experience. Give machines a clean markdown version.

3. Repeat your pricing on your product pages

Many companies have a single pricing page. That simplifies maintaining it, and also makes it easy for buyers to find all pricing information in one place. For AI, it can make it harder to match individual products and features to the correct pricing.

Anthropic does this well. Each model page includes the pricing for that specific model (like this .md file for Fable 5.1), and an overview llms.txt file points AIs to ALL .md files. They also still have ONE consolidated pricing page for humans designed for the buyer (individual, developers, teams, API).

In addition to your pricing page for humans, repeat the relevant pricing wherever you describe an individual product.

4. Structure your pricing data and show when it was updated

Use structured data to make your pricing explicit, and signal the information is current.

The Value Project provides open JSON schemas for machine-readable pricing and value models, all available on GitHub.

And add a dateModified to show when information was last updated (not necessarily when your pricing changed). That gives AI a stronger signal that your page is more current than a years-old comparison article.

5. Fix pricing on review sites like Capterra and G2

VisibilityStack found that every contact-sales brand it checked had pricing published on Capterra or G2 anyway.

Prioritize the directories that already appear in AI answers. Claim the profiles you can control, and audit them after every pricing change. If an outdated number is posted online, AI may treat it as current.

6. Engage creators and Reddit communities

VisibilityStack found that Reddit and YouTube account for 16% of pricing citations. Another study found that the average Reddit thread cited by AI had 33 comments but only 7 upvotes, while 96% were plain-text posts. Discussion matters more than virality.

Find Reddit threads and YouTube videos that mention your product pricing. If they’re wrong or outdated, engage transparently as the vendor and provide the current information.

If creators have already covered your product, give them a reason to revisit it with the current pricing.

7. Publish a YouTube pricing explainer

Create a simple video explaining your plans, actual prices, and who each plan is designed for.

Clay Co-founder Varun Anand and Head of Finance Karan Parekh made this YouTube video explaining their new pricing model when they split out platform actions from their pricing.

8. Explain your pricing changes everywhere

When Clay changed its pricing in March 2026, it explained the change in an 8-minute video, a launch announcement, an internal pricing memo, multiple FAQs, its Community, Clay University, and LinkedIn posts.

That repetition matters. Each becomes another current source AI can retrieve when someone asks why your pricing changed or what an existing customer should expect.

When pricing changes, update your pricing page and talk about it - in multiple pages.

Your pricing strategy is no longer communicated only through your pricing page and your sales team alone. AI is explaining it for you, so make sure it has the right information.

Can your customers commit for a discount or pay as they go?
What if their usage shifts from one team or product to another?
What happens when they have an unusually busy few days?

How you handle these situations shapes your customer’s experience with you.

Here is what 10 SaaS and AI companies do in two areas: choices when customers buy, and flexibility when usage changes.

Give customers choices in how they buy

1. Let customers pick their pricing metric

Decagon gives customers two ways to pay for its AI agents:

• Per conversation: Pay for each incoming conversation.

• Per resolution: Pay a higher unit price for conversations the agent fully resolves.

Decagon says most customers choose per-conversation pricing.

Let buyers choose the metric that fits their preferences for predictability and risk.

2. Let customers choose whether to commit or pay-as-you-go

Salesforce Agentforce offers three ways to buy Flex Credits:

• PayGo: Pay monthly for actual usage at list price, billed in arrears.

• Pre-Commit: Commit to minimum usage for lower rates, without paying upfront. Billing is monthly and customers still owe the minimum if actual usage falls short.

• Pre-Purchase: Buy credits upfront to get the highest discount.

Let customers choose the balance of flexibility, cash flow, and savings that fits their business.

3. Offer pooled credits for teams

Figma lets admins purchase a shared pool of additional AI credits. Approved users draw from it after exhausting their individual seat allowance. Admins can manage access and track team-wide credit usage in a dashboard.

Let teams allocate capacity where it is needed without predicting every person’s usage upfront.

4. Make credits portable across deployments

Palo Alto Networks lets their customers use its Software NGFW credits across their deployments, including different cloud providers like AWS and Azure. Customers can shift allocations as their infrastructure and workloads change.

Allow customers to commit and reduce their risk of committing to the wrong product.

5. Allow dual licensing for migrations

Atlassian’s dual licensing lets eligible customers purchase an annual Cloud subscription and extend their existing Data Center license at a 100% discount for up to six months during the migration.

Customers can migrate in phases without paying full price for both products.

Subscribe now

Add flexibility when their usage changes

6. Let overage slide until it becomes a pattern

Jetpack Stats excludes the two busiest traffic days each month when checking usage against plan limits. Paid customers must upgrade only after they exceed their limit for three consecutive months.

This leniency with clear limits avoids sizing capacity on spikes instead of typical needs.

7. Allow backdating upgrades for higher limits

Acquia Cloud Platform allows customers to backdate the effective date for upgrades to cover prior overages. Customers pay for the larger plan from its effective date, avoiding the overage fees.

This gives customers a way to correct an underestimate without the penalty of an overage.

8. Allow users to reset usage limits in the product

OpenAI allows Plus and Pro users to reset their weekly and 5-hour limits (for Codex, Work) when they need it. In addition to paid resets, they offer promotional resets that can be banked and have an expiration date (which avoids customers accumulating unused resets indefinitely).

These resets help users handle a temporary spike without permanently increasing their plan.

9. Allow credits to roll over a month

Runway allows up to one month of unused credits to roll over on Max and Team plans, while on Standard and Pro they expire monthly. As Rob Litterst points out, they are even making this a reason to upgrade.

A 1-month rollover mitigates the customer’s loss of capacity they already paid for.

10. Give customers spending controls

Vercel lets teams set spending thresholds and choose one or more actions when these are reached: send an alert, trigger a webhook, or pause production deployments. They are also transparent that spending can exceed the threshold before those actions take effect since checks run only every few minutes.

Let your customers choose how to balance service continuity and spending.

Build flexibility into the rules

The pattern across all these examples is flexibility with clear boundaries: Pay as you go (at a higher price), forgiving overages (up to a limit), a way to reset limits (with an expiration date).

Start with the billing rule that creates the most disputes or sales objections. Give customers a clear way to handle it. Measure the effect on adoption, retention, and cost to serve.

Make the rules easy to understand before the customer needs them.

Subscribe now

Keep reading

So how do you pick the right model for your task? Two public benchmark sites score AI models across 24 benchmarks specific to use cases like analysis, reasoning by comparing Quality vs. Cost.

1. LiveBench: cost vs. quality for 7 categories

LiveBeLnch publishes a neat interactive chart (scroll down to “02 Insights”) that plots models on a 2x2 matrix: quality vs. cost.

The logic is simple. Paying more for a model that scores higher on your task can be worth it. Paying more for one that scores lower is not.

Above the chart, you can select one of 7 categories:

1. Reasoning

2. Coding

3. Agentic Coding

4. Mathematics

5. Data Analysis

6. Language

7. Instruction Following (IF)

The insights can save you a lot of money. GPT-6 Astra for coding costs almost triple per successful task to that of GPT-5.6 Sol Max Effort. Its quality score is also lower (80.6 vs. 83.9). LiveBench calls that region the “kill zone.” Select GPT-5.6 Sol (max) and Astra greys out inside its ‘kill zone’: everything that scores lower and costs more.

2. Artificial Analysis: 17 use case benchmarks

Artificial Analysis is the other main player benchmarking cost per task. They define their Artificial Analysis Intelligence Index based on a suite of benchmarks and publish an overall ranking. Even more importantly, they also publish the individual benchmarks.

The “total cost” view is informative, but model choice is always a trade-off: how much outcome do I get at what price? Click on “Intelligence vs. Total Cost“ and the view changes to a 2x2, adding the Artificial Analysis Intelligence Index score as the y-axis.

The per-benchmark pages are even more useful. 17 of them offer the quality vs. cost view.

Business Workflow Benchmarks

1. Analyze spreadsheets and documents → AA-AnalystAgent

2. Execute consulting, investment banking and law tasks → APEX-Agents-AA

3. SaaS workflow automation → AutomationBench-AA

4. Customer service, HR, ITSM → EnterpriseOps-Gym-AA

5. Create spreadsheets, presentations and memos → AA-Briefcase

6. Produce professional deliverables across 44 occupations → GDPval-AA

7. Legal Agent tasks → Harvey LAB-AA

Document Reasoning & Instruction Following Benchmarks

8. Answer questions based on long PDFs → GDP.pdf

9. Long-context reasoning → AA-LCR

10. Precise instruction following → IFBench

11. Medical records reasoning → MLCR-AA

Coding Benchmarks

12. Scientific research code writing → SciCode

13. Coding and terminal tasks → Terminal-Bench

14. Diagnose root cause of Kubernetes incidents → ITBench-AA

Research & Science Benchmarks

15. Answer academic questions → Humanity’s Last Exam

16. Research-level physics problems → CritPt

17. Graduate-level physics, biology, and chemistry questions → GPQA Diamond

Take AA-AnalystAgent. It tests models on 80 analyst questions, each with its own folder of spreadsheets and documents. Gemini 3.7 Flash scores slightly higher than Claude Fable 5.1 (60% vs 58%) at roughly a third of the cost. That is a decision you can act on when choosing which LLM your agent runs on.

How to read and tailor the score vs. cost view

Using AA-AnalystAgent as the example:

1. Open AA-AnalystAgent and scroll down to “Cost”

2. Select Score vs. Cost per Task to compare cost against performance

3. Hover over a model’s dot to see its “cost per task“ and benchmark score

4. Use the models dropdown above the chart to add/remove individual LLMs

Closing Thought

Stop choosing models by published token prices. They’re not comparable between AI providers. Pick the cheapest model that offers the right quality for your use case.

Keep Reading

And with “Pricing team” I mean Pricing strategists. People whose primary job is pricing. Not your deal desk, RevOps, FP&A, or external consultants.

The bad news: there is no one answer that one universal answer. But there are credible market and expert benchmarks. Here they are.

What the benchmarks say

The Professional Pricing Society (PPS) survey of 752 pricing professionals in 2023:

“Respondents from American companies reported an average of 3.58 Pricing FTEs per billion dollars in annual revenue (lower than last year’s survey), while European companies reported an average of 3.73 Pricing FTEs per billion euros in revenue (higher than last year).”

Tim Smith from Wiglaf Pricing shared in August 2025:

“My favorite benchmark is and has been pricing professionals per revenue dollars in the range of 1/$100 to 1/$500 million. I have been using this ratio for the past three years in writing my Pricing Spineometer case studies with little objections in my scoring.”

The Boston Consulting Group (BCG) in 2017 recommended that

“pricing organizations have two to six full-time employees per $1 billion in company revenues. Small companies—those with revenues of $500 million or less—need pricing teams of at least five full-time people. … Midsize companies, with revenues of $1 billion to $5 billion, often need 10 to 12 full-time employees; larger companies will need even more dedicated resources.”

Simon Kucher recommended in March 2024 as a rule of thumb:

“about 40-200 million euros in revenue per FTE. With an increasing team size, …
…the more responsibilities are allocated to it,
…the more deals per year the company closes,
…the more complex the deals are, and
…the larger the deals are on average.”

Complexity decides where you land

Once you have the range based on your revenue, four factors determine whether the size of your pricing team should be on the higher end of that range:

• Commercial complexity: multiple segments, geographies, channels. Sales-led growth needs more discount governance than PLG.

• Monetization complexity: usage-pricing, credits, outcome-based metrics, hybrid models, meter governance, and complex rating structures.

• Change velocity: frequency of pricing changes from rapidly changing AI costs and competitive reference points.

• Scope: Discount analytics and guidance for sales-led growth, and pricing product management.

The more combinations of complexity, high velocity and responsibilities beyond pure pricing strategy, the more headcount you should plan for to maximize EBIT impact.

What I would use

Do not try to get it right by turning those benchmarks into scientific precision. Use revenue to get some guidance on the right range (e.g. 1-6 FTEs for $250M revenue). Then think about complexity, velocity and what you want the Pricing team to own.

A stable, seat-based SaaS company with one product, one geography, and standardized discounting (or PLG) can operate near the lean end. An AI company with frequently changing competitive landscape and AI costs that sells into multiple geographies should staff toward the heavier end, even at the same revenue.

Transaction-level deal authority (i.e. Deal Desk) also requires much more than most pricing teams are staffed for, and those benchmarks do not include those.

Keep Reading

What the experts found, and what they left open

• Simon-Kucher reported in 2024 that pricing has “a pretty even distribution of functional homes across B2B organizations.”

• BCG found the same in 2017. Pricing reports either to the CEO, or to sales or marketing or is embedded in the finance department.

• Homburg, Jensen, and Hahn in “How to Organize Pricing? Vertical Delegation and Horizontal Dispersion of Pricing Authority” found that strategic pricing and transaction-level discount authority need different governance. Delegate too much and margin leaks. Delegate too little and deals stall.

That last point gave me an idea. When overall there is no dominant function, what factors impact where pricing should sit. Here is what I found.

Scope decides the reporting line

In this dataset of job descriptions, the CPO is the most common home for a pricing strategy role at 35%. But that means 65% of the companies put pricing somewhere else. CMO, CFO, CRO, and Strategy EVP are all at roughly 15% each.

Pricing strategy roles that also own deal desk responsibilities report much more often to the CFO (33%) or CRO (27%). A deal desk structures deals, approves discounts, weights trade-offs with payment terms. That work lives next to revenue recognition and quota attainment, so the reporting line follows.

Homburg’s paper called this the difference between strategic authority and transaction-level authority. Companies are organizing exactly that way. If you are hiring a pricing leader, be clear if the job is the former or if it includes the latter too. One word of caution: Very few Pricing experts have experience in both.

An industry-focus moves pricing away from CPO

Industry-agnostic companies put pricing under the CPO at every stage: SmartBear below $100M, Mistral at $400M, OpenAI at $25B.

Industry-specific companies don’t. Below $100M, the most common home is the CRO at 36%. Between $100M and $500M it is the CMO at 36%. Above $500M it lands with Strategy EVP at 26%, with CPO and CFO close behind.

The mechanism I suspect is the buyer and their budget. Vertical software gets priced against a budget line that already exists inside the customer’s business. Breezeway sells into property operations, where the comparison is a labor cost. Flock Safety sells to police departments and municipalities, where the comparison is a public budget. Procore sells into construction, where the buyer already funds project management. Whoever knows the customer’s budgets best, ends up owning pricing.

Subscribe now

AI-native companies have not settled it

Below $100M, three roles of AI-native companies land in three different places: Legora hires a Sr. PMM for Pricing & Packaging under the CMO, LangChain hires a “Senior Fullstack Engineer, Growth & Monetization” under the CTO, and Decagon hires their first pricing hire in their Business Operations team. Each have a different focus - packaging optimization vs. pricing experimentation vs. deal structuring.

Between $100M and $500M fragmentation continues. Six roles, six homes. Mistral under the CPO. Perplexity under the CTO. AlphaSense under Strategy EVP. Glean under the CMO. ElevenLabs under the COO. Sierra under the CRO.

Only above $500M do you start seeing a pattern. CPO becomes most common with 50%, CFO at 33%, CTO at 17%. OpenAI actually has multiple pricing roles under the CPO and the CFO. Cursor keeps it under the CTO. Anthropic runs it under the CFO.

AI-native companies price against a cost curve that moves rapidly, so the person who owns inference economics is critical to understand economics with growth. Metering and rating also is critical for usage-based AI pricing. That is why pricing is under a CTO at a coding tool. And beyond $500M, product strategy and gross margin are what investors ask about, which explains the function sitting in again product & finance.

Company ownership affect who owns pricing

Among founder-owned and VC-backed companies, the most common home for pricing depends on the size / stage. Below $100M, it is the CPO at 32%. Between $100M and $500M, the CMO at 35%. Above $500M, the CFO at 36%. The smaller ones focus on growth, the large ones shape their financial story for an exit.

PE-controlled companies want change fast. Between $100M and $500M, half report to the CRO. Post-buyout, price realization through sales discipline is the fastest lever to optimize: reduce discount leakage for new deals, push for more renewal uplift. Above $500M it splits evenly between Strategy EVP and CPO at 38% each - at this stage, a coherent pricing strategy across a portfolio is a big lever by itself.

Public companies between $100M and $500M prefer Pricing roles under the CPO and CMO at 40% each. They are closest to customer needs and decide what to build and launch next. Above $500M the CPO still leads with 43%, but the CMO loses share likely because their focus at that scale moves further away from Product Marketing.

The first pricing hire happens earlier than you think

Out of 17 confirmed first-time pricing hires, 47% were at companies below $100M ARR and another 35% were between $100M and $500M. The median for first time hire Pricing roles in this dataset is $132M ARR, the top quartile is $36M ARR.

Companies also hire a pricing strategy role before an exit. 82% of first-time pricing roles were at founder-led, VC-backed companies, even though those companies made up only about half the JDs in the dataset. A dedicated pricing role before an exit strengthens financials, ensures consistency, and signals maturity.

At scale, one owner stops working

OpenAI and Google both have multiple C-level pricing homes. OpenAI splits product pricing from GTM and finance pricing. Google splits Cloud business pricing from AI product monetization.

Such federated pricing appears at large companies once the product portfolio outgrows a single buyer. AI subscriptions and cloud infrastructure are sold to different buyers and are also quite different on the cost side. Splitting the pricing roles under two executives can drive collaboration across functions, with each function optimizing for its own price.

What to do with this

The reporting line is a design decision, so make it on purpose:

• If you are a founder, start out with pricing early under the CPO or CMO.

• If you are industry-specific, hire under who knows the customer’s budgets.

• If you are AI-native below $500M, align it with who owns your unit economics.

• If the role approves deals, hire under the CFO or CRO for margin discipline.

• If you are PE-backed, hiring under the CRO might fix price realization fastest.

• If you are past $500M with multiple product lines, you likely need multiple homes (e.g. CPO for product pricing + CFO for GTM pricing).

Closing Thought

Pricing has no one dominant home, but it depends on four things you can actually check. Whether the role owns deal approval, whether you sell to one industry or many, how much revenue you have, and who controls your cap table.

One caveat on all of it. These insights come from analyzing job descriptions, so they capture where companies are actually hiring pricing leaders right now. With AI reshaping how companies price, we might see these patterns shift over time.

For now, it is what I would trust. It beats relying on gut instinct. Answer those four questions and you know where pricing should sit for you specifically.

Keep Reading

→ Spotify: https://lnkd.in/gquubN6G
→ Apple Podcasts: https://lnkd.in/gM5tVKJk

What we covered
↳ Pricing (user, usage, outcome) vs. Billing (subscription, consumption)
↳ Why buyers need predictability
↳ AI pricing tactics that work for customers and vendors
↳ The Monetization Flywheel
↳ What’s next: more optionality, and whether AI agents will replace humans

Listen to the full episode:
→ Spotify: https://lnkd.in/gquubN6G
→ Apple Podcasts: https://lnkd.in/gM5tVKJk

Keep reading

Halfway through 2026, the big acquirers started reaching across several layers at once. This locks in the control points where leverage compounds, switching costs rise, and architectural choices become durable.

The main difference in 2026 so far is also in concentration: SpaceX moved across the layers fast, and at larger scale, than anyone else. Here are the 10 biggest deals:

1) Vertically integrated AI infrastructure

(Silicon → power & cooling → interconnect → compute)

Deals:

• SpaceX acquired xAI (the Colossus supercomputers, Grok, X) for $250B in February 2026.

• Texas Instruments acquired Silicon Labs (connectivity and edge silicon for AI devices and data centers) for $7.5B in February 2026.

• Ecolab acquired CoolIT Systems (direct liquid cooling for AI data centers) for $4.75B in March 2026.

• Qualcomm acquired Modular (hardware-agnostic AI inference and compute stack) for ~$3.9B in June 2026.

• Marvell acquired Celestial AI (silicon photonics and optical interconnect) for $3.25B, up to $5.5B with earnouts (announced Dec 2025, closed Feb 2026).

What is converging: Last year this layer collapsed simulation, silicon, networking, and compute into one stack. In 2026 it extended in two directions. Down into power and thermal, and up into the supercomputer itself.

xAI is the apex of that. It covered three layers at once: It captured the compute (Colossus’ 555,000-GPU cluster drawing two gigawatts, the gas turbines and substations feeding it & the planned orbital data centers), the model (Grok), and the distribution and identity surface (X).

And SpaceX now even rents xAI compute out to other AI providers. Anthropic took all of Colossus 1, more than 220,000 GPUs and over 300 megawatts, to serve Claude. Google took capacity in Colossus 2 through 2029 for $920M per month!

Why it matters: The AI infrastructure moat is now physical and full-stack. Power, cooling, photonic interconnect, and the supercomputer itself. Owning an AI data center is becoming as defensible as making the AI chips.

The year’s biggest compute commitments that were not acquisitions point the same way. Meta and OpenAI both signed commitments with CoreWeave for $21B and $22.4B respectively. Those were capacity contracts though, not transfers of ownership, so I left them out.

2) Agentic experience & interaction

(Apps → agents → ambient AI)

Deals:

• SpaceX acquired Cursor / Anysphere (AI coding) for $60B in June 2026. SpaceX used its newly public stock to fold Cursor in days after the IPO.

• Salesforce acquired Fin / Intercom (autonomous customer-service agents) for $3.6B in June 2026.

• Apple acquired Q.ai (silent-speech and on-device audio AI) for ~$1.6B in January 2026.

What is converging: The interaction surface is splitting two ways. Toward agents that resolve work end-to-end (Fin in service, Cursor in code). And toward ambient, on-device sensing that removes the interface (Q.ai reads mouthed and whispered words from facial micro-movements).

Why it matters: SpaceX bought its way into enterprise workflows by buying the coding AI. Apple bought its way back toward frontier AI through audio and imaging in its second-largest acquisition ever.

3) Agent identity & trust

(Human identities → agent identities)

Deals:

• ServiceNow acquired Armis (AI-powered cyber asset and exposure management) for $7.75B (announced December 2025, closed April 2026).

• ServiceNow acquired Veza (AI-native identity for human, machine, and agent access) for ~$1B weeks earlier.

What is converging: Asset visibility, exposure management, and permission mapping are merging into a single agent-aware trust layer. ServiceNow stitched what an enterprise has (Armis) to what every identity can do (Veza). It built the combination explicitly to govern autonomous agents.

Why it matters: As agents take real actions inside enterprises, the control point is shifting from the user login to the permission graph for non-human identities. Similar to the Wiz/CyberArk logic of 2025, ServiceNow in 2026 has bought its identity stack.

4) End-to-end data & agent context

(Raw data → telemetry → grounded agent context)

Deals:

• Salesforce acquired Contentful (composable content layer for AI agents) for ~$1-1.5B in June 2026.

• Snowflake acquired Observe (AI-powered observability and telemetry) for ~$1B in January 2026.

What is converging: Data, content, and telemetry are merging into the context layer that grounds what agents do. The platform has to feed agents the right content to act on and watch the actions they take.

Why it matters: “Models churn. Data moat compounds.” That was my 2025 data-moat thesis, and it extended in 2026 to governance of the live content and telemetry that agents read and write. Owning the context is owning the agent’s version of reality.

5) The June cluster & the meta-convergence

In the first half of 2025, inference became better understood and confidence in enterprise demand for AI rose. That fueled a consolidation of point solutions across the 4 layers of the AI stack in the second half of the year (I wrote about it here).

And just as more than half of 2025’s top deals landed in a single July spree, 2026’s biggest deals clustered around one event. SpaceX went public on June 12. Days later it used its freshly liquid stock to buy Cursor for $60B. Salesforce announced Fin and Contentful in the same stretch.

Four things changed the shape of the year:

SpaceX dominated. xAI and Cursor together were $310B, that is ~90% of the total 2026 H1 M&A. That concentration is extreme. Strip out the Musk empire and the top 10 are ~$35B and more evenly spread across every layer. Salesforce was the other major player that consolidated across two layers (Contentful data + Fin experience). SpaceX wrote the headlines and reached across infrastructure (xAI Colossus), model data (Grok), and experience (X) with the xAI deal, then added Cursor too.

AI demand growth is unprecedented. Stanford’s 2026 AI Index Report found that Generative AI adoption reached 88% of organizations (up from 78% in 2025), and 53% of the population use AI. That rapid adoption within three years is faster than that of the PC or the internet.

AI capability continues rising fast. The same Stanford 2026 AI Index found that performance kept rising across model providers. On a key coding benchmark performance rose from 60% to near 100% in just one year. And compute capacity grew 3.3x per year since 2022.

Prices for chips and memory skyrocketed. TrendForce reported record highs for DRAM and NAND memory prices as suppliers prioritized higher-margin data-center demand. This increased cost pressure on consumer devices such as laptops and tablets. Apple already raised prices across the board in June 2026 since “The rapid expansion of AI data centers has created an extraordinary surge in demand for memory and storage.”

Closing Thought

SpaceX is the first AI behemoth to assemble this much leverage across the AI stack. It owns frontier-model infrastructure (xAI and Colossus), consumer distribution (X), and now also enterprise workflow (Cursor). It also sells infrastructure capacity to Anthropic and Google. Grok competes with Claude and Gemini, while SpaceX controls compute capacity both rivals use.

Infrastructure and hardware have become critical to AI. Data center capacity, GPUs, memory, power and cooling are becoming strategic constraints. Recent memory shortages have already flowed through to higher prices for consumer electronics. As AI demand and model capability keep rising, long-term access to infrastructure becomes critical for AI providers and the companies building on top of them.

For AI builders, the strategic risk is vendor lock-in. In 2025, enterprises piloted AI use cases while large buyers consolidated individual layers of the stack. In 2026, one supplier can control multiple layers of the AI stack you rely on. Infrastructure costs, roadmap dependency, and pricing power become linked.

The companies that preserve leverage will know which control points they must own, which they can rent, and where they need a credible second supplier.

For recurring pricing work, scheduled automations beat AI agents in most cases. They cost less, can do multiple pre-defined steps (in a defined sequence) and are more predictable.

Start with understanding your AI cost. Uber’s CTO burned through his entire annual AI budget in four months. One consultant told Axios that a single client spent $500M on AI in a month. Cursor had to refund customers after surprise bills last year. The pattern repeats: usage scales faster than expected.

The per-token price you compared up front does not tell you what a task actually costs. The 2026 paper “The Price Reversal Phenomenon” found the cheaper-looking model is often the more expensive one, because reasoning models spend very different amounts of tokens on the same job. Gemini 3 Flash’s list price is 80% below the GPT-5.4 the paper tested, yet was 38% more expensive across the test set. In about one third of model-pair comparisons the model with the lower per-token price ended up being more expensive overall. I broke this down in full here.

Agents make this worse. Model-driven loops can increase token consumption above what a simple scheduled run would have cost. Anthropic’s own guidance states

“When building applications with LLMs, we recommend finding the simplest solution possible, and only increasing complexity when needed. This might mean not building agentic systems at all.”

Automated AI workflows are often far cheaper and offer the predictability and consistency needed for many tasks. Barry Zhang from Anthropic advises to use workflows instead of agents for many use cases:

“If your budget per task is around $0.10, for example, you’re building a high volume customer support system… in that case, just use the workflow to solve the most common scenarios and you’re able to capture the majority of the values from there.”

Workflows win on predictability for well-defined tasks. Agents earn their cost only when you genuinely need decisions made on the fly at scale. Most recurring pricing tasks are the first kind.

Subscribe now

Schedule AI automation instead of AI agents

Here is how you schedule AI automations for pricing work with the major AI tools. You can run prompts on an automated schedule with all the major chatbots (ChatGPT, Google Gemini, Claude Cowork, Grok). There are two caveats:

1. Only ChatGPT can run automations AND update existing Google sheets.

2. Claude Cowork needs your laptop to be turned on to run automations.

Here is a step-by-step for each:

Automate with ChatGPT: Scheduled Tasks

In ChatGPT, you can create a new scheduled task from most chats; just not from within GPTs. You simply type a prompt that indicates the frequency, how often it should run, and the prompt you want to run. For example:

This will create a scheduled task that runs automatically as per the cadence you asked it to run. Make sure you select the model you want it to use when prompting it to create it. The model selection is locked in and cannot be changed later. If you want it to run it in a different model later, you would have to recreate a new task for it.

When you click on the scheduled task, you can edit the prompt or change its cadence:

You can view and edit all your scheduled tasks by clicking on “Scheduled” in the sidebar, or simply by bookmarking https://chatgpt.com/schedules:

Once created, ChatGPT will email you every time your prompt runs.

Connect it to Google Drive: If you want the task to update a particular file, like a Google sheet tracker, you first need to connect it to Google Drive. To do that, click in the sidebar on “Apps”, find Google Drive, and follow the prompts to grant it access.

Once you connected ChatGPT to Drive, you can then add this to your prompt:

The first time you run the task, you might see a card asking you to confirm your permission to allow ChatGPT to use Google Drive like this

Being able to edit an existing Sheet is a unique capability that only ChatGPT has (as per my knowledge), and is excellent if you track changes over time in a file. The version history works accurately too, so you can see changes made with the appropriate time stamp.

Automate with Gemini: Scheduled Actions

In Google’s Gemini, you can in a similar way create scheduled actions, as long as you have a paid Gemini plan like Google AI Pro or Ultra. You can prompt creating scheduled actions from any chat by typing for example:

This will create a scheduled action that runs automatically as per the cadence you asked it to run. Make sure you select the model you want it to use when prompting it to create it. The model selection is locked in and cannot be changed later. If you want it to run it in a different model later, you would have to recreate a new action.

You can view all your scheduled tasks by clicking on the cogwheel at the bottom left of your Gemini console, and then on “Scheduled actions”, or simply by bookmarking https://gemini.google.com/scheduled. This shows you the list like this, and you can also create new scheduled actions from here:

Connect it to Google Drive: If you want it to run a workflow outside of Gemini, it is by default connected to Google Drive and can ask it to update a particular file, like a Google sheet tracker.

One important limitation: Even though you can connect Gemini to Google Drive, Scheduled actions cannot change existing documents. It can only read your existing files and create new ones.

Automate with Claude Cowork: Scheduled Tasks

Claude can also run automations, but there is a catch: Anthropic so far has scheduled tasks not in the UI of the web app, but only in the Claude desktop app. So the main difference is that scheduled tasks run locally on your machine in Cowork and require your laptop to be powered on, or they will not run.

To create a scheduled task with Claude Cowork, first you’ll need to download Claude to your laptop. Once you have the app installed, select “Cowork” in the top left, then either

• Click on “Scheduled” in the sidebar and then on “New Task” on the right, or

• Type /schedule in the task input to launch a Skill that walks you through creating a task, or

• Describe your task in a prompt like this:

Then click on “Schedule”.

Scheduled tasks only run while the Claude app is open. You can edit Scheduled tasks by clicking on “Scheduled” in the Cowork sidebar. If you scheduled a task to run during peak hours (i.e. weekdays 5AM-11AM PT), you likely see a warning here that this task “will consume your usage limits faster”.

If you want to consider running automations overnight to save cost, you will have to keep your laptop on. Claude has the “Keep awake” option for that. If your laptop is turned off at the time you scheduled a task, the automation runs once you launch Claude next time.

Clicking into the task details shows you how Claude already enhanced your prompt. You can also edit the prompt further here and even change which model it runs with.

Once you run the task for the first time, you might see a notification asking to grant specific permissions, which are then added to the task for future runs.

Connect it to Google Drive: If you want it to run a workflow outside of Claude, you can connect it to Google Drive and have it read existing files or create new files. To connect it, click in the Claude Cowork sidebar on “Customize”, then on “Connect your apps”, search for “Google Drive”, then on the black button “Connect”, and follow the prompts in your browser to grant Claude read/create permissions.

One important limitation: Even though you can connect Claude to Google Drive, it cannot change existing documents. It can only read your existing files and create new ones.

Automate with Grok: Scheduled Tasks

I found no way to create a new scheduled task from just a normal chat in Grok. However, you can navigate to Tasks by clicking on your name in the bottom left and then on “Tasks”.

Or simply bookmark the URL directly: https://grok.com/tasks.

Once there, click on “Create task” or “New task”:

That will open a popup window where you can select your cadence, enter your desired prompt, and select how you prefer to be notified (email, app, both, or not at all):

There is no model selection here, and given the only option I found to create a task is via navigating to Tasks, I am not 100% certain what model the tasks run with.

Connect it to Google Drive: If you want it to run a workflow outside of Grok, you can connect it to Google Drive and have it read existing files or create new files. To connect it, click in the Grok sidebar on “Skills and Connectors”, then click on “Google Drive”, then on the black button “Connect”, and follow the prompts in your browser to grant Grok read/create permissions.

One important limitation: Even though you can connect Grok to Google Drive, it cannot change existing documents. It can only read your existing files and create new ones. I’ve seen Grok do a smart workaround though: create a new file with the additional rows added, saved as a new file and the old file moved to trash.

Keep Reading

AI usage pricing requires metering usage at scale, and billing accurately based on that usage. That created a new infrastructure category: usage-native metering and rating platforms such as Metronome, m3ter, Orb, OpenMeter, and Togai.

Zuora, Kong, Stripe, Salesforce, and Adyen have all moved to acquire that layer that turns product consumption into billable revenue. The meter is getting absorbed into the systems that already own the revenue workflow: payments, CRM, API infrastructure and subscription billing. The pace has accelerated.

Why metering tools exist

Billing AI usage has two hard parts: metering and rating.

Metering is observing and recording raw usage events. Events are API calls, tokens processed, GB transferred or GPU-second consumed. That sounds simple until you are ingesting millions of events per minute in real time across thousands of customers, with full auditability for every one.

Rating takes those counted events and applies your pricing logic to them, converting usage into numbers, e.g.

• Tiered limits that reset monthly

• Overage charges once included balances are drawn down

• Minimum commitments that floor the invoice

• Volume discounts that kick in at thresholds

• Custom contract terms that override standard pricing

The two functions are separated by design. You can meter perfectly and rate wrong. You can have correct pricing logic but incorrectly meter events upstream. B2B contracts make this specifically hard for three reasons:

The scale problem: AI companies’ token consumption has orders of magnitude more events than what SaaS billing systems were designed to handle. Think of metering millions of events per minute vs. thousands of seats per month.

The flexibility problem: Usage-based pricing models iterate constantly. Rob Litterst from PricingSaaS reported on average more than 3 pricing changes per company in 2025, with fast-growing companies making multiple pricing changes per month. Each change needs to flow through the rating engine without an engineering sprint.

The auditability problem: Revenue recognition requires a complete event ledger. Every metered charge needs a traceable lineage from raw event to billable line item. If you cannot defend a charge when a customer disputes an invoice, you will lose trust.

A small set of purpose-built companies emerged to solve this. They became the infrastructure layer for billing AI usage. Now larger platforms started acquiring that capability rather than building it.

The M&A wave

Five acquisitions, five different acquirers, the same underlying logic: a platform that already owns part of the revenue workflow acquires the metering and rating layer it was missing.

Adyen + Orb (announced June 11, 2026)

Adyen already serves customers like Uber and Spotify. Orb’s customers include Perplexity, Replit and Vercel. Orb lets engineering teams define billing metrics directly with SQL queries, simulate pricing changes against historical usage data before deploying in production, and report on usage in real-time. Orb CEO Alvaro Morales on the rationale of bringing both together:

“Standalone billing systems are fundamentally limited because they operate blind to transaction execution … By joining forces with Adyen, we can connect this ingestion layer directly to real-time financial health signals, closing the loop between billing logic and payment success.”

Prior to its acquisition by Adyen, Orb had raised $44M, including $25M from its Series B in September 2024.

Salesforce + m3ter (announced June 8, 2026)

Salesforce already owns the CRM, CPQ, and increasingly the agentic workflow layer. Without a metering engine, they couldn’t close the loop on consumption-based AI monetization. m3ter gives them enterprise-grade rating inside Revenue Cloud.

Salesforce framed it directly: acquiring m3ter to

“bring high-volume mediation, metering and rating capabilities natively to Agentforce Revenue Management, enabling enterprises to launch, track, scale, and bill with the flexible usage and outcome-based pricing models needed for the AI era.”

m3ter had raised $31.5M before its acquisition. Latka estimated the deal cost Salesforce around $140M.

Stripe + Metronome (completed January 2026, reported ~$1B)

Metronome’s customer list reads like an AI infrastructure directory: OpenAI, Anthropic, Confluent, and Nvidia. Stripe CEO Patrick Collison was plain:

”Metered pricing is the native business model for the AI era.”

Metronome had raised $128M, including a $50M Series C just months before the deal. Stripe reportedly paid around $1B, ~2.1x Metronome’s last $470M valuation, per Sacra.

Kong + OpenMeter (acquired September 2025)

Kong is the API gateway and already sits in the traffic path for API calls. Adding OpenMeter means Kong can now translate that traffic directly into billable events. The API gateway becomes part of the monetization system.

“OpenMeter’s capabilities will be integrated into Kong Konnect, enabling usage-based pricing, entitlements, and invoicing for APIs, events, and AI workloads.”

OpenMeter had raised $3M prior to the acquisition.

Zuora + Togai (acquired in May 2024)

Zuora has owned subscription billing for years but was architecturally weak on usage.

“Togai’s low-code builder and developer-friendly interface to quickly configure metering and rating of raw events” filled the gap without Zuora having to rebuild its core engine.

Zuora acquired Togai for $24.3M. Togai had raised $3.1M prior to its acquisition.

The pattern across all five:

Five different players, all move into usage metering and rating:

• Orb → Adyen = Payments + Billing + AI consumption metering

• m3ter → Salesforce = Enteprise CRM + AI consumption metering/rating

• Metronome → Stripe = Payments + Billing + AI consumption metering

• OpenMeter → Kong = API gateway + API usage rating

• Togai → Zuora = Subscription monetization suite + usage rating

Who’s left standing

Not everyone got acquired. Two notable independent players are still building as standalone companies, and their positioning reveals what they’re betting on.

Sequence goes wider than pure metering. It positions itself as a CPQ + billing + usage metering + revenue recognition + receivables automation platform. Their customers include Cognition, Legora and 11x. Their bet is fast-scaling AI companies need one platform for the entire workflow from contract signature to accurate billing. Sequence raised $38M from investors like 645 Ventures and a16z, including $20M from a Series A in December 2025.

Lago is the open-source bet. Customers include Groq and Mistral.ai. It allows AI labs to self-host and control their billing infrastructure. Lago’s differentiation is deployment flexibility: fully managed cloud or on-premises, with full access to the underlying data. For companies with data residency requirements or platform independence as a strategic priority, Lago is the natural alternative to the lock-in that comes with being inside Stripe’s or Adyen’s billing stack. Lago raised $22M from investors like FirstMark, SignalFire and Y Combinator. Its Series A valued it at reportedly $100M.

What this means for pricing strategy

For Pricing teams, metering is becoming a platform capability. Instead of having to buy a separate metering tool, Pricing teams can leverage their existing CRM / Billing / API gateway. That might make it a little easier to secure a budget for metering.

For founders building on AI, billing infrastructure is now a platform alignment choice. The vendor that owns your metering has full visibility into your unit economics: consumption patterns, pricing model structure, which customers have custom contract terms, where you draw overage thresholds. The vendor that owns your metering owns your pricing intelligence. That leverage matters the next time you’re renewing with them.

For the remaining standalone players, Lago and Sequence, a strategic acquisition looks like the most plausible path. The obvious buyers are platforms that already own parts of the revenue workflow but have not yet acquired a usage-native metering and rating specialist: Microsoft, ServiceNow, HubSpot, or larger billing incumbents like Chargebee, BillingPlatform, and keylight. Those billing incumbents already support subscription, usage-based, and hybrid models, but as AI monetization keeps pushing toward event-native metering, real-time rating, and constant pricing iteration, they may still decide that owning deeper metering infrastructure matters.

The window for those acquisitions is closing. Once the category fully consolidates, founders building AI products will have to commit to a wider platform when choosing their metering and rating engine.

Keep Reading

Billing engines from Metronome, Stripe and Charbee made it easier to bill customers based on actual consumption. But just because you can doesn’t mean you should.

Customers have come to hate the unpredictable AI usage bills. They need budget certainty. They want adoption without the risk of runaway spend. And they do not want high usage bills without clear outcomes.

Token consumption varies on top of token pricing

Tokens are still relatively new to customers, and companies are figuring out how many they need and how to control usage. On top of that, token consumption varies by vendor and model, making token prices hard to compare.

This study from March 2026 found that the cheapest AI model on the pricing page is not always the cheapest in production, because token consumption varies among models. Though Gemini 3 Flash’s listed token price is 80% cheaper than GPT-5.4’s, in their test the cost across tasks was 38% higher.

Not knowing how much tokens a particular action or outcome takes for each model makes understanding token usage and total cost hard. Putting the risk of varying token consumption on the customer when models don’t deliver the expected outcome consistently (because they are not deterministic) makes this problem worse.

Customers need predictability

And companies are struggling with unexpected AI usage bills:

Uber CTO Praveen Neppalli Naga acknowledged in April 2026 that he had already spent his AI budget for the entire year, 4 months into the year:

Axios reported a company even spent $500 million in a single month on AI usage. They had not set usage limits, and usage went unexpectedly high.

Cursor faced a backlash after changing its Pro pricing in the summer of 2025. Their CEO Michael Truell had to publish a pricing clarification, acknowledging “we missed the mark” and offer refunds to customers who had a “surprise usage bill.”

OpenAI CEO Sam Altman said in May 2026 “customers are increasingly asking us for certainty on capacity”, and that OpenAI is now selling “discounted tokens for 1-3 year commits”. He sees it as “a big win-win” since it also helps OpenAI plan. He’s right.

AI companies need growth

He hit the nail on the head. AI vendors need predictability to plan too. They need to fund huge capital investments in the billions and justify forward-looking revenue. That’s harder if you don’t have ARR. Anthropic reported in April 2026 that they have surpassed $30 billion “run-rate revenue”, up from $9 billion at the end of 2025, which justifies their $50 billion investment in computing infrastructure.

Until now though, AI companies focused on glorifying token consumption (not recurring commitment) to fuel growth and have been great at this:

OpenAI started giving physical awards in 2025 to companies crossing 10B, 100B and 1T tokens last year. The Tokens of Appreciation program was so successful that OpenAI increased thresholds for 2026 by 10x to 100B, 1T and 10T.

That was the start of token-maxing: employees now use AI tokens to show they leverage AI more than others, that they are the ones that lead innovation.

A Meta employee even created a dashboard so coworkers can compete to become the company’s highest token consumer.

So…

AI vendors want usage to keep growing and are great at promoting it.
Customers need the spend to be predictable.
That tension is why pure pay-as-you-go billing will break down in B2B AI.

Subscriptions don’t have to be per seat

Usage pricing and usage billing are not the same thing. Pricing defines the logic of value, what customers pay for. Billing defines the rhythm of cash, when customers pay.

Companies and CFOs are used to SaaS subscriptions that are priced per seat and usually paid annually or monthly. If the pricing metric changes from users to usage, that does not mean billing has to become pure usage-based billing. For example

• If your pricing metric is usage, contract capacity.

• If your pricing metric is outcomes, contract outcome volume.

• If your pricing metric is credits, sell recurring credit subscriptions.

You can still contract commitments on AI usage, capacity, tokens, credits, outcomes, or resolutions. Then bill that commitment as a subscription.

I wrote more about this distinction in Untangling billing and pricing.

AI is moving away from pure pay-as-you-go

This isn’t just me saying this might happen. The shift is already underway:

AWS sells Bedrock Provisioned Throughput alongside its on-demand pricing since September 2023. AWS has also sold EC2 Reserved Instances since at least 2009.

Microsoft started selling Azure OpenAI Service Provisioned Reservations in late Summer 2023, alongside its hourly no-commitment purchasing.

Google started selling Provisioned Throughput for its Vertex AI in June 2024.

Salesforce introduced Agentic Enterprise License Agreements (AELA) in late 2025, offering flat rate, unlimited usage pricing for its AI agents.

Zendesk has priced AI Agents based on Automated Resolutions since November 2025. Customers can choose to commit to volume at $1.50 per resolution, or pay a higher price of $2 per resolution without commitment.

OpenAI announced it is introducing Guaranteed Capacity on May 19, 2026, saying:

Even if AI pricing stays usage-based for the foreseeable future, AI billing is moving toward contractual commitments and predictable recurring revenue.

B2B contracts are a Win-Win

For customers, commitments help with:

• Budget certainty: CFOs prefer predictable spend over volatile usage bills that add another risk to EBIT.

• Internal planning: Teams can forecast AI spend, allocate budgets, and avoid monthly surprises.

• Adoption pressure: Longer commitments create a commitment effect: “We already bought it. Let’s use it.”

• Commercial leverage: Larger commitments can justify better rates, capacity guarantees, stronger support, or roadmap access.

For AI providers, they help:

• ARR: Usage becomes contracted recurring revenue, instead of uncertain pay-as-you-go consumption revenue.

• Better forecasting: Finance can model revenue, capacity needs, GPU commitments, and gross margin with better confidence.

• Infrastructure funding: Long-term customer commitments can support long-term compute commitments.

• Lower churn risk: Every renewal is a chance to reconsider competitors. Longer terms reduce evaluation frequency. And after 1 to 3 years, status quo bias works in your favor.

So what’s next

Recognize that pricing and billing are two separate decisions.

Usage-based pricing is fine. But don’t bill customers against a metric they can’t predict, at rates that vary by model, for outcomes that aren’t guaranteed. Don’t put all the risk on the customer while you capture all the upside of growing consumption, while eroding the trust you need for a long-term relationship.

Pick the pricing metric that’s right for your business. But let customers contract a commitment against it, and bill that commitment as a predictable subscription. Give them visibility and controls to manage their spend without a mid-year crisis.

That way AI vendors get ARR, and customers get predictability. And AI moves from a line item CFOs fear to a partnership that deepens over 1-3 years.

Subscribe now

Keep Reading

This builds on the classic price waterfall from Managing Price, Gaining Profit, which showed how companies leak revenue from list price, invoice price, to pocket price. It’s a classic and useful, but with AI we need to consider variable costs.

For years, the standard SaaS price waterfall was:

List price → discounts → net price:
Pricing Teams were focused on determining the list price and questioning standard discounts and sales concessions. But with AI, one more step is critical:

Net price → variable cost → contribution margin:
A customer can look attractive at the net price level and still be margin dilutive once you account for AI inference.

The AI Price Waterfall

The AI price waterfall puts variable cost in perspective to the discounts sales is already used to from B2B SaaS and AI pricing.

List Price: The standard published price before any discounts, concessions, fees, or customer-specific adjustments are applied.

• Packaging Discounts increase product adoption or deal size by incentivizing customers to buy multiple products, modules, bundles, or the full suite together.

• Volume Discounts reduce the effective unit price as customers commit to higher seat counts, usage levels, transaction volumes, or enterprise-scale deployments.

• Commitment (Billing & Contracting) Discounts incentivize longer contract terms, annual billing, or prepayment upfront. This article is a deep dive on these.

• Channel Discounts (or fees) compensate intermediaries such as resellers, MSPs, or marketplaces for selling, managing, or distributing your product.

• Discretionary Discounts extend deal-specific discounts selectively to win early adopters, match competitors, or secure strategically important accounts.

= Net Price: The actual realized customer price after all discounts, credits, rebates, channel margins, and negotiated concessions.

Variable Cost of Revenue are the variable costs to deliver your product such as AI inference / usage, cloud hosting, third-party APIs, or data acquisition and processing.

= Contribution Margin: The revenue remaining after subtracting variable cost of revenue, used to assess deal profitability and discount headroom.

Why This Matters More for AI

Notice the waterfall does not stop at the net price. It continues through variable cost of revenue and ends at contribution margin.

Traditional SaaS companies ignored that step, because variable costs were generally low. They had high gross margins and relatively predictable cost to serve. That made list price determination and discount management the two main pricing problems.

AI changed that. Two customers paying the same net price can generate very different margins. One customer is a light user with predictable workflows. Another runs high-volume inference, relies on expensive frontier models, and triggers APIs often. If both customers get the same discount, they look similar in ARR but very different in contribution margin. Your pricing strategy needs to account for that, or you end up subsidizing your heaviest users that get the most value.

A better waterfall shows where price is being given away and where margin is being eroded. Net price is no longer the finish line. For AI products, it is the midpoint. The waterfall helps pricing teams see whether a pricing strategy works after discounts AND usage costs.

Waterfall Template for Google Sheets

I made you a simple Google Sheets template for this: the list-to-net price waterfall template. It has two tabs: one with instructions, the other with the waterfall.

Make a copy and use it as you wish. Use it to compare customer segments, align your leadership on which discounts are creating value versus eroding margin, or to simply compare deals. My goal is to make it easier for you to do your job and show:

List price → discounts → net price → variable cost → contribution margin

If you share it, please just be so kind and mention where you got it from. 🙂

Each team needs incentives that reward revenue quality, retention discipline, value realization and focus. So how do you set these incentives for your teams?

Sales: Prioritize quality revenue, not just more ARR

Most sales incentive plans still overweight one metric: new ARR.

That is simple to manage, but it misses an important point: not all revenue is equally valuable. A deal closed with heavy discounting or poor pricing discipline may help bookings while hurting long-term economics.

BCG argues that pricing metrics should be part of Sales Incentive Plans (SIPs), yet fewer than 10% of SIPs actually include any price-based metrics. That means 90+% still focus on New ARR with no indicator of revenue quality.

Two better metrics you should consider:

Price realization measures how well reps hold price versus list or target price.

• Pro: keeps product cost and margin confidential

• Con: does not directly steer portfolio mix or favor higher margin products

Margin performance measures the profitability of what is sold.

• Pro: directly connects rep behavior to the company bottom line

• Con: exposes margin and cost data broadly, which may change over time

Rep performance can also be evaluated in different ways:

• Absolute performance

• Improvement over time

• Hybrid model combining both

If you want more profitable growth, don’t pay Sales as if every dollar of ARR is equally good. Set targets that align with your strategy and consider price realization or margin performance targets. For most companies, price realization is the easier starting point because it improves discipline without exposing sensitive margin data.

Customer Success: Separate retention from expansion

The same logic applies after the initial sale. For years, many CS teams were measured primarily on NRR, which combines retention and expansion into one number.

Move from NRR to GRR to isolate retention

That worked well in an era when expansion was abundant and growth the overriding goal. The TSIA State of Customer Growth and Renewal 2025 reports that

“The days of unchecked expansion have been replaced by a strategic focus on profitability, forcing companies to rethink how they allocate resources, manage renewals, and drive expansion.”

And GainSight’s Customer Success Index 2025 shows that companies are separating those motions and putting more emphasis on GRR:

• 62% of CS orgs use GRR as their primary metric (up from 45% 2 years ago)

• CS orgs using NRR as their primary metric declined from 59% to 51%

NRR is still valuable, but it can mask underlying retention problems when expansion is strong. GRR makes customer health and renewal performance easier to see. It answers a more disciplined question: are we keeping the revenue we already earned?

Companies are adopting a more disciplined approach to financial management, moving beyond rapid expansion to focus on sustainable, long-term gains.

Companies centralize renewal ownership in CS

Once companies separate retention from expansion, the operating model usually changes too. Renewal ownership becomes more specialized, and companies start asking whether expensive AE capacity should be used for every renewal motion.

They centralize renewal ownership within Customer Success rather than leaving it with higher-cost AEs. According to TSIA, shifting renewal and upsell responsibilities from AEs to specialized CS renewal teams can lead to:

• ~2/3rds lower renewal costs

• 10% higher net renewal rates

• Nearly 10% higher attached upsells

AI reinforces the efficiency play

AI adoption in CS reflects the same shift toward productivity and disciplined growth. The highest adoption use cases of AI in CS focus on efficiency:

• Auto summarization (82%)

• Drafting email/follow-ups (71%)

• Sentiment analysis (54%)

• Churn or risk prediction (39%)

• Next best action recommendations (30%)

These are not “AI for AI’s sake” use cases. AI is being adopted first where it can reduce manual work, improve consistency, and help teams manage retention at scale.

The incentive lesson is that CS teams should not be rewarded only for coverage or activity. You should reward them for scalable retention outcomes: lower renewal cost, faster risk identification, and better GRR.

Product: Value creation trumps feature velocity

Product teams used to be rewarded for shipping: roadmap delivery, launch volume, and shipping velocity.

In the AI era, incentives for Product teams need to shift to reward adoption, usage, retention, expansion and creating value to increase willingness to pay. There is a large gap between what teams build vs. what customers actually use.

The Product-Led Alliance’s The State of Product Analytics Report 2025 shows a similar shift towards value realization:

• Customer feedback is increasingly used to decide on new features (44% → 49%)

• Retention has overtaken revenue as the #1 goal for analytics programs

Role of AI: Define use cases with ROI & set targets

The same principle applies to AI investment. McKinsey’s State of AI in 2025 found that while 88% of companies use AI, only 6% get 5+% EBIT impact. “AI high performers” are more disciplined. They set growth and/or innovation as an objective of AI efforts, define use cases, redesign workflows and align leadership.

They follow 10 best practices across 4 areas:

Use cases and leadership alignment on value:

- Have defined a road map with specific AI initiatives and use cases across priority business domains, aligned with broader AI strategy

- Top leaders understand how AI can create value for the business

Rapid iteration:

- Have an agile product delivery organization or an enterprise-wide agile organization with well-defined agile team delivery processes

- Have an established process for building AI solutions and iteratively improving them (eg, guardrails, approach to development)

- AI efforts process quickly and are adaptive (ie, characterized by quick decision-making and iterative learning

Require Processes and role model:

- Embeds AI solutions into business processes effectively (eg, changing frontline employees’ processes, creating users interfaces)

- Senior leaders are actively engaged in driving AI adoption, including role modeling the use of AI

- Have developed a clear workforce plan (for technology and nontechnology roles) that incorporates the anticipated changes from AI

Frontier tech with human in the loop:

- Technology infrastructure and architecture allow implementation of core AI initiatives using the latest technologies

- Have defined processed to determine how and when model model outputs need human validation to ensure accuracy

Scaling agents are downstream of process redesign. High performers are almost 3x as likely to fundamentally redesign workflows and push for transformative change. They also scale use of agents 3x faster though, and get >5% EBIT impact.

The overall lesson

Incentives drive behavior. Behavior shapes the quality of your growth.

If you pay Sales only on new ARR, you incentivize discounting.
If you measure CS on NRR, early signs of churn can get masked.
If you manage Product by velocity, you incentivize features regardless of value.
If you experiment with AI without clear use cases, you are not going to see returns.

Profitable growth requires that you set incentives that reflect your goals.

Reward Sales for acquiring high-quality revenue.
Reward CS for protecting existing revenue.
Reward Product for prioritizing value creation.
Focus AI investments on measurable use cases with clear ROI.

Then your incentive system rewards profitable growth.

In SaaS, companies defaulted to per user pricing because seats were easy to explain, tracked value well enough, and variable costs were manageable. AI changed that. Costs are real, usage is often concentrated in power users, and outcomes can be attractive but hard to attribute.

In AI, the pricing metric often matters at least as much as the price point. Pricing teams spend a lot of time thinking through which pricing metric is ideal, so I pulled together what the experts say, and synthesized them into one practical framework.

What the experts say about a good pricing metric

A lot of smart people in pricing have already described what makes a good pricing metric. They use different language, but they are largely pointing at the same underlying truth: a pricing metric has to work for the customer and for the company at the same time.

Here is what some great pricing leaders said makes an ideal pricing metric:

1. Nagle & Mueller’s classic “The Strategy and Tactics of Pricing” has 5 criteria:
   - Tracks with Difference in Value Across Segments
   - Tracks with Differences in Cost-to-Serve
   - Is Easy to Measure and Enforce
   - Facilitates Favorable Positioning Versus Competition
   - Aligns with How Buyers Experience Value in Use

2. Kyle Poyar gave a SaaS monetization list of criteria in 2023 recommending metrics should be Value-based, Flexible, Scalable, Predictable, Feasible.

3. Simon-Kucher focuses on value and fairness: “The most effective price metrics are the ones that are aligned with the value customers derive from the offering. This means that the price scales alongside customers’ success or their usage of the software. How predictable and measurable these metrics are will depend on the industry your company operates in. At a minimum, however, price metrics should be accepted and considered fair by customers.”

4. Ed Arnold says that “Good pricing strategies align the product’s price metric with either the customer’s usage or value metric.“

5. Amy Konary has 5 criteria with a heavy focus on the customer view:
   simplicity, predictability, transparency, control, and behavior.

6. Mark Stiving names these 5 criteria for a pricing metric: Fairness & trust, Comprehension, Predictability, Growth & expansion, and Profitability.

Each of these focus on something important. Nagle & Mueller and Kyle Poyar lean more toward value capture and feasibility. Simon-Kucher, Ed Arnold, and Amy Konary put more emphasis on customer value alignment, predictability, and acceptance. Mark Stiving comes closest to balancing both sides by making the commercial consequences explicit. What I wanted, though, was one practical framework that separates and brings together the needs of both the customer and the provider.

Bringing the expert frameworks into one

These experts focus on different dimensions of the same underlying challenge. The core pattern across all of these expert frameworks is that a good pricing metric has to do two jobs at once. It must work for the customer, and it must work for the provider.

My synthesis is to collapse them into six practical criteria:

• For the buyer: be intuitive, controllable and predictable

• For the seller: be attributable, scalable and feasible

Customer-facing criteria

1. Intuitive

   • Is easy to understand and meaningful to the buyer persona

   • Reflects how customers perceive the value your product creates

2. Controllable

   • Customers can influence what they pay through their behavior

   • Feels reasonable relative to the perceived value received

3. Predictable

   • Does not vary materially due to inputs outside the customer’s control

   • Customers can forecast future spend within a reasonable range

Company-facing criteria

1. Attributable

   • The metric captures willingness to pay across segments and use cases

   • It helps isolate and monetize the value your product creates

2. Scalable

   • The metric increases as your product creates more value over time

   • You can reasonably control and forecast the cost-to-serve

3. Feasible

   • The metric is measurable within the product

   • You can measure it without depending on customer-reported inputs

Evaluating common AI pricing metrics with this

Once you apply the framework, you see the focus of the common AI pricing metrics:

User metrics

User metrics were a great proxy for value for SaaS providers, and they are easy to explain, control and budget for buyers. That is why they were popular in SaaS, and why they remain popular for AI in B2B, e.g. Canva as well as ChatGPT (the UI version).

However, for AI products, seats don’t track value creation well, and they cause issues for scalability and attributability. When users connect their Canva account with ChatGPT Agent, their Canva account still tracks this as 1 user seat, even though a human and an Agent both use their product. Identity providers are starting to separate agent identities from human ones, which could solve this also for pricing.

Usage metrics

Usage in AI is the best indicator for scaling revenue with cost, protecting margins when variable costs are high. They tend to also be feasible since a product can track its own usage. They are very common in infrastructure already, e.g.:

• Per API call: Google Cloud API Gateway

• Per duration and/or request: AWS Lambda

• Per token: ChatGPT API

However, they are often not great for the customer. They tend to be overly technical (e.g. input + output tokens, API calls) and volatility causes issues for budgeting.

AI credits

To solve for volatile usage metrics, many AI companies allow purchasing AI credits. These are not a pricing metric in itself but rather a fungible resource unit that can be used for consumption later. So they’re a billing mechanism. How much one credit is worth is often not transparent or at least hard to understand, which causes issues to connect back to perceived value for the customer.

Outcome metrics

These tend to be more intuitive and controllable for customers, and also easier to predict because they are close to the customer value, e.g.:

• Per successful payment: Stripe

• Per recovered chargeback: Chargeflow

• Per customer resolution: Intercom Fin AI

The challenge for AI vendors is to define outcomes they can attribute and control the cost-to-serve for. This is where many vendors fail, as we discussed in this article.

Hybrid metrics (Platform fee + usage or outcome)

This is where many AI companies will land, but from a practical standpoint they are also NOT a pricig metric per se. They combine multiple metrics in one pricing architecture as discussed here. The still leaves the choice of which metrics to combine into one architecture unanswered.

So how do you choose your right pricing metric?

The goal is to choose the one (or combination of multiple) that balances what your customers need and what you need. Here is the process that has worked for me:

1) Create a long list of ideas of potential pricing metrics

Start listing all plausible options that align with how the customer receives value:

• Outcomes

• Outputs

• Usage

• Activity

• Users

• Employees

• Workflows

2) Score each metric

Take the ideal pricing metric criteria and rate each one on a simple 5-point scale: High, High-to-medium, Medium, Medium-to-Low, Low.

Don’t overthink this. The point is to force tradeoff discussions.

3) Aggregate the scores

Average the ratings to create an overall view of which metrics look strongest. For example, High for Intuitive + Low for Predictable = Medium overall.

This will never be perfect, but it helps structure your discussions.

4) Model the pricing using real prospects

Consider a hybrid pricing architecture of multiple metrics (e.g. seats + outcomes) if you believe that balances the customer and your company needs best.

Then model out the financials! This is the step novice pricers skip. You need to see what pricing would actually look like for real customers with each potential metric.

5) Pick the metric that works for your customers and you

Choose the metric where the scoring indicates strong customer fit as well as strong company fit, and where the resulting pricing works best for your target customer profile.

The AI pricing metric that works

In AI, pricing power starts with charging with the right unit of value. If you choose the right metric, packaging and price points fall into place more naturally. If you choose the wrong one, even the most elegant pricing architecture will struggle.

That is why pricing teams should spend more time determining the right metric. Customers need to be able to understand it, budget for it and control their spend. Providers need a metric that tracks with value, scales economically, and works for billing.

The best pricing metric works for both sides.

Subscribe now

It has four phases: You start with quantifying value for a buyer to justify their investment. Then you create value for your customer by focusing on their success. Track value and outcomes against clear baselines, and finally articulate that value back to customers and the market.

When these phases work together, monetization becomes an engine that fuels growth where every customer outcome strengthens retention, expansion, and the next deal.

Here are that four phases, and what to focus on during each:

1. Quantify Value

To close a deal, Sales needs to convince a prospect that they get 3x-5x in potential value to justify their investment. Value often isn’t guaranteed, while spend is committed and certain. This is how B2B sales work. To convince a prospect to buy requires them to see they have a problem, and that you can solve the problem for them.

Sales Engineering / Presales quantifies the value a customer gets. To close a deal, they translate product capabilities into economic impact: Revenue lift, risk reduction, cost savings, or productivity gains. The problem is this is often done ad hoc. Value calculators standardize this across customers so reps don’t have to start from scratch for every deal. They ensure the main drivers of value are considered when negotiating deals, reducing discounting, increasing win rates and making sales more efficient.

Sales needs to convince the prospect that the value applies to them. Benchmarks and independent value studies give reps credible reference points, but prospects can question their applicability: Different industry dynamics, varying geographic needs, differences in company size. There are many reasons why they would not get the same ROI. Customer success stories from comparable customers who achieved real ROI help here. A product without success stories won’t survive very long.

That way you move from selling features and capabilities to quantifying and selling outcomes comparable customers have actually achieved.

Common pitfalls to avoid:

• Forgetting about the status quo: The prospect always has the option to change nothing. Change = risk. Always address the status quo in your ROI model, and make the case for change!

• Overly ambitious ‘benchmark’: Just because McKinsey or another reputable source states on average customers achieve something, doesn’t mean your customer believes this is achievable for them.

What works:

• Business case based on a value calculator → standardize how value is quantified

• Benchmarks & independent value stories → give reps credible reference points

• Customer stories → prove that value already happened elsewhere

2. Create Value

Once a customer buys, you need to deliver on what you promised, and time-to-value becomes critical. This is where many companies struggle. A customer bought based on expected business outcomes. If the implementation and adoption is not focused on achieving business outcomes, the buyer easily ends up disappointed.

Customer Success is tasked with making sure value is realized. They build a Success roadmap with the customer, align stakeholders, and accelerate adoption to get the customer to outcomes. But they can only do that if they know what outcomes Sales has sold. In large organizations, the handoff from Sales to CS is often a failure point.

Over time things also change, commercial context gets lost. The customer might have a reorg, or just normal attrition. Your CSM might have moved to a new opportunity, Sales has long moved on to the next pitch. Chances are your signed Order Forms are in some repository, but the pitch decks? Not very likely. Then CS is left guessing what business problem the customer was trying to solve. Make sure you have ONE value repository for all docs around value, be it the value model from Sales Engineering, the pitch deck and business case from Sales, or the Success roadmap from CS.

Common pitfalls to avoid:

• Treating onboarding just as a project: Timely setup and successful training are table stakes, but time-to-value means more, i.e. measurable business outcomes

• Sales and CS have separate siloed repositories: Handover between teams are THE most common failure points I’ve seen over the years in many environments

What works:

• Clear handoff from Sales to CS → preserve the business case and value model

• Success roadmaps → tie product rollout to measurable business outcomes

• ONE value repository → give Sales and CS visibility on value promised over time

3. Track Value

Aligning on outcomes is not enough. To prove value was created, you need to measure improvement against a baseline. If the first time you think about measuring impact is after the implementation is completed; or worse: during a renewal discussion, you are too late. You lost the chance to show evidence.

Customer Success needs to quantify success numerically. Outcomes need to have clear KPIs with targets so the finish line isn’t constantly moving. Stakeholders need to be aligned, especially when the buyer in the organization is not the same as the one overseeing the implementation - think of a BI tool used by Finance, but implemented and rolled out by the CIO org. The CFO signed off on the business case with certain outcomes to justify the ROI. Then the CIO organization manages your rollout. Different stakeholders, different targets.

It’s tempting to skip the baseline measurement when stakeholders ask for progress. You’re accelerating the implementation, training, and adoption to get the customer to value faster. The baseline? Gets skipped. How many staff did the team have when we sold ‘efficiency gains’? How long did the process take before we ‘accelerated revenue’? Improvements cannot be measured without a baseline though! Without it, value realization stays vague. And vague does not renew well.

Common pitfalls to avoid:

• No aligned baseline: Not documenting baseline measurements of your success KPIs, or measuring but not agreeing with the customer often leads to debates later

• Tracking activity instead of outcomes: Implementation progress and adoption are table stakes, but are not the same as outcomes or Return-on-Investment (ROI)

What works:

• KPIs & targets per outcome → define outcomes, quantify each and set a target

• Baseline measurements captured up front → set a reference point of what the KPI was before you started, so you can measure improvement later

4. Articulate Value

Once you created and measured value, articulate it back to the customer. Consistently, on an ongoing basis. If you do not communicate the impact, you are relying on the customer to connect the dots. Value not perceived might as well not exist.

Customer Success is usually responsible to protect ARR, sometimes to drive expansion. According to Gainsight’s Customer Success Index 2025, 62% of CS organizations utilize GRR as their primary metric, so they’re isolating expansion and focus purely on avoiding churn. Communicating value repeatedly, whether in QBRs or executive reviews, is the critical role CS plays to proactively address churn risk.

And for smaller customers that don’t have a CSM, value dashboards need to fulfill the same role. They need to focus on business outcomes, KPIs and progress against those. No Executive renews to increase weekly active users or other adoption metrics.

Marketing also needs to document proof points from one customer as inputs for the next prospects. Success stories with quantified outcomes, before-and-after metrics, and Executive’s quotes all strengthen the funnel and help close the next deal. That is how the Monetization Flywheel comes to spin. Value delivered to one customer becomes the proof point for the next one.

Common pitfalls to avoid:

• Adoption reporting: adoption metrics are easy via telemetry, and standard across customers. Not tying them back to outcomes invites questions on why to renew

• Treating customer proof points as marketing assets: Marketing usually collects “success stories”, but Sales and Presales also need them further down the funnel

What works:

• QBRs and executive reviews → translate product impact into business language

• Value dashboards → make progress visible for smaller customers

• Capturing success stories → turn customer outcomes into future proof points

The Monetization Flywheel that works

The Monetization Flywheel only works when value is treated as a continuous commercial discipline, not a one-time sale or pricing exercise. Companies that continuously quantify, create, track, and articulate value systematically are better positioned to reduce discounting, improve renewal rates and increase expansion.

In that sense, monetization is not just about price setting and price getting, which most pricing teams focus on. Implementing the Monetization Flywheel requires a lot more cross-functional collaboration, but the effort pays off to compound growth.

1) Naming conventions stayed similar to SaaS

Many SaaS companies had 4 tiers (with few exceptions like Salesforce having more).

“Free” or “Starter” was the entry tier for many SaaS companies. Though ”Starter” is not common for AI companies, “Free” remains the default entry tier name for more than two thirds of AI native companies such as ChatGPT Free and Claude Free.

“Pro” remains a very common name for a paid tier. In our dataset, over 80% of AI-native companies have a tier named “Pro” (e.g. Cursor Pro, Perplexity Pro).

“Business” persists as the most common name for the tier that has some limited business features. 50% of companies in our data set had a tier that contained “Business”, such as ChatGPT Business, Grok Business and n8n Business.

“Enterprise” remains the most common top tier name, e.g., Lovable Enterprise, Replit Enterprise. Companies still require SSO, admin controls, and centralized billing. It’s actually not easy to find a company that does not have an “Enterprise” tier.

2) Packaging added power user tier(s): Ultra/Max

AI-native companies also have generally an additional middle tier. This tier is aimed at power users (developers, analysts, researchers) and priced in the range of $100-$200 per month. The name varies with “Ultra” and “Max” being the most common at the moment, e.g. Cursor Ultra ($200), Claude Max 5x and Max 20x ($100 and $200).

In B2B SaaS, tiers were mainly differentiated by features and company size. “Land and Expand” sales motions often started with one team, expanding enterprise-wide later. AI adoption often starts with a few individuals experimenting too, but power users have high costs. The additional power user tier(s) capture higher prices earlier in the adoption cycle.

To maximize price capture in B2C, some AI companies also introduced a new lower-priced tier to monetize the many light users not willing to upgrade to Pro. ChatGPT Go ($8, ad-supported) promises 10x the usage of their Free plan, and Gemini AI Plus ($7.99) has twice the credits of Free. Both were launched globally in January 2026.

3) Differentiators shifted: features → usage

While in SaaS, tiers were mainly differentiated by features and company size, AI-native companies like Claude differentiate tier much more around “More usage” and “Priority access at high traffic times”. The reason is simple: AI products incur real compute costs, and power users can cause a significant cost far beyond what they pay, even with the new Ultra/Max tiers.

So usage must be controlled, and companies have two levers for this:

1. Throttle usage to indirectly ensure value scaling. Claude has these limits:

   1. Session limits: A limit for each session, which resets every 5 hours.

   2. Weekly limits: Weekly limits across all models + some model specific ones.

2. Meter usage directly: scale pricing with usage within a tier (often via credits), e.g.:

   1. Replit Pro pricing has a drop down to select between $100 and $4,000 (40x)

   2. Lovable Pro pricing allows selecting between 100 to 10k credits (100x)

Your operator takeaway

Even though AI changed the economics, the tiers naming did not radically change. AI-native companies kept what buyers were already familiar with (“Free”, “Pro”,
”Business”, “Enterprise”).

They evolved the tier structure and names to reflect their economic reality by adding one or two “Ultra” / “Max” tier(s) for power users, differentiating tiers with usage throttling, and allowing to purchase metered usage (up to 100x) within a tier.

I touched these three levers before in my post on Discounting as a Growth Lever. This is the deeper benchmark cut focused on billing and contracting discounts.

Across the pricing pages I reviewed, billing and contracting discounts show up in three forms.

1) Multi-year vs annual: Churn reduction

Multi-year commitments are usually still paid annually. The customer typically does not prepay 3 years up front. The main benefit is churn reduction through fewer renewal moments.

A 3-year commitment means the customer questions the purchase 67% less often. Once every 3 years instead annually. A 2-year commitment still reduces the decision points by half.

One observation before the numbers. Very few B2B companies publish multi-year discounts on pricing pages. That makes the few that do publish them unusually valuable. If you work in pricing, these data points are gold.

In my dataset, the median for multi-year vs 1-year discount is 20%.

3 examples:

• Codeanywhere Premium: 13% ($23 per month on 2-year vs $20 on 1-year)

• TaxDome Pro: 10% ($900 per year on 3-year vs $1,000 on 1-year)

• SEO PowerSuite Professional: 32% ($239 per year on 3-year vs $349 on 1-year)

2) Annual vs monthly: Cash flow optimization

In most B2B products, “annual” typically means annual contracting plus annual prepayment. This is a working-capital lever as much as it is a commercial one.

Annual prepay improves cash conversion and lowers billing friction. On top of getting the cash up front, your Finance team has to issue and reconcile 92% less invoices.

In my dataset, the median discount for annual prepayment is 17%.

3 examples:

• Cursor Pro: 20% ($16 p.m. billed annually vs $20 monthly)

• Replit Core: 20% ($20 p.m. billed annually vs $25 monthly)

• GitHub Copilot Pro: 17% ($100 annually vs $10 monthly)

3) Monthly vs Paygo: Behavioral economics

Monthly subscriptions (vs annual commitments) are much less predictable and create a lot of uncertainty for AI companies like Lovable. Their Growth leader Elena Verna recently said this:

Monthly subscriptions (vs paygo) don’t change that, but they use the same behavioral economics pattern that makes gym memberships so successful. People overestimate future usage. Paygo forces you to feel the pain of EVERY single use. A subscription removes that friction, makes continued usage feel like a gain (since you’ve already prepaid), and charges even if you use it less.

In my dataset, the median discount here is 21%.

3 examples:
• Lovable: 17% ($25 for Pro including 100 credits vs $15 for 50 top-up credits)

• Luma AI: 20% ($20.99 for Plus including 10k credits vs $4 for 1,200 credits)

• ElevenLabs: 27% ($22 for Creator including 100k credits vs $0.30 for 1k credits)

Stacking discounts

Buyers (and you Account Executives) stack these billing and contracting discounts. Multi-year plus annual prepay plus commit instead of paygo.

Using the percentile table, the implied 3-year vs paygo discount lands at 48% at the median. Top quartile is even 58%.

Most companies don’t publicly offer all three levers, but many B2B companies already discount annual and multi-year commitments. If you layer AI credit pricing on top, make sure you understand unit economics well, and define stacking rules clearly.

To capture your fair share of the value you create, you need to:

1. Explain value in a way a buyer can quantify quickly.

2. Personalize it so the buyer see this is my situation.

3. Validate it with evidence the buyer trusts.

Build a simple chain: value calculation → relevance → proof. Here are the three key tools to do it, with practical guidelines and real-world examples:

• ROI calculators

• Customer success stories

• Independent value studies

1) ROI calculators

You need to articulate value with an ROI calculator. A good calculator teaches the value model, then shows the math with the buyer’s inputs, so they can see why buy.

A great example is ServiceNow’s Value Calculator: it

• has 4 inputs to tailor value to a prospective buyer

• Articulates value broken down into 3 buckets

Another one is Crowdstrike’s Cybersecurity Consolidation Value Calculator:

• Has 5 inputs to tailor value to the specific buyer

• Articulates value in 4 buckets

The best ROI calculators follow these practices:

• Focus on 3-4 business outcomes (e.g. cost saved, risk avoided, revenue gained)

• Design the formula (e.g. savings = incidents * time/incident * % reduction)

• Limit to 3-5 inputs (e.g. # incidents, time/incident, team size)

• Use defensible default assumptions (e.g. industry ranges) for inputs

• Pre-fill with typical values so the buyer can edit instead of starting blank

Common pitfalls to avoid

• Hidden assumptions / unrealistic savings → ROI inflated, not credible

• Too many outputs → Lack of focus

• Too many inputs → The prospect moves on

2) Success Stories with KPIs

Make the value feel specific with customer stories that show achieved KPIs. Case studies are not just proof of value, they make value relevant. The buyer is asking: “Is this my world?”

Your story library should:

• Lead with KPIs and time to value
  (e.g. “Reduced onboarding from 14 days to 5 days in 90 days.”)

• Let prospects filter to match their situation

A great example are Snowflake’s 342 stories that can be filtered by industry, product categories, department and region.

Each success story then shows the top 2 KPIs upfront. On the left, they show metadata conveniently: the industry, location and product used. Here is an example:

Many leaders do this well (e.g. Salesforce 231 success stories, Adobe’s 304 success stories) and allow filtering by product and industry.

Buyers are not going to scroll through dozens of stories, so filters and metadata matter! Think about how value varies, and consider these ones in this order:

• Industry

• Business size

• Region

• Product or use case

• Buyer department or persona

• Tech stack or integration context

Common pitfalls to avoid

• Soft outcomes only → Lack of KPIs signals ROI is not tracked

• KPIs not business-critical in industry → Hard to establish why it matters

• Stories with no KPIs → Buyers want to know if they get the ROI you articulated

3) Independent Value Studies

The final step is to build trust through independent value verification. Buyers discount vendor claims. Third-party validation reduces that discount rate. It also gives internal champions something they can cite without looking biased.

Here are three companies offering them:

Forrester’s Total Economic Impact (TEI) studies, e.g. ServiceNow, Zoom

IDC Business Value (BV) white papers, e.g. Microsoft Dynamics 365 for Manufacturers, AWS

Nucleus Return on Investment (ROI) studies, e.g. SFDC at Pearson, Oracle at SNFC

Zoom’s TEI study shows the benefits over time by type, calculates ROI and payback period. The most important part is it is independent and Forrester states how many customers they interviewed, and what type of customers these are (see example).

Independent value studies help you defend assumptions in your ROI calculator by establishing a credible value range for a category and solution type, verified by a trusted third party. It arms your sellers and champions with language their finance and procurement teams accept.

How to best use third-party studies:

• Position the study as a benchmark (not a promise), a note its limitations

• State what must be true for similar outcomes (baseline, adoption, scope)

• Use it alongside your ROI calculator + success stories to increase credibility

Common pitfalls to avoid

• Not acknowledging ranging outcomes → Perception it might be cherry picked

• Citing ROI with no context → Prospect may not believe it applies to them

The system that works

The three components reinforce each other:

• ROI calculator: “Here is the value model with your inputs.”

• Filtered KPI stories: “Here are customers like you who achieved it.”

• Independent verification: “Here is a third party validating the results are real.”

If you’re just starting

1. Build your calculator first and test with your next pitch (with buyer’s inputs)

2. Turn early wins into KPI stories once value is realized

3. Commission an independent study only after you have traction (they’re expensive)

Use this guide to design your pricing architecture to support predictable revenue and profitable growth in your reality:

• Is your customer’s demand seasonal?

• Do “power users” consume multiples of everyone else?

• Is usage hard to forecast for customers, and your marginal costs are high?

• Do you incur meaningful costs even for small accounts?

• Do your enterprise customers need 24x7 expert support and strict SLAs?

• Do your customers want budget predictability?

Recurring Subscription Components

1) Committed Volumes (Users, Usage or Outcomes)

A recurring contract with committed volumes in a chosen metric. The defining feature is committed recurring revenue, not the unit of measure. Recurring credit bundles can fit here if they are contracted and revenue-recognizable under ASC 606.

Examples:

• Lovable Pro is $25 for 100 credits per month

• Zendesk AI Agents are $1.50 per committed automated resolution

2) Platform Fee

A fixed recurring fee that captures baseline value and helps cover non-scaling costs. There are 3 common variants:

• 2-part tariff: platform fee does not include any usage

• 3-part tariff: platform fee includes base allowance, overages are billed

• All inclusive: unlimited use, protected by rate limits or a fair use policy

Examples:

• NextMatter charges 7,500 EUR per month, in addition to per user charges

• Basecamp Business is $299 per month all-inclusive

3) Support & Success

Paid support subscriptions layered on top of licenses. These are typically priced as a % of product spend and tied to access, SLAs, and expert resources.

Examples:

• Microsoft Unified Enterprise Support is ~7.5-10% of product spend, min $50k

• Salesforce Premier Success Plan is 30% of license fees

One-time Revenue Components

4) Up-front Services

One-time services for onboarding, integration, testing and training.

Examples:

• Factor A/E charges a one-time implementation fee plus subscriptions

• Link2Feed requires a one-time implementation fee on top of its subscription

5) Usage-based Consumption

Pay-as-you-go with no upfront volume commitment. These are common for variable or AI usage and value-added services.

Examples:

• OpenAI API calls are priced per 1M tokens and usage is billed monthly

• Amazon Bedrock Flows: $0.035 for 1k node transitions, metered & billed monthly

6) Migration Services

One-time services to extract, transform, or move data, or fees for data egress or assisted migration.

Examples:

• MongoDB Atlas charges $0.125 per GB exported plus transfer fees

• AWS Snowball Edge export jobs are priced per device fee plus shipping

Closing Thought

Let’s go back to our questions from the beginning. Here is how you can design you pricing architecture so revenue is predictable and your business grows profitably:

• Is your customer’s demand seasonal? → Use subscriptions with annual caps

• Do “power users” consume multiples of everyone else? Offer pooled allowances

• Is usage hard to forecast for customers, and your marginal costs are high? → Offer committed credits with mid-term top-ups (pre-priced incremental add-ons)

• Do you incur meaningful costs even for small accounts? → Add a platform fee to cover non-scaling cost

• Do your enterprise customers need 24x7 expert support and strict SLAs? → Consider an Enterprise Support priced as a % of ARR with a minimum

• Do your customers want budget predictability? → Price committed volumes significantly lower than pay-as-you-go rates

Which tactics companies choose to turn volatile AI usage into more stable, recurring revenue comes down to answering two questions:

1. Who bears the risk? Do you shift volatility risk to the customer, or absorb it?

2. How do you control it? With contractual terms or in-product levers?

Drawing this as a 2x2 matrix shows four combinations, and all 4 are gaining traction:

1) Committed capacity at a Premium
(vendor-friendly, commercial)

In SaaS, committed volume meant discounts. In AI, committed capacity can cost more because it comes with reserved performance.

OpenAI Scale Tier pricing offers higher reliability and better latency. Under standard pay-as-you-go API pricing, 1M input tokens for GPT-5.2 cost $1.75. With Scale Tier, customers purchase 25,000 “tokens per minute” capacity for $105 per day. In a best-case utilization scenario (100% utilization for all 1,440 minutes of a day), this is a 67% premium over PAYG API pricing.

Amazon Bedrock’s Provisioned Throughput similarly lets customers provision a higher level of throughput at a fixed cost. Longer commitments reduce price, but the core idea is the same: Pay for reserved capacity, not just consumption.

2) Auto-upgrades as default
(vendor-friendly, in product)

Auto-upgrades when credits are exhausted lead to an immediate tier jump. It prevents disruption, but locking the customer into a higher commitment by default can cause backlash.

Sleekflow’s AI Agent Flow Builder has auto-scaling enabled by default. It automatically upgrades the subscription to the next pricing tier when monthly flow enrollment credits are reached. It is framed as preventing disruption, and the upgraded tier remains moving forward.

HubSpot made auto-upgrade the default for HubSpot credits in December 2025. Once purchased credits are exhausted, accounts are automatically upgraded to a higher Credits pack for the rest of a commitment term. Enterprise customers can opt out of auto-upgrades to pay for usage overages instead.

3) AI credit pooling & controls
(buyer-friendly, in product)

Shared credit pools smooth variance across users. Heavy users draw from the same entitlement as light users. Admins get visibility, guardrails and governance.

Figma announced it will allow admins to purchase a shared pool of additional credits starting March 11, 2026. Approved users can draw from the pool when they exceed AI credits included in their seat, and admins can track team-wide credit usage in their dashboard.

Palo Alto Networks includes pooled flexibility and burst protection for its Next-Generation Firewall via pooling credits across deployments and clouds, and averaging usage over a month.

4) Burst protection against usage spikes
(buyer-friendly, commercial)

Burst protection waives charges for short spikes. It typically works by excluding peak days, ignoring extreme daily overages, or capping billable overages, usually with limits on frequency or duration.

Honeycomb’s burst protection does not count events on days where usage exceeds daily event targets by more than 2x. This can be triggered up to 3 times in a calendar month.

Login VSI offers 1 month of burst protection per year. Customers can scale up to 2x contracted usage at no extra cost.

Jetpack Stats excludes the 2 highest traffic days each month. If the customer exceeds the limit in three consecutive months, they require an upgrade to a higher tier.

5) Backdating upgrades
(two flavors, commercial)

Backdating upgrades retroactively applies a higher tier so the customer pays for a higher commitment instead of one-time overage fees.

Customer-requested backdating (forgiveness): Acquia Cloud Platform offers a 30% capacity buffer before overages are invoiced. Customers can backdate the effective date of an upgrade to avoid one-time overage fees above the buffer. They also offer full overage protection for the first 3 months of production launch.

Audit-driven retro tiering (enforcement): Redress Compliance reports that SAP SaaS audits check Fieldglass usage and sometimes push customers into a higher subscription tier retroactively.

Closing Thought

Volatility is a real problem for B2B AI buyers. If you want to reduce tension and churn risk, stop treating surprise usage bills like a punitive tax.

Use these tactics based on your comfort to take on risk:

• Committed capacity when reliability is the value, price it like an SLA.

• Auto upgrades to prevent disruption, but beware of buyer pushback if it feels like a trap.

• Pool credits to smoothen seat-level variance, only when paired with admin controls.

• Burst protection to separate true growth from temporary spikes.

• Backdating to reframe overages into a win-win commitment.

Thanks for reading, and see you in the next one.

Funding builds options. It keeps architectures flexible and narratives open.
M&A hardens decisions, collapses layers, and locks in the control layers where leverage compounds, switching costs rise, and architectural choices become durable.

Those control layers are infrastructure, identity, data, and experience.

1) Vertically integrated AI infrastructure

(Simulation → silicon → networking → compute)

Deals:

• Synopsys acquired Ansys (AI simulation and chip design) for $35B in July 2025.

• Hewlett Packard Enterprise acquired Juniper Networks (AI-native data center networking) for $14B in July 2025.

• CoreWeave acquired Core Scientific (High-density AI compute) for $7B in July 2025.

What is converging:
AI workloads are forcing tight coupling across layers that were historically modular. Simulation, chip design, networking, power, and compute are collapsing into a unified layer. Scale is becoming the competitive moat.

Why it matters:
This signals the end of best-of-breed infrastructure for frontier AI. The winners own the entire infrastructure stack, from silicon to compute.

2) Agent identity & trust

(Human identities → agent identities)

Deals:

• Alphabet acquired Wiz (Cloud and AI-native security posture management) for $32B in July 2025.

• Palo Alto Networks acquired CyberArk (Identity for machine and agent access) for $25B in July 2025.

What is converging:
Identity security, cloud posture, and runtime protection are merging into agent-aware trust layers. The perimeter is no longer users or workloads. It is autonomous AI agents acting independently.

Why it matters:
This is the creation of control infrastructure for non-human identities as agents inside enterprises require new controls.

3) End-to-end data

(Raw data → labeled data → real-time inference)

Deals:

• Meta Platforms invested $14B in Scale AI (Data labeling for AI applications) in April 2025.

• IBM acquired Confluent (Real-time data streaming for AI systems) for $10B in May 2025.

• Salesforce acquired Informatica (Enterprise AI data governance) for $8B in June 2025.

What is converging:
Data ingestion, labeling, governance, and activation are collapsing into AI-ready data layer. The distinction between operational data and big data is disappearing.

Why it matters:
Owning the data lifecycle becomes more valuable than owning models. Models churn. Data moats compound.

4) AI experience and interactions

(Dashboards → continuous, ambient AI)

Deals:

• Qualtrics acquired Press Ganey Forsta (AI experience measurement) for $8B in June 2025.

• OpenAI acquired io Products (Always-on AI device) for $6B in September 2025.

What is converging:
Experience measurement, feedback loops, and interaction surfaces are becoming continuous and embedded. Hardware, software, and inference are merging into always-on experience systems.

Why it matters:
The experience is shifting from UX to continuous presence. AI is no longer something you open as an app. It is always on and with you.

5) The July 2025 deal cluster & the meta-convergence

More than half of the top 10 AI deals in 2025 closed in July alone. I had posted about the spree at that time here. The July cluster was structural, because by mid-year three things had changed:

Cost for compute and inference were better understood. Stanford’s 2025 AI Index Report from April 2025 found hardware costs declined ~30% annually, and energy efficiency improved by ~40% each year.

Confidence in enterprise demand for AI showed and Nvidia’s stock price increased 70+% from low of $94 in April 2025 into new all time highs of $160+ for the first time.

Buyers had H1 performance visibility and board-level approval to commit capital at scale. AI wasn’t experimental any longer.

That made future cost for real use cases predictably cheaper, increased confidence that growing demand is real, and allowed large transactions to act swiftly.

Closing Thought

The 2025 M&A wave showed where power actually compounds. Point solutions get absorbed. Control layers become the long-term profit centers.

If you are building or investing in AI, the critical question is which layer you own, and how durable that position is as the stack continues to consolidate.