5 guests 5 episodes 2,261 words

How to Charge for AI: The Pricing Model Decision That Nobody Has Solved Yet

Should AI products charge per seat, per usage, or per outcome?

The AI pricing question is urgent because the wrong model does not just leave money on the table -- it actively distorts your product incentives. Charge per seat and your AI agent has no reason to be efficient. Charge per usage and customers fear runaway bills. Charge per outcome and you need to define what "success" actually means. Each model shapes not just revenue but the product itself.

Bret Taylor, the co-founder of Sierra who also co-created Google Maps, served as CTO of Meta, co-CEO of Salesforce, and now chairs OpenAI's board, says outcome-based pricing is "so obviously the correct way to build and sell software." Madhavan Ramanujam, a senior partner at Simon-Kucher who has led pricing engagements with over 400 companies and 50 unicorns, says the model matters more than the price. Krithika Shankarraman, who was the first marketing hire at both OpenAI and Stripe, says nobody has actually solved this yet. Meanwhile, Intercom's Fin is printing money with 99-cent-per-resolution pricing and Cognition charges by Agent Compute Units.

As AI transforms what software can do -- from assisting humans to autonomously completing work -- which pricing model captures value most effectively: per-seat subscriptions, usage-based pricing, or outcome-based pricing?

The 5 Positions

1 / 5

The Model Matters More Than the Price

Madhavan Ramanujam

Madhavan Ramanujam, senior partner at Simon-Kucher and author of Monetizing Innovation, has a deceptively simple insight that most founders miss. After working with over 400 companies on pricing strategy, he argues that founders obsess over the wrong question entirely.

"We usually say how you charge is way more important than how much you charge."
Madhavan Ramanujam
▶ 01:00:04

The pricing model must match how customers derive value. Ramanujam illustrates this with Michelin: in a commodity tire market where a 20% price premium was impossible, Michelin shifted from per-tire to per-mile pricing. Truckers loved it -- they could pass costs through as a variable expense invoiced per journey. The product was identical; only the model changed. But that change unlocked demand a premium price never could have.

He extends this to B2B SaaS with Segment (before the Twilio acquisition). Segment priced on API calls, but customers did not think about value in terms of APIs. They thought about monthly tracked users. Switching the pricing metric -- while keeping the product identical -- aligned the price with how customers perceived value.

Critically, Ramanujam warns against chasing whatever model is fashionable. When subscription dominated, everyone went subscription. When Snowflake made usage-based famous, everyone rushed to usage. Each model has specific conditions where it fits:

"If they say, 'Okay, I'm going to price based on usage,' that would be dramatically wrong pricing model. So in those cases, subscription makes sense, or simplifying the pricing conversation is to your advantage."
Madhavan Ramanujam
▶ 01:04:53

He uses LifeLock as the counterexample: a product with episodic usage (you only need it when your identity is compromised) but ongoing value (protection runs constantly). Usage-based pricing would be dramatically wrong there, because it would charge only when the worst happens.

When This Applies

when you need stable base revenue plus upside from outcomes. HubSpot combines a subscription base with usage overage. Many companies in transition from copilot to agent use a seat-based floor with outcome-based upside. Works when different customer segments derive value differently.

Outcome-Based Pricing Is Obviously Correct

Bret Taylor

Bret Taylor makes a structural argument, not a technical one. His painful experience building Quip, a productivity docs company he sold to Salesforce, taught him that selling tools that make humans faster is extraordinarily difficult.

"The whole market is going to go towards agents. I think the whole market is going to go towards outcomes-based pricing. It's just so obviously the correct way to build and sell software."
Bret Taylor
▶ 00:00:07

"It's so hard to sell productivity software, which I learned the hard way."
Bret Taylor
▶ 00:00:21

When AI runs autonomously and you can measure its impact, outcome-based pricing becomes the only model that perfectly aligns vendor and customer incentives. Sierra, Taylor's AI agent company, charges for customer support outcomes resolved autonomously. No resolution, no charge. This creates a virtuous flywheel: better AI produces better outcomes, which generates more revenue, which funds further AI improvement. Per-seat pricing for productivity tools lacks this self-reinforcing loop because value capture is indirect.

The hidden prerequisite is attribution. You can only charge for outcomes if you can prove the AI caused them. In domains where human and AI contributions are entangled -- creative work, strategic decisions, multi-step processes -- isolating the agent's contribution is genuinely difficult.

When This Applies

when AI augments an existing workflow already priced per seat. Microsoft Copilot adds AI to Office 365 seats. Notion AI layers onto existing per-member pricing. Works when value is proportional to headcount and procurement teams need predictable costs.

99 Cents Per Resolution Proves It Works at Scale

Eoghan McCabe

Eoghan McCabe, who founded Intercom in 2011 and returned as CEO to lead its AI transformation, provides the proof point that outcome-based pricing works not just in theory but at massive scale. His advocacy comes from hard-won experience with pricing failure -- Intercom's old pricing was so confusing it became a viral meme on Twitter.

"We charge 99 cents to resolve tickets, customer problems, and we have a higher resolution rate than anyone else, and we are proud of that and we obsess over that. It is the metric by which these agents are assessed, and we wanted our revenue to be 100% aligned with the value that they attained because we had all this scar tissue from pricing prior that felt unfair to customers."
Eoghan McCabe
▶ 00:23:33

When Fin first launched, each resolution cost 120 cents to run. Intercom was losing money on every transaction. But McCabe bet that costs would drop and that pricing should come from value, not costs:

"I always believe that pricing should come from value and not from costs. The cost is our problem."
Eoghan McCabe
▶ 00:25:20

The bet paid off. SaaS businesses were spending $20-30 per human-resolved ticket. At 99 cents, Fin's value proposition was overwhelming. On the podcast, Lenny Rachitsky connected Intercom's model directly to Ramanujam's framework, calling it the "magic quadrant" of pricing advice: beautifully simple and outcome-based. McCabe's response captures the relief of a founder who lived through years of pricing wars: "Thank God our pricing wars are over."

Fin is now on track to exceed $100M ARR in under three quarters -- a growth rate that dwarfs Intercom's traditional SaaS business.

AI Pricing Is Genuinely Unsolved

Krithika Shankarraman

Krithika Shankarraman, who was the first marketing hire at both OpenAI and Stripe (sole marketer at Stripe for three years) and early marketing leader at Retool and Dropbox, offers the most honest assessment from someone at the frontier.

"The reality is it's not a solved problem. And a lot of folks, a lot of companies in the AI domain are trying to figure out the right pricing model. There is a value creation aspect to using AI that doesn't kind of neatly fit the mold of SaaS-based pricing or seed-based pricing, or even usage-based pricing."
Krithika Shankarraman
▶ 00:08:47

She specifically flags agents as the complicating factor. How do you price AI workers? What is the unit of completion for a code generator? For an AI marketer?

"It's going to be a Wild, Wild West before we come up with something that is as internalized now as seed-based pricing or usage-based pricing."
Krithika Shankarraman
▶ 01:01:30

Her framing gives permission for experimentation. She points to OpenAI's own journey -- from Sam Altman's famous quip "At some point we will ask ChatGPT how to make money" to a monthly subscription that just works -- as evidence that the answer sometimes emerges through iteration rather than analysis.

When This Applies

when the AI acts autonomously and outcomes are measurable. Intercom's Fin charges 99 cents per resolved support ticket. Sierra charges per successful customer interaction. Works when you can absorb revenue variability and when the outcome metric is unambiguous.

Usage-Based Pricing for Autonomous Agents

Scott Wu

Scott Wu, CEO of Cognition, the company behind Devin (the first autonomous AI software engineer), uses Agent Compute Units (ACUs) as a usage-based metric -- a pragmatic middle ground between seat-based and outcome-based pricing.

"We have since the beginning been laser-focused on agentic coding, and that is the one thing that we've really believed in. It's the one thing that we've designed for and that goes all the way to even the revenue model with ACUs and having the usage-based setup."
Scott Wu
▶ 00:56:27

ACUs track the computational work Devin performs -- a proxy for effort rather than result. This is a pragmatic choice: coding outcomes are notoriously hard to define (what constitutes "one feature"?), but compute consumed by an autonomous agent is trackable and transparent. You delegate a task via Slack or an issue tracker, Devin runs autonomously for however long it takes, and you pay for the compute it consumed. Harder tasks cost more, creating rough value correlation.

When This Applies

when AI consumption varies dramatically and outcomes are hard to define. Cognition's Devin charges per ACU for autonomous coding tasks. AWS and cloud infrastructure providers charge per compute unit. Works when the unit of work is clear, low commitment matters, and customers demand transparency.

Evidence from the Archive

Simon-Kucher

HubSpot as a hybrid model combining subscription base with usage overage

Michelin shifted from per-tire to per-mile pricing, unlocking a commodity market by matching the monetization model to how truckers derive value

Madhavan Ramanujam · Senior Partner ▶ 01:00:04

Intercom

At launch, each resolution cost 120 cents but McCabe bet costs would drop -- they did

Intercom Fin charges 99 cents per resolved customer support ticket, on track to exceed $100M ARR in under three quarters

Eoghan McCabe · Founder & CEO ▶ 00:23:33

Sierra

Sierra charges for customer support outcomes resolved autonomously by AI agents

Quip (Taylor's productivity startup) as evidence of how hard it is to monetize productivity tools on a per-seat basis

Bret Taylor · Co-founder & CEO ▶ 00:00:07

OpenAI / Stripe / Retool

OpenAI's pricing journey from 'we'll ask ChatGPT how to make money' to a working subscription model

ChatGPT's monthly subscription that 'just works' despite not fitting traditional SaaS pricing theory

Krithika Shankarraman · VP Marketing; Executive in Residence at Thrive Capital ▶ 00:08:47

Cognition (Devin)

Devin is reportedly writing 50% of Cognition's own code, validating the autonomous agent model

Devin uses ACUs (Agent Compute Units) as its billing metric for autonomous coding tasks

Scott Wu · CEO ▶ 00:56:27

The Synthesis

These five voices reveal a pricing landscape that is more structured than it appears. Apply Ramanujam's framework and a clear decision tree emerges -- but only after you answer the right questions.

Model Before Price

Why do founders agonize over the wrong pricing question?

Product Architecture Match

Which pricing model fits which product type?

Copilot-to-Agent Trend

How will AI product evolution change pricing models?

The pricing model decision must be made before the price-point decision. Most founders agonize over '$29 or $49 per month?' when the real question is 'per seat, per usage, or per outcome?' Getting the model wrong makes any price point wrong.

Seat-based works when AI augments humans already paying. Usage-based works when AI operates autonomously and consumption varies significantly. Outcome-based works when AI impact is clearly measurable -- like Intercom's $0.99 per-resolution pricing. Each has distinct structural advantages and downsides.

As AI moves from copilot (human-in-the-loop) to agent (autonomous), the pricing model naturally shifts from seat-based to outcome-based. But the transition will be messy, with hybrid models and experiments, before the market standardizes.

Which Approach Fits You?

Answer 3 questions about your situation. We'll match you to the right approach.

Question 1

How does your AI product deliver value?

Question 2

What do your customers care about most in pricing?

Question 3

Can you absorb revenue variability?

Notable Absences

The Bottom Line

The trend line is clear: as AI moves from copilot (human-in-the-loop) to agent (autonomous), the pricing model naturally shifts from seat-based to outcome-based. But Shankarraman's "Wild, Wild West" assessment tempers this -- the transition will be messy, with hybrid models and experiments, before the market internalizes a standard.

Lenny's newsletter on freemium versus trial adds another dimension: free is an acquisition strategy layered *on top of* a monetization model, not a replacement for one. Microsoft Teams disrupted Slack by bundling free chat into Office 365 (seat-based). Robinhood disrupted trading with free trades (transaction-fee-based). The pricing model and the free-tier decision are independent choices that can be combined.

Sources

Madhavan Ramanujam — "Pricing your AI product: Lessons from 400+ companies and 50 unicorns | Madhavan Ramanujam" — Lenny's Podcast, July 27, 2025
Bret Taylor — "He saved OpenAI, invented the “Like” button, and built Google Maps: Bret Taylor on the future of careers, coding, agents, and more" — Lenny's Podcast, July 31, 2025
Eoghan McCabe — "How Intercom rose from the ashes by betting everything on AI | Eoghan McCabe (founder and CEO)" — Lenny's Podcast, August 21, 2025
Krithika Shankarraman — "Growth tactics from OpenAI and Stripe’s first marketer | Krithika Shankarraman" — Lenny's Podcast, May 25, 2025
Scott Wu — "How Devin replaces your junior engineers with infinite AI interns that never sleep | Scott Wu (Cognition CEO)" — Lenny's Podcast, September 8, 2025