LLenny's Podcast
← All frameworks
Sales

Outcome-Based Pricing Qualifier

If the work is autonomous and the result is measurable, price the outcome — not the tokens.

Difficulty
Moderate
Time to result
~months to results
Steps
4
Confidence
90%

A pricing philosophy for AI products: charge for the business outcome the software achieves rather than for usage, seats, or tokens. It applies when two conditions hold — the agent does a job autonomously and the outcome is measurable — which aligns vendor and customer as partners rather than as vendor and line item. Usage-based pricing (tokens) can be actively misleading because effort and value aren't correlated.

Origin

Bret Taylor's model at Sierra, which prices on resolved customer-service interactions; he ties it to Madhavan Ramanujam's outcome-pricing thesis and Sierra being cited as an example.

Core principles

  • 01Outcome pricing requires autonomy plus measurability
  • 02Tokens/usage measure effort, not value — and effort can even be negative value
  • 03Charging on outcomes makes you a partner, not a vendor
  • 04A measurable, autonomous job is a self-evident productivity driver, so it can be valued directly

How to run it

  1. 1

    Confirm the agent works autonomously

    Verify the software actually accomplishes the job on its own, not just making a human incrementally faster. Autonomy is what makes the value self-evident.

    Watch out Productivity software that only makes people '10% more productive' is notoriously hard to sell because the value is unattributable.

  2. 2

    Confirm the outcome is measurable

    Define a concrete, countable outcome — e.g., a resolved/contained support call, a closed sale, a produced pull request. If you can't measure it, you can't price on it.

    Pro tip Anchor the price to the cost being displaced — a human-answered call costs ~$10–20, so a deflected call has a clear reference value.

    Watch out A long, effortful interaction that fails still has zero or negative value — don't price the effort.

  3. 3

    Pre-negotiate a rate per outcome

    Agree a rate the customer pays only when the defined outcome is achieved (a resolution, a sales commission), aligning your business model with theirs.

  4. 4

    Reorient the whole company around delivering outcomes

    Because you're only paid when customers succeed, the company must be built to help them achieve the outcome — deeply customer-centric, not 'throw software at the wall.'

    Watch out You'll never get paid if the outcome doesn't happen, so this pricing demands a different company shape and operating posture.

In the wild

Sierra's resolution-based pricing

Sierra's agents (Harmony for Sirius XM, agents for ADT, Sonos) handle customer service. A human-answered call costs ~$10–20, mostly labor. When the AI agent resolves the customer's problem without a human picking up — a 'containment' or 'deflection' — the customer pays a pre-negotiated rate.

Pricing aligned to saved cost per resolved interaction; Sierra cited as a leading example of outcome-based pricing working in practice, with high CSAT (e.g., a Weight Watchers agent at 4.6/5).

Common mistakes

Pricing on tokens or usage

Tokens measure effort, not value — like the manager who counted lines of code and got a negative number after a refactor. A high-token session that fails to solve the problem delivered no value, or negative value.

Applying outcome pricing where the outcome isn't measurable or autonomous

If the software only assists a human or the result can't be attributed and counted, outcome pricing collapses into the same unattributable value problem that makes productivity software hard to sell.

Is it for you?

Best for

AI/agent companies selling autonomous, outcome-measurable software into enterprise buyers

Not ideal for

Assistive tools that augment humans without an isolable, measurable outcome

From the transcript

these agents need to be autonomous and the outcome has to be measurable. That's not always possible, but I think it's broadly possible

1:06:00

there's a pre-negotiated rate for that uh and that's we call it like resolution based

1:06:00

yeah, you used a lot of tokens, like good for you. Did the, you know, did it produce a poll request

1:07:00

every technology company aspires to be a partner, not a vendor

1:08:00

From the episode

He saved OpenAI, invented the “Like” button, and built Google Maps: Bret Taylor on the future of careers, coding, agents, and more