Pricing and packaging · Answered by Golden Section from more than 400 B2B software companies observed
Price AI separately once you know who controls the outcome it produces; until then, put a rate-limited allowance in an existing tier and lift that tier's price. Every customer buys an outcome. If results depend heavily on how the customer uses the AI, charge for usage, such as tokens or documents processed. If they depend on your product, charge for the outcome itself, such as a resolved ticket or fraud caught. Carry inference as cost of revenue, launch any change on new business first and at renewal second, never mid-contract, and expect a correct move to cost a few quarters of growth optics. Track AI inference revenue and inference cost on their own lines from the start. Next step: write the one-sentence outcome your AI moves, and note whether the customer or your product controls it.
Separate pricing follows knowing who controls the outcome: usage pricing when the customer does, outcome pricing when your product does. Until that is proven, put a rate-limited allowance in an existing tier with a price lift on that tier, and absorb inference as cost of revenue. Sequence any change to the existing base behind new business, and decide before launch what you will tell the board about the growth it may cost.
| Metric | Value | What it means | Source |
|---|---|---|---|
| Choosing the AI pricing unit | customer-dependent outcome: usage; vendor-dependent outcome: outcome; unproven: rate-limited allowance in a tier with a price lift | pricing unit chosen by who controls the result the customer buysall customers buy outcomes; the question is who drives them | Golden Section operating viewGolden Section operating view |
| Companies pricing AI on outcomes | 4 of 24 | software companies with an explicit, on-record move to outcome pricingQ2 2026 earnings season; folding AI into existing seats and tiers was the quiet majority position | Golden Section, publishedWhere the AI Budget Lands |
| Cost of a pricing transition | stock down about 19% on 20% revenue growth | HubSpot's share-price reaction after re-basing AI pricing on outcomesoutcome pricing lengthened sales cycles and reset near-term seat growth, even as it aligned price to value | Golden Section, publishedWhere the AI Budget Lands |
| AI product gross margin | roughly 53% (2026); we underwrite 50–60% | gross margin on AI-delivered productsICONIQ State of AI, as cited in our addendum; traditional SaaS runs 76–80% | Golden Section, publishedInvesting in Software addendum, September 2026 |
| Inference share of spend | about 23% | inference cost as a share of total spend at the scaling stageICONIQ via our addendum; cost of revenue now varies with usage | Golden Section, publishedInvesting in Software addendum, September 2026 |
The AI pricing model play sorts the options into five: outcome, bundled, usage, subscription plus credits, and absorbing the cost without repricing. None is right everywhere. Outcome pricing is the most defensible when you can draw a line from price to a number the customer already tracks, and the hardest to execute honestly. Bundling, as a rate-limited allowance in an existing tier with a price lift on that tier, is the lowest-risk choice while you learn whether the outcome depends on the customer or on you. When the customer controls it, usage pricing is fair; when your product does, outcome pricing is. Usage pricing survives only when the customer wants more of what is metered. Our review of the Q2 2026 earnings season found no convergence, and the clearest lesson came from the company that moved: even a correct pricing change can reset growth for several quarters.
Margin is the second reason to decide deliberately. AI-delivered revenue runs well below traditional software margins, and buyers will inspect whether AI revenue scales and sticks. You can answer only if AI inference revenue and inference cost are tracked as separate lines, with the question our addendum asks in diligence ready: what happens to gross margin at three times today's usage?
Willingness to pay comes from the customer ROI you already sell. AI that moves that number can carry a price. AI that does not is a feature.
The figures are invented. A company at $5M in annual revenue serving independent insurance agencies ships AI document intake. Customers in a pilot report about six staff hours saved a week, and inference costs roughly $40 per account a month. Rather than meter documents, it adds the feature to its top tier with a monthly document cap and a $150 monthly uplift, sells it to new logos for two quarters, then offers it as an upgrade at renewal. It reports AI-tier revenue and inference cost on their own lines. After two quarters, 30% of new customers choose the top tier and upgraded accounts renew at a higher rate, which becomes the AI revenue story the board sees.
If AI is the product rather than a feature, and the customer buys it for a measurable output, outcome or usage pricing can be right from the first sale. Where inference cost is large relative to price, absorbing it may not be affordable; pass it through on a unit the customer can understand.
From the Golden Section mistakes list, each paired with the play that prevents it.
Picking a meter without knowing who controls the outcome is pricing in a vacuum, disconnected from what the customer values.
AI does not sell itself; the outcome it produces has to be shown in the customer's own numbers.
Outcome pricing lengthens sales cycles, and the price has to justify the longer cycle.
In the order we would run them. Each is on its own page, most with a free Excel template.
Finds where AI moves the return the customer already pays for.
Chooses among outcome, bundled, usage, hybrid and absorb, and sequences the rollout.
Places AI revenue and inference cost on their own lines so margin can be seen.
Puts AI attach rate, AI revenue and inference margin in front of the team weekly.
Sales & marketing plays AI pricing is a pricing decision first, built on the same value proposition and matrix as the rest of the product.
They pay when the AI moves a number they already track and budget against. Ask three customers to name the outcome and its value in their own terms, then offer it as a paid upgrade to new logos and watch attach rate and retention. Until you know who controls the outcome, put a rate-limited allowance in an existing tier, lift that tier's price and keep measuring.
It depends on who controls the outcome, because every customer is buying one. Per token or per usage unit fits when the result depends heavily on how the customer uses the AI. Per outcome fits when the result depends on your product and is clear and attributable. Until you know which you are in, keep it in the seat or tier with a rate limit and a price lift on that tier. Remember that usage can fall as customers get more efficient.
Building AI into the outcome a customer already pays for is the kind of structural change our equity is meant to fund, and we underwrite a pricing transition explicitly rather than assuming a clean step-up. A pricing change by itself needs no capital.
Growth equity →Reviewed by Dougal Cameron, CEO & Co-Founder on 2026-10-22. Golden Section observations are labeled separately from external benchmarks and illustrative arithmetic.