#114 — Pricing for AI

Outcome-based pricing means customers only pay when the AI delivers a specific, agreed-upon business result—such as a resolved support ticket or revenue generated. This model perfectly aligns incentives between supplier and buyer, speeds up sales cycles, and can justify premium pricing. For example, Intercom’s Fin AI charges only for support tickets fully resolved by its autonomous AI agent, not when human intervention is needed. This clarity accelerates enterprise adoption and helps startups capture 25-50% of the attributable value delivered.

Never give away pilots for free—charging even a nominal fee filters out non-serious buyers. Structure pilots as business case development projects with the customer: co-create clear ROI benchmarks tied to their real KPIs. For instance, if you’re offering an AI coding assistant, agree up front on metrics like hours saved or fewer bugs, and use those to measure pilot success. This approach both validates the product’s value and creates internal champions who can later justify larger workflow investments.

The Autonomy x Attribution matrix classifies AI products by two axes: (1) autonomy (how independently the AI operates), and (2) attribution (how directly its impact can be measured). For example, Slack has low autonomy and low attribution, so seat-based pricing fits. Cursor (AI coding assist) is low autonomy, high attribution, making hybrid/usage+seat optimal. Twilio (communication APIs) is high autonomy, low attribution, so usage-based works best. The top right quadrant—high autonomy, high attribution—enables outcome-based pricing, the most lucrative and defensible model for startups.

Use the 'acceptable-expensive-prohibitively expensive' method: after demonstrating value, ask prospects what price feels acceptable, which feels expensive, and which is prohibitively expensive. Analyze the demand curve for psychological 'cliffs,' and consider anchoring your price at the 'expensive' level (not just 'acceptable'), as this often best aligns with perceived value, growth, and willingness to pay. Superhuman used this method to price at $30/month—against a crowded field of free alternatives—and validated customer demand for premium productivity.

GitHub Copilot started with seat-based pricing (low autonomy, low attribution). As competitors like Cursor demonstrated clearer time-saving value (high attribution), pricing shifted to hybrid models, combining seats with usage. Intercom’s Fin AI moved even further, charging outcome-based fees for each support ticket solved autonomously. These shifts mirror increases in product maturity and value-measurement clarity, which allow startups to capture more value as they scale.

Articulate your product’s impact in three buckets: (1) incremental top-line (e.g., more revenue, higher conversion, lower churn), (2) cost savings (e.g., reduced licensing or headcount), and (3) opportunity cost (e.g., freeing up team time for higher value work). Provide dashboards or reports that make these value stories easy to repeat—so the VP Sales, CFO, and CEO each see what matters most for their area. Always make your value pitch so clear that a customer can paraphrase it back; otherwise, it’s too complex for scalable adoption.

While pricing innovation (like outcome-based models) is a wedge for disrupting incumbents, true defensibility comes from network effects, proprietary data, integration into workflows, and brand. For example, Salesforce pioneered new pricing, but its lasting moat became its ecosystem and data. Founders should use pricing to win customers, but invest windfalls into building these less-replicable strategic moats.

According to the '20/80 axiom,' 20% of features create 80% of willingness to pay—yet founders commonly give this value away in free MVPs, hoping to monetize later. Instead, frame your offering as the 'most valuable product,' and charge for core value early. Free should be reserved for limited use, with clear upsell paths, or for land-and-expand strategies (where usage triggers monetization thresholds).

Unlike traditional SaaS (where 1:10 ROI is considered great), outcome-driven AI solutions have been able to capture 25-50% of attributable value. For instance, some AI chargeback recovery tools take a 25% cut of money recouped for clients. As long as value creation attribution is clear and outcome delivery is repeatable, startups can justify much higher take rates than previous generations of software companies.

Track customer cohort retention and 'workflow' usage, not just ARR headlines. Segregate experimental revenue (pilot/POC) from contracted, recurring workflow usage. Analyze customer engagement and monetization by segment (not just aggregate averages), and be cautious of overextrapolating peak revenue periods. Durable revenue comes from real adoption—renewals and expansion within customer accounts—not from a series of short-lived experiments or non-repeatable windfalls.

Outcome-based pricing isn’t suitable when value delivery is highly episodic (e.g., fraud detection that rarely triggers) or when ROI attribution is ambiguous/noisy. In these cases, recurring or fixed pricing may be more appropriate, as customers may balk at unpredictable, episodic charges or struggle to justify payment for infrequently realized outcomes. For example, B2C fitness apps commonly use gamification or subscription, not outcome-based pricing, due to attribution and consumer complexity.

Start with models that fit your product’s autonomy and attribution. As you gain clearer outcomes and higher autonomy, build features (e.g., dashboards, usage attribution tools) that justify a shift toward usage-based or outcome-based pricing. For example, coding assistants that move from being co-pilots to fully autonomous can migrate from seat- or usage-based pricing to outcome-based models tied to code deployed, bugs resolved, or features shipped—capturing more value as product and attribution mature.

The major AI pricing models are seat-based, usage-based, hybrid (seat + usage), and outcome-based. Seat-based pricing is common for tools needing human input (e.g., Slack, Figma). Usage-based is used for autonomous infrastructure services (e.g., AWS, Twilio). Hybrid models combine seats and usage for products offering measurable, user-driven value (e.g., Cursor). Outcome-based pricing charges only for AI-delivered results—for example, Intercom Fin AI charges per support ticket resolved without human aid.

Common mistakes include: giving pilots away for free, setting pilot prices as long-term anchors, choosing a pricing model that doesn’t fit product maturity, failing to co-create value metrics with customers, and undercharging for features that deliver the bulk of value. Many founders also conflate pilot revenue with sustainable, workflow revenue, creating misleading growth signals.

Frame your pilot as a business case development, not just technical validation. Define clear, mutually agreed ROI metrics (e.g., revenue increase, hours saved, cost reduced). Charge for your pilot to ensure buyer commitment. Use the pilot to empower your internal champion with data they can use to secure future budget. For example, set a 60-day pilot that aims to reduce support costs by 25%; success can justify ongoing, higher-value contracts.

Yes—innovative pricing (like outcome-based) can help startups win customers from slow-moving incumbents who are locked into traditional models. However, pricing alone is not a lasting moat; startups should leverage initial wins to build defensibility via data, integrations, workflow embedding, or brand.

Early willingness-to-pay (WTP) interviews clarify real market value before launch. Instead of asking 'What would you pay?', prompt users after a demo for prices they think are 'acceptable,' 'expensive,' and 'prohibitively expensive.' This reveals true demand cliffs. Superhuman, for instance, used this method to identify $30/month as their optimal price point despite free alternatives in the market.

Value stories should be categorized: (1) incremental top-line impact (sales leaders), (2) cost savings (CFOs), (3) opportunity cost (senior management). Use dashboards and regular reporting to make each ROI dimension visible and repeatable. Equip your champion to pitch for you internally by simplifying and rehearsing your value proposition.

Traditional SaaS often aims for clients to see a 10x return (pay $1 for $10 of value). AI solutions with high, provable, and autonomous impact can often command 25–50% of the value they directly create. For example, chargeback recovery tools routinely take a 25% cut of recovered funds, showing the potential for much higher value capture in AI.

Upgrade your product’s attribution (with dashboards, API hooks, reporting), and demonstrate that your AI delivers fully autonomous, measurable results. Start with usage or hybrid, and move to outcome-based once you reliably close the feedback loop. For instance, once your AI QA agent can autonomously resolve bugs with clear documentation, consider charging per bug resolved rather than per hour or per seat.

The 20/80 axiom states that 20% of your product drives 80% of willingness to pay—typically the features that are easiest to build. Instead of giving your MVP away for free and hoping to monetize later, identify and charge for this 'must-have' 20% from day one to set the right market anchors and create sustainable revenue.

Measure cohort usage, engagement, expansion, and renewal rates—not just contract value or headline ARR. Separate pilot/POC revenue from recurring workflow revenue. Identify segments with the highest retention and upsell rates, and avoid over-representing peak or seasonal months in projections. Durable revenue comes from sticky, repeatable business usage.

Outcome-based pricing is rarely used in consumer products because results are hard to verify and customers are less sophisticated buyers. Gamification (like refunding part of the subscription on fitness goal achievement) exists, but most B2C AI apps stick to subscriptions or freemium. Outcome-based works best for B2B when impact is measurable and the buyer has accountability for results.

At seed stage, focus on pricing model selection (seat, usage, or outcome), validating with real customer willingness to pay, and qualifying engaged customers with paid pilots. At growth, shift efforts toward optimizing for both market share and wallet share, moving up the autonomy and attribution spectrum to capture more value and defend your market position as you scale.

Seasonal products need true annualized reporting—don’t extrapolate a peak month across the year, as this creates misleading ARR. For example, tax automation AI may have huge Q1 sales but minimal use other quarters. Account for this in projections, and prefer usage or fixed pricing that matches customer demand patterns.

Move quickly to use pricing wins to build sticky integration (embedding in workflow), collect unique dataset feedback, create seamless user experiences, and drive network effects. For example, an AI assistant that plugs deeply into enterprise workflows and tailors recommendations using proprietary data will be harder to replace, regardless of pricing shifts by competitors.