--- title: "Outcome-Based Pricing for AI: How to Build Profitable AI Products" description: "Intercom charges $0.99 per resolution. Zendesk moved from $115/seat to outcomes. Here's how the smartest AI companies are pricing for value, not tokens." date: 2026-02-02 author: Kat Laszlo canonical: https://tansohq.com/blog/outcome-based-pricing --- # Outcome-Based Pricing for AI: How to Build Profitable AI Products By [Kat Laszlo](https://www.linkedin.com/in/katrinalaszlo/) · February 2, 2026 Here's a pattern I've seen a few times now: an AI founder looks at their data and realizes their power users—the ones who love the product—are actually losing them money. One founder mentioned spending $2.50 in compute for every $1 those users pay. Another found that 78% of high-usage accounts were underwater. It's a weird problem. The more someone uses your product, the worse it is for your business. Outcome-based pricing is one way out. Instead of billing for tokens or API calls, you bill for what the customer actually wanted: a ticket resolved, a lead generated, a question answered. Intercom charges $0.99 when Fin resolves something without a human. They don't charge for the AI—just for the result. Zendesk is moving this direction too. Last August they announced they're shifting away from $115/agent/month toward outcomes. Sierra's pitch is literally "pay only when we complete a task." This is happening now, not someday. --- ## The Problem with Charging for Tokens Usage-based pricing seemed like it would work. Charge per API call, per token, per minute. Simple enough. Maps to your costs. Except AI costs are deflating 10x per year. Lock in pricing based on today's costs and in 12 months you're either overcharging or scrambling to cut prices while competitors undercut you. And usage doesn't equal value. A customer sending 100,000 tokens on the wrong task gets less value than one sending 10,000 on the right one. You end up charging your best customers the most. ### The Usage-Based Trap Source: BCG AI Monetization Research, a16z Enterprise Newsletter Zendesk figured this out the hard way. Their traditional model charged $115/month per agent seat. Sounds fine until you realize: the better their AI gets at handling tickets, the fewer human agents customers need. Their own success was cannibalizing their revenue. So they're switching to outcome-based. Charge for resolved tickets instead of headcount. Now when the AI handles more volume, Zendesk makes more money. Incentives finally point in the same direction. --- ## What Outcome-Based Actually Means Instead of billing for what goes into your AI (tokens, calls, compute), you bill for what comes out. Leads generated. Tickets resolved. Documents processed. Fraud prevented. ### Pricing Model Comparison | Dimension | Usage-Based | Value-Based | Outcome-Based | | --- | --- | --- | --- | | What you charge for | Resources consumed | Perceived value tier | Measurable results | | Example metric | Tokens, API calls | Feature tier, seats | Resolutions, leads, verified users | | Margin protection | Low (cost pass-through) | Medium | High (decoupled from costs) | | Customer alignment | Low (penalizes usage) | Medium | High (shared success) | | Implementation complexity | Low | Medium | Medium-High | --- ## Why This Model Works GPT-4 class models have dropped 90% in cost. If you're on usage-based pricing, that's pressure to cut prices. If you're on outcome-based, that's pure margin expansion. The value of a resolved ticket doesn't change just because it got cheaper to deliver. Sierra puts it plainly: "Paid only when we complete a task for you." That's not just marketing copy. When your revenue is tied to customer outcomes, suddenly improving the product isn't a cost center. Better AI means higher resolution rates means more revenue. The incentives finally line up. Try explaining token pricing to a buyer. Now try this: "Each resolution costs $0.99. Your average ticket costs $15-20 to handle with humans. Do you want to save $14-19 per ticket?" That's not a 5-week procurement cycle. That's a yes. When competitors charge for usage and you charge for outcomes, you're saying something about confidence in your product. Decagon offers both per-conversation and per-resolution options, letting customers pick their risk tolerance. That flexibility becomes a selling point. Riskified only charges for approved, fraud-free transactions. No fraud detected? No charge. That's a bold promise, and it commands premium pricing. Customers aren't paying for "AI fraud detection." They're paying for guaranteed clean transactions. When customers only pay for results, what's left to object to? "What if it doesn't work?" isn't an objection anymore; it's a non-issue. Companies report 40-60% compression in sales cycles after switching to outcome-based models. You'd think tying revenue to outcomes would add volatility. But customers have pretty stable outcome needs. A support team resolves roughly the same number of tickets each month. A sales team needs a consistent pipeline of leads. The outcomes are more predictable than raw usage patterns. --- ## Who's Actually Doing This Only charged when AI successfully resolves a customer issue without human escalation. Moving from $115/seat/month to charging for resolved tickets. Announced August 2024. "Paid only when we complete a task for you." Zero risk for customers trying the platform. Charges only for approved, fraud-free transactions. Takes on chargeback liability. No upfront cost. Fees charged only when chargebacks are successfully recovered. Charges only for successfully verified users. Failed verifications don't cost customers. Revenue cycle management priced on claims processed and revenue recovered. Credit-based system for completed AI agent tasks. Flex Credits at $0.10/action for workflows. Offers both per-conversation and per-resolution options. Customers choose risk profile. --- ## This Isn't New Outcome-based pricing has been working at massive scale for decades: Airlines pay per flight hour, not for engines. Rolls-Royce maintains ownership and incentive to maximize uptime. Running since 1962, now 50%+ of their service revenue. Rail operators pay per passenger-mile. Hitachi owns trains and handles maintenance. Alignment: Hitachi profits when trains run reliably with high ridership. Same insight: when the supplier profits from the customer's success, the incentives line up. Rolls-Royce makes more money when planes fly more hours. Hitachi makes more money when trains run reliably. --- ## Making It Work in Practice ### Pick the right metric This is where most companies get stuck. You need something measurable, attributable to your product, and tied to real business value. ### Value Metrics by AI Product Type | AI Product Type | Good Metrics | Bad Metrics | | --- | --- | --- | | Customer Support AI | Tickets resolved, CSAT maintained | Messages sent, API calls | | Sales AI | Qualified leads, meetings booked | Emails drafted, contacts enriched | | Document AI | Documents processed, fields extracted | Pages scanned, API requests | | Fraud Prevention | Fraud blocked, clean transactions approved | Transactions scanned | | Code AI | PRs merged, bugs fixed | Completions generated, tokens used | | Content AI | Published pieces, engagement metrics | Words generated | ### Build in volume tiers Outcome-based doesn't mean flat pricing. Different customers have different volumes and different willingness to pay. Tiering lets you capture value across segments: ### Example Tier Structure (Customer Support AI) - • Up to 500 resolutions/mo - • Standard response time - • Email support - • Up to 5,000 resolutions/mo - • Priority routing - • Dedicated support - • Unlimited resolutions - • Custom SLAs - • Dedicated CSM ### Add guardrails Don't confuse outcome-based with unlimited. You still need protections: - →**Minimum commits:** Base subscription that includes X outcomes, overages priced per-outcome - →**Quality thresholds:** Define what counts as a "successful" outcome (e.g., resolution without escalation within 24h) - →**Complexity tiers:** Simple vs complex outcomes may have different pricing - →**Monthly caps:** Optional ceiling to give customers budget predictability ### Offer hybrid options Some customers want pure outcome-based. Others want predictability. Let them choose. Decagon handles this well: customers pick between per-conversation (safer, higher price) or per-resolution (riskier, lower price). Same product. Different risk profiles. Customers who are confident in the AI's performance can opt for outcome-based and pay less per unit. Skeptical customers pay more but take less risk. --- ## How Long This Actually Takes This isn't a two-week project. You need time to figure out the right metric, test it with real customers, and build the measurement infrastructure. ### Outcome-Based Pricing Transition Timeline - • Identify candidate outcome metrics - • Analyze current customer cost-to-serve - • Model revenue impact scenarios - • Interview customers on willingness to pay - • Define outcome measurement methodology - • Design tier structure and guardrails - • Build pricing calculator/simulator - • Create migration path for existing customers - • Pilot with 5-10 friendly customers - • Shadow bill existing customers - • Refine outcome definitions based on edge cases - • Train sales and success teams - • Launch for new customers - • Grandfather existing customers with opt-in path - • Monitor margin impact closely - • Iterate pricing based on data --- ## The Objections You'll Hear (And How to Think About Them) This confuses cost-based pricing with value-based pricing. You're not trying to maintain a fixed margin on every transaction. You're pricing based on what the outcome is worth to customers. If a resolution is worth $15 to them, it's worth $15 whether it costs you $0.50 or $2.00. Your margin varies; their value doesn't. Try explaining token pricing to a non-technical buyer. Now try: "You pay $0.99 every time we resolve a ticket without a human." Which is easier? Outcomes map directly to business results. Tokens don't map to anything customers care about. This is a feature, not a bug. The value of a resolved ticket doesn't decrease when your costs do. With usage-based pricing, you're forced to pass through savings. With outcome-based, you capture them as margin. That's the whole point. If you can't measure it, you can't price it. This is a real constraint, not an excuse. Build measurement into your product: track resolutions without escalation, integrate with CRMs for lead attribution, count successful document extractions. The instrumentation work pays for itself. --- ## Where This Is Going The companies moving to outcome-based pricing aren't doing it because it's trendy. They're doing it because usage-based is broken for AI. It punishes your best customers, exposes your margins to cost volatility, and makes value invisible to buyers. Intercom, Zendesk, Sierra, Riskified: these aren't experiments. They're signals of where the market is heading. The transition isn't trivial. Expect 11-20 weeks from analysis to migration. You'll need to instrument outcomes, test pricing with real customers, and train your sales team on a new conversation. But the result is a business model where your revenue grows when customers succeed, your margins improve as AI gets cheaper, and your pricing actually makes sense. ### Key Statistics at a Glance ## Key Takeaways - →Usage-based AI pricing is broken. Your best customers become your least profitable. - →Outcome-based pricing charges for results: tickets resolved, leads generated, fraud prevented. Not tokens. - →Intercom charges $0.99/resolution. Zendesk is moving away from $115/seat. Sierra only bills when tasks complete. - →The switch takes 11-20 weeks. Worth it: margins improve as AI costs drop, and sales cycles shrink 40-60%. *[Kat Laszlo](https://www.linkedin.com/in/katrinalaszlo/) is co-founder of Tanso, flexible pricing infrastructure for SaaS and AI.* ### Want to price with confidence? Book a quick call and we'll walk you through it.