Outcome-based pricing became the default pricing model for AI agents in 2025-2026 — and it's reshaping how enterprises evaluate, procure, and budget for AI deployments. Here's how it works, where it wins, where it breaks, and how to evaluate it honestly.
What outcome-based pricing means in practice
You pay per measurable outcome the AI produced. Concrete examples:
- Customer support: Pay per resolved conversation ($0.50-$3.50/resolution typical)
- AI SDRs: Pay per qualified meeting booked ($50-200/meeting typical)
- Healthcare patient outreach: Pay per completed call ($0.50-$2/call typical)
- Document processing: Pay per processed document ($0.20-$5/doc depending on complexity)
- Code generation: Per accepted PR (less common; Sweep does this partially)
Contrast with per-seat ($X/user/month) or per-usage ($Y per API call) models. Outcome-based ties vendor revenue to delivered value.
Why vendors moved to outcome-based
Three reasons drove the 2024-2026 shift:
Buyer alignment
Enterprise CX, GTM, and ops leaders pushed back on per-seat AI pricing through 2023-2024. The pitch was "AI replaces seats" — but the pricing assumed seat-replacement value without seat-replacement risk. Outcome-based realigns: vendor only wins if outcome ships.
Differentiation from per-seat incumbents
Sierra's outcome-based pricing was its sharpest differentiator vs Ada / Zendesk's per-seat pricing during 2024-2025 procurement wars. Buyers preferred the alignment; Sierra won deals partly because of it.
Validates the AI
If vendors price per outcome, they're betting their P&L on AI capability. That's a stronger trust signal than per-seat pricing, which guarantees revenue regardless of performance.
Where outcome-based wins for buyers
Pilot risk is bounded. You pay only for what the AI delivered. Failed pilots cost you implementation labor but not platform fees. Lower-risk path to validate.
Scale aligns naturally. Successful deployments grow vendor revenue and your value together. No "we used the platform less but still paid the same" frustration.
Vendor incentive aligns to your KPI. Vendor wants to deflect more (in CX) or book more meetings (in sales) — same as you.
Easier ROI math. "$2 per resolution vs $15 per human-handled ticket" is a one-line ROI calculation. Per-seat math is harder to defend.
Where outcome-based breaks for buyers
Forecasting is harder. Per-seat: $30K/month, predictable. Outcome-based: somewhere between $5K-150K/month depending on volume + deflection rates. Finance teams hate this variance.
Vendor gaming risk. Vendors optimized for "outcomes that maximize their billing" — not always identical to outcomes that maximize your value. Examples:
- AI SDR booking "qualified meetings" that don't convert
- Support agent resolving conversations by closing them rather than truly solving
- Document processing AI flagging documents as "processed" with minimal actual analysis
Counter this with strict outcome-definition contracts + quality SLAs.
At very high volume, the math inverts. Once you process 500K resolutions/month, the per-resolution premium on outcome-based can exceed what an all-you-can-eat per-seat tier would have cost. Negotiate volume tiers aggressively or revisit pricing model.
At very low volume, unit economics don't work. Outcome-based usually has minimums + implementation costs that need 2K+ monthly conversations to amortize. Small CX teams find Sierra's outcome-based model effectively too expensive at their scale.
The vendors and their outcome-based models
Sierra
Per-resolution pricing: $1-3 typical, $2-4 for tool-using workflows, $3-5 for identity-verified actions. Enterprise volume tiers at 50K, 100K, 500K resolutions/year.
Decagon
Similar per-resolution model to Sierra, often 15-30% cheaper at mid-market volumes. Voice pricing per-minute ($0.20-0.40/min typical).
Intercom Fin
Hybrid: bundled per-resolution within Intercom subscription. $0.99/resolution after included tier. Easier procurement (one Intercom contract) but resolution definition is platform-controlled.
11x (and other AI SDR vendors)
Per-booked-meeting pricing in some deployments ($50-200/meeting). Many AI SDR vendors still use per-seat as primary; per-meeting is offered selectively.
Forethought
Per-conversation + tier-based. Closer to hybrid; still substantially outcome-aligned.
Hippocratic AI
Per-completed-call + platform fee. Healthcare-specific outcome definitions (intake completed, follow-up completed, etc.).
How to evaluate outcome-based pricing
Step 1: Pin down the outcome definition
Make sure "resolution" or "qualified meeting" or "completed task" is defined contractually with examples + edge cases. This is where vendor incentive misalignment hides.
Step 2: Run the math at 3 volume scenarios
- Low (50% of forecast): does the platform still pencil?
- Expected: what's the per-seat-equivalent cost?
- High (200% of forecast): are there volume tiers that protect you from runaway costs?
Step 3: Negotiate volume tiers
The list rate is rarely what you pay at scale. Multi-year commits + volume tiers should reduce 15-40% off list.
Step 4: Lock quality SLAs
Outcome-based + low quality = vendor gets paid for bad outcomes. Lock CSAT, conversion-rate, accuracy SLAs to prevent gaming.
Step 5: Build in an off-ramp
12-month contracts max for outcome-based deals. Multi-year locks remove your leverage if the platform underdelivers.
The market direction
By Q4 2026, the dominant pricing models in AI agents shake out roughly:
- Outcome-based: Customer support (Sierra, Decagon), AI SDR (top tier), specific verticals (healthcare, legal automation)
- Per-seat: Coding agents (Cursor, Copilot, Devin), knowledge/search (Glean, Notion AI), most general-purpose
- Hybrid (per-seat + per-outcome): Increasingly common — base platform + per-outcome upside
The "all-outcome-based future" predicted in 2023-2024 didn't fully arrive. Per-seat is sticky for tools that augment work rather than replace outputs. Outcome-based dominates where the AI's job is identical to a quantifiable human outcome.
Bottom line
Outcome-based pricing reshaped AI agent procurement through 2024-2026 and remains the dominant model for customer support, AI SDRs, and several vertical applications. It's not strictly better than per-seat — it trades predictability for alignment. The right evaluation: pin down outcome definitions, run the math at multiple volume scenarios, negotiate volume tiers, lock quality SLAs, build in an off-ramp. Done well, outcome-based aligns vendor incentives with yours. Done badly, it's vendor-gaming with extra steps. The 12-month evaluation discipline applies either way.
Sierra review → · Sierra vs Decagon → · How to choose AI customer support platform →