You are choosing what you bill for: a seat, a token, a workflow task, or a successful outcome. That choice sets GTM and margin, so treat it as more than a Stripe setting. Bessemer’s AI pricing playbook (Feb 10, 2026) is the named public spine: copilots lean seat or consumption; agents lean workflow or outcome; hybrid (base fee plus usage or outcome tiers) is the early-stage middle ground when you are still learning cost variance.
What you are deciding
| Charge metric | Fits when | Breaks when | Bessemer signal |
|---|---|---|---|
| Seat (per user) | Copilot beside a human; value scales with headcount | Agent replaces work that does not grow with seats | Copilots typically priced per seat or consumption, like SaaS |
| Consumption (token / API / inference) | Technical buyer who wants control; costs track cleanly | Non-technical buyers cannot forecast; they throttle usage | Leena AI: consumption made customers wary; shift to outcomes gave clearer ROI |
| Workflow (per completed task) | Discrete, recognizable jobs (book meeting, draft contract) | Task complexity swings 10× and margin erodes | Clearer value than tokens; more cost variability |
| Outcome (per successful result) | Unambiguous, measurable win; you can absorb compute variance | Outcome definition is fuzzy or failure modes are expensive | Intercom Fin: $0.99 per ticket resolved, not per message or token |
| Hybrid (base + usage/outcome) | Early stage; need predictability and expansion upside | You never harden one model and custom deals proliferate | Bessemer: middle ground for early startups; example $12K + 100 included, then $5K / 100 |
fn-content has no verified atom yet for how common seat vs usage vs outcome is across AI SaaS cohorts. fn-content tracks it as benchmark request: AI SaaS seat vs usage vs outcome pricing adoption mix. Until then, use Bessemer’s named examples and principles, not an invented category share.
Predictability vs value alignment vs margin risk
| Model | Buyer predictability | Value alignment | Your margin risk | Who it fits |
|---|---|---|---|---|
| Consumption | Low for non-technical buyers (they must estimate tokens) | Weak (pay for activity, not result) | Low if you meter well | API / platform buyers |
| Seat | High (budget by headcount) | Medium for copilots; weak for agents | Medium (usage per seat can spike) | Human-in-the-loop copilots |
| Workflow | Medium (pay per task they recognize) | Stronger than seats or tokens | Medium–high | Bounded task complexity |
| Outcome | High on ROI story; variable on volume | Highest | Highest | Agents / services that close the loop |
| Hybrid | High floor + expandable ceiling | Strong if credits map to outcomes | Contained if platform fee covers 2× delivery cost | Early AI products still proving reliability |
Bessemer’s pattern: as you move from consumption → workflow → outcome, you accept more cost risk for tighter value alignment. Choose what customers will pay for, then build the discipline to make it profitable.
Map the product type before the price list
Bessemer describes three emerging models (pricing playbook and Part III, Dec 5, 2024):
| Model | What it is | Typical charge | Named public cue |
|---|---|---|---|
| Copilot | AI beside the human; person stays in the loop | Per seat or consumption | Microsoft Office 365 roughly $15–$30 per license; Copilot add-on about $30 more (Bessemer Part III) |
| Agent | Executes workflows with minimal human intervention | Workflow, outcome, or ROI vs incremental hire | Intercom Fin agent: $0.99 per AI resolution |
| AI-enabled service | Automation plus human oversight as a service | Consumption → outcome; often vs FTE or legacy service rate | EvenUp: per AI-generated demand package (not hourly paralegal) |
Part III also notes copilots at public companies (Microsoft, Google, Salesforce) have supported healthy price increases via add-ons. That is incumbent seat expansion. It does not justify seat pricing on an autonomous agent.
Named public examples (numbers only as Bessemer states them)
| Company | Model type (Bessemer) | Pricing mechanism (Bessemer) |
|---|---|---|
| DeepL | Hybrid | Per user + per editable file |
| EvenUp | Outcome-based | Per AI-generated demand package |
| Intercom (Fin) | Outcome-based | $0.99 per AI resolution |
| Leena AI | Outcome-based | ROI basis on tickets closed by agents; often a minimum threshold |
| Sett.ai | Hybrid | Per generative module + share of ad spend on winning campaigns |
| Zenskar | Hybrid | Annual subscription (tiered) with fees that scale by usage and complexity |
Use these as proof that hybrid and outcome models are live in market. Do not treat any row as your Series A pricing template.
Practical starter: hybrid until you can absorb outcome variance
Bessemer’s early-stage formula:
- Calculate true delivery cost (inference, human-in-the-loop, support drag). Include founder time if founders are still selling or answering tickets.
- Set a platform fee at about 2× that delivery cost so the floor covers COGS with room.
- Bundle a starter pack of outcome credits (their example: 100 resolutions inside a $12K annual platform).
- Price overages in blocks (their example: $5K per additional 100). As volume rises, price per outcome can fall while total revenue from the account rises.
Friction test they describe: start at a price. If buyers say “sold” instantly, raise it. Stop short of the point where price becomes a real blocker. Bessemer says multi-billion-dollar companies found their sweet spots this way over years five to ten, through buyer friction rather than a spreadsheet.
Stay hybrid while reliability is still climbing. Move toward pure outcome only when the success definition is unambiguous and you can absorb the long-tail compute cases (FounderNexus session: seats for humans, outcomes for agents when the product closes the loop on its own).
Soft ROI, hard ROI, and the 2026 renewal cliff
Bessemer maps products on revenue vs efficiency and hard vs soft ROI:
- Copilots often sit in softer ROI: advice and suggestions without closing the loop. Buyers ask whether they are getting value, and that question cuts willingness to pay at renewal.
- Agents that finish the job create harder ROI and stronger pricing power.
- Service replacement sells on total cost of ownership versus the legacy approach. Enterprises often undercount that legacy cost; your job is to make the comparison explicit.
Their warning: much of 2025 ran in “AI adoption at all costs” with low price sensitivity. As those pilots reach 2026 renewals, pricing must reflect delivered value, not promise. Soft-ROI copilots that never close the loop are the exposed class.
This page is operator judgment on charge metrics, not legal, tax, or securities advice.
Worked situations
Seed copilot, you copied Microsoft’s seat add-on math. Part III’s ~$30 Copilot add-on is an incumbent expansion story on Office seats. If your buyer is not buying more seats, seats will under-monetize. Prefer hybrid: platform fee covering 2× delivery cost plus credits for the jobs the copilot finishes.
Agent that resolves tickets, still billing tokens. Bessemer’s Leena AI lesson: consumption made customers wary of using the product. Intercom’s public shape is $0.99 per resolution. If you can define “resolved” cleanly and instrument it, move the charge metric to the outcome. Keep a platform floor until you know the variance.
Custom outcome deals for every logo. Bessemer’s complexity trap: nine pricing shapes across contracts break down at Series B. Pick one hybrid formula that works at 10 and at 1,000 customers. Push exceptions through a written approval path, not tribal AE creativity.
Soft-ROI pilot renewing in 2026 with weak usage proof. Reprice around a measurable outcome or a tighter workflow unit before the renewal. Or keep seats but attach expansion to hard adoption metrics. “AI potential” is not a renewal metric in Bessemer’s framing.
Sources
- Bessemer Atlas, The AI pricing and monetization playbook. Atlas Editors, published 10 Feb 2026. Three models (copilot / agent / AI-enabled service). Consumption vs workflow vs outcome trade-offs. Hybrid formula (platform fee at 2× delivery costs + outcome credits; $12K / 100 / $5K per 100 example). AI margins often 50–60% vs SaaS 80–90%. Intercom Fin $0.99 per AI resolution. Company examples table (DeepL, EvenUp, Intercom, Leena AI, Sett.ai, Zenskar, and others). Soft vs hard ROI; 2026 renewal cliff for soft-ROI pilots.
- Bessemer Atlas, Part III: Business model invention in the AI era. Feinstein, Rea, Bennett, Deeter, et al., published 5 Dec 2024. Copilot seat framing; Microsoft Office 365 roughly $15–$30 per license and Copilot add-on about $30 additional. Early vertical pricing examples including Fin at $0.99 per AI resolution.
- FounderNexus. Founder sessions, 2026. Seats for humans / outcomes for agents when the product closes the loop; hybrid while reliability climbs. Not a survey. No closed-session numbers.
- fn-content #18. Benchmark request: AI SaaS seat vs usage vs outcome pricing adoption mix.
Related
Founders who have picked a charge metric and a hybrid shape for an AI product will pressure-test yours in a FounderNexus session.