Per-seat pricing charges for each human user, so when AI agents do work that people used to do, the seat count and the revenue fall even while the customer gets more value. Hybrid pricing (a fixed base fee plus a metered charge for usage, credits or outcomes) ties revenue to the work instead of the headcount. The trade is real: you gain expansion and margin protection, and you give up some predictability, a clean NRR series and an easy migration.
This article is about financial planning and modeling. It is not accounting, legal or financial advice; confirm revenue recognition treatment with your auditor.
Founders mostly need a narrow answer: if we move from seats to hybrid, what happens to our numbers next quarter and next year? This guide answers that with a worked model. For the definitions of each pricing model, see our guide to SaaS pricing models; this page covers the transition economics.
Why seat-based pricing is under pressure
The mechanism is simple. A seat is a proxy for value: one more person using the product means one more unit of work done in it. That proxy held for twenty years because work and headcount moved together. AI agents break the link. If an agent resolves tickets, drafts pipeline emails or triages bugs, the customer can do more work with fewer licensed humans. Under pure per-seat pricing the vendor is paid less precisely when the product does more. That is "seat compression".
The forecasts that get quoted most need care. The one that holds up comes from IDC's FutureScape: Worldwide Agentic Artificial Intelligence 2026 Predictions (October 2025): "By 2028, pure seat-based pricing will be obsolete, with 70% of software vendors refactoring their pricing strategies around new value metrics." Note the wording: pure seat pricing, and refactoring, not abandoning. Several widely shared Gartner numbers on the same theme could not be traced to a primary source, so we do not repeat them here.
Survey data points the same way, toward hybrid rather than toward pure usage:
- Growth Unhinged, 2026 State of B2B Monetization (Kyle Poyar, more than 230 software companies, surveyed April to May 2026): "25% of respondents said they were hybrid 12 months ago. That number has jumped to 37%." Large companies above $150M ARR still hold on to per-seat pricing (29% adoption), and only 5% of respondents thought investors would prefer seat-based pricing.
- ICONIQ, State of AI: Bi-Annual Snapshot (January 2026, AI products, N=297, multi-select): subscription or platform fee 58%, consumption 35%, seat-based 23%, outcome-based 18% (up from 2% in Q2 2025). Most AI products already combine models.
There is counter-evidence worth taking seriously. Atlassian told shareholders in October 2025 that customers using AI code generation tools "expand their paid seats on Jira at a rate that's approx. 5% higher" than those who do not. Seat compression bites hardest where the work done in the product can be automated (support, SDR outreach, data entry), and least where AI creates more work for humans to coordinate.
What vendors actually changed (2024 to 2026)
Every move below is taken from the vendor's own pricing page or announcement. The pattern is consistent: almost nobody removed seats. They kept a seat or platform fee as the base and added a meter for agent work.
| Vendor | Date | What changed | Pricing shape |
|---|---|---|---|
| Zendesk | Aug 2024 | AI agents billed only for issues "resolved autonomously by AI" | Seats + per-resolution |
| Salesforce Agentforce | Oct 2024 to 2025 | Started at $2 per conversation; in May 2025 added Flex Credits at $0.10 per action, plus unmetered per-user licenses | Seats + credits, buyer's choice |
| Intercom Fin | Current | $0.99 per outcome; helpdesk seats $29 / $85 / $132 per seat per month (annual) | Seats + per-outcome |
| HubSpot Breeze agents | Apr 14, 2026 | Customer Agent moved from $1.00 per conversation to $0.50 per resolved conversation; Prospecting Agent to $1 per recommended lead | Seats + per-outcome |
| Microsoft 365 Copilot | Current | Per-user Copilot license; custom agents billed "on a metered basis" | Per-user + metered agents |
| Atlassian | Announced Sep 1, 2026, effective Dec 3, 2026 | Seat plans keep built-in allowances (for example 25, 70 or 150 Rovo credits per user per month on Jira Standard, Premium, Enterprise); extra usage at $0.01 per credit | Seats + metered allowance |
Two details matter for finance. First, Salesforce now sells three structures at once, which tells you the market has not settled on a unit. Second, Atlassian's allowance is sized per user, so seats still set the floor and usage sets the upside. That is the hybrid shape this guide models. For the pure models, see usage-based pricing, outcome-based pricing and credit-based pricing.
The hybrid revenue formula
A seat book has one revenue driver per account. A hybrid book has two, and only one of them is contracted:
Seat MRR = accounts × seats per account × price per seat
Hybrid MRR = (accounts × base fee) + (active accounts × usage per account × price per unit)
The base term behaves like classic recurring revenue. The usage term behaves like a volume business: it moves with customer activity, it has a distribution rather than a single value, and it can fall without anyone churning. Everything that changes in your P&L and metrics comes from that second term.
Worked example: 200 customers, 40% seat compression
All inputs below are illustrative. They are chosen to make the mechanics visible, not to represent any company. Every number was re-derived in Python.
Today (seat model): 200 customers × 25 seats × $40 = $200,000 MRR. Each account runs 1,000 AI agent tasks a month, which costs you $0.08 per task in inference (200 × 1,000 × $0.08 = $16,000). Other cost of revenue (hosting, support) runs $6 per seat (200 × 25 × $6 = $30,000). Gross profit is $154,000, a 77.0% gross margin.
The hybrid design: $400 base per account plus $0.60 per agent task. At today's 1,000 tasks, an average account pays $400 + $600 = $1,000, exactly what it pays on seats. Revenue-neutral on day zero, by construction.
Now run two scenarios over the next year. In both, customers cut seats by 40% (25 to 15) because agents absorb part of the work.
- Scenario A, agents take the work: tasks per account rise 50% (1,000 to 1,500).
- Scenario B, budget cuts: seats fall but task volume stays flat at 1,000.
| MRR | AI cost | Other COGS | Gross margin | NRR (no churn) | |
|---|---|---|---|---|---|
| Today | $200,000 | $16,000 | $30,000 | 77.0% | 100% |
| A: seat model | $120,000 | $24,000 | $18,000 | 65.0% | 60% |
| A: hybrid | $260,000 | $24,000 | $18,000 | 83.8% | 130% |
| B: seat model | $120,000 | $16,000 | $18,000 | 71.7% | 60% |
| B: hybrid | $200,000 | $16,000 | $18,000 | 83.0% | 100% |
The arithmetic for Scenario A on seats: 200 × 15 × $40 = $120,000 of revenue, against $24,000 of inference (200 × 1,500 × $0.08) and $18,000 of other COGS (200 × 15 × $6). Gross profit $78,000, margin 65.0%. On hybrid: 200 × ($400 + 1,500 × $0.60) = $260,000, same costs, gross profit $218,000, margin 83.8%.
Three things to read from the table. Seat compression hits the seat model twice: revenue falls 40% while AI cost rises, because the agents doing the work still consume inference. The hybrid book keeps or grows revenue in both cases. And the 130% NRR in Scenario A assumes the per-task price holds as volume grows. In practice large accounts negotiate volume tiers, so treat it as an upper bound.
Now the case hybrid boosters skip. Scenario C: no seat compression, but customers move 30% of their agent work to another tool (1,000 to 700 tasks). The seat model keeps $200,000 and its margin improves to 79.4%. The hybrid book drops to 200 × ($400 + 700 × $0.60) = $164,000, an 82% NRR with zero logo churn, and margin falls to 74.9% because the per-seat costs did not shrink. Hybrid moves your exposure from headcount risk to activity risk. It does not remove exposure.
What changes in revenue predictability
On seats, intra-year revenue is close to locked: seat counts change at renewal or at true-up, and annual prepayment makes the next twelve months mostly contracted. On hybrid, only the base is contracted. In the example, the base is $80,000 of a $200,000 book on day zero, or 40%. The other 60% depends on activity each month.
Three practical consequences:
- Committed share shrinks as you succeed. In Scenario A the base is still $80,000 but total MRR is $260,000, so only 31% is committed. Growth arrives in the uncommitted portion.
- Annual commitments come back as a hedge. Prepaid usage commits or credit packs restore predictability, at the cost of deferred revenue and breakage accounting.
- Seasonality appears. Usage follows your customers' business cycles. A December dip in agent tasks becomes a December dip in revenue, which a seat book never showed.
What changes in NRR
Net revenue retention is calculated the same way, but its components behave differently:
NRR = (starting MRR + expansion - contraction - churn) ÷ starting MRR
On seats, contraction is a discrete event: a customer removes ten seats at renewal. On hybrid, contraction and expansion happen every month on the usage line, often in the same account. A customer that runs 900 tasks one month and 1,100 the next generates $120 of contraction and then $120 of expansion with no real change in behavior. Measured monthly, gross expansion and gross contraction both inflate, and gross revenue retention looks worse than the business is.
Fixes that keep the metric honest:
- Compute NRR on a trailing three-month average of usage revenue per account, or annually on cohorts, not month over month.
- Report base NRR and usage NRR separately. Base NRR tells you about logo health; usage NRR tells you about adoption.
- Split usage contraction into "customer did less work" and "customer moved work elsewhere". Only the second is a competitive signal.
NRR is now measuring activity, not licenses. Cohort analysis by signup quarter is the cleaner view through a migration.
What changes in gross margin
Seat plans quietly cross-subsidize heavy users. Run the example per segment, keeping total task volume at 200,000 a month:
| Segment | Accounts | Tasks / account | Seat price | Seat-plan GM | Hybrid price | Hybrid GM |
|---|---|---|---|---|---|---|
| Heavy | 40 | 2,200 | $1,000 | 67.4% | $1,720 | 81.0% |
| Typical | 100 | 1,000 | $1,000 | 77.0% | $1,000 | 77.0% |
| Light | 60 | 200 | $1,000 | 83.4% | $520 | 68.1% |
On seats, every account pays $1,000, but a heavy account costs $176 in inference (2,200 × $0.08) against $16 for a light one. As agents take on more work, the heavy segment grows and blended margin erodes, which is exactly the slide we describe in AI COGS and gross margin. Usage pricing passes the variable cost through to whoever creates it.
The surprise is the light segment: its margin falls on hybrid, from 83.4% to 68.1%. Each light account still carries $150 of per-seat cost (25 × $6) but now pays only $520. The lesson for design: the base fee has to cover the fixed cost of serving an account, or low-usage customers become your worst-margin customers. Use the AI cost and margin calculator below to test your own per-task cost against your price, including the heavy-user band.
Revenue recognition for base plus usage
We cover prepaid credits, breakage and deferred revenue in detail in our credit-based pricing accounting guide, so here is only the part specific to a base-plus-overage hybrid.
PwC's "Revenue recognition: A Q&A guide for software and SaaS entities" works through exactly this structure under ASC 606-10-32-40, the exception that lets you allocate variable consideration to the period it relates to:
- Allowance resets monthly (PwC's Case A): the fixed fee is recognized over the term and each month's overage is allocated to that month. Revenue tracks usage.
- Allowance applies to the whole annual contract (Case B), and customers typically exceed it late in the year: recognizing overage only when incurred would load revenue into the final months without more effort from the vendor. PwC says the vendor "likely has to estimate the variable fee and recognize the total transaction price (fixed minimum and estimated variable fee) over the contract term", unless the contract is a promise of a specified quantity of service rather than access.
The planning point: the same customer behavior can produce different quarterly revenue depending on whether your allowance resets monthly or annually. Decide the allowance period with your accountant before you set it in the price book, because it changes the shape of your revenue line.
Forecasting a hybrid book
A seat forecast needs seats per account and a churn rate. A hybrid forecast needs a usage distribution. Build it as two drivers:
- Base revenue = active accounts × base fee. Forecast it like any subscription line, with new logos and logo churn.
- Usage revenue = active accounts × usage per account × price per unit. Forecast usage per account from cohort history (how usage ramps in months 1 to 12 after signup), not from a single average.
Then forecast a band, not a point. For Scenario A at month twelve, with an illustrative usage distribution across accounts:
| Case | Tasks / account | MRR | Base share |
|---|---|---|---|
| p10 (low) | 1,100 | $212,000 | 38% |
| p50 (median) | 1,500 | $260,000 | 31% |
| p90 (high) | 2,100 | $332,000 | 24% |
Plan fixed commitments (hiring, long leases) against the low case and variable spend against the median. Watch the cost side too: at p90, inference cost rises in step with revenue, so margin holds, but cash needs for prepaid compute can arrive before the customer pays. The revenue forecasting guide covers the driver-based method in general, and scenario planning covers how to turn these bands into trigger points.
Migration math: where the revenue dip comes from
The design above is revenue-neutral on average. Averages hide the problem. With the segment mix from the margin table, moving everyone to $400 + $0.60 per task means:
- 40 heavy accounts go from $1,000 to $1,720 (+72%).
- 100 typical accounts stay at $1,000.
- 60 light accounts go from $1,000 to $520 (-48%).
The light accounts' discount is certain and immediate: 60 × $480 = $28,800 of MRR leaves the day they migrate. The heavy accounts' increase is only collected if they stay. Here is what four common migration paths produce on day one after full migration:
| Migration path | MRR after | Change |
|---|---|---|
| Force-migrate, nobody churns | $200,000 | 0.0% |
| Force-migrate, 25% of heavy accounts churn | $182,800 | -8.6% |
| Force-migrate, cap increases at +20% for the first term | $179,200 | -10.4% |
| Let each customer choose seats or hybrid | $171,200 | -14.4% |
| Grandfather everyone, hybrid for new customers only | $200,000 | 0.0% (no dip, slow learning) |
The worst result comes from the option that sounds friendliest. When customers choose, light users switch (they save 48%) and heavy users stay on seats (they would pay 72% more). That is adverse selection: you take every decrease and none of the increases. The +20% cap costs $20,800 of MRR versus a neutral migration but sharply reduces churn risk among your highest-value accounts.
Timing matters too. If accounts migrate at renewal and renewals are spread evenly, the capped path falls in a straight line from $200,000 to $179,200 over twelve months ($194,800 at month three, $189,600 at month six, $184,400 at month nine), before any usage growth. Usage growth in the migrated accounts is what earns it back, so a migration plan should state the month in which the hybrid cohort's MRR is expected to cross its pre-migration level, and track it.
Ways to shrink the dip: raise the base so it covers per-account fixed costs (which also fixes the light-segment margin), include an allowance of tasks in the base so light users do not drop as far, and show customers a shadow bill for a cycle or two before switching, as our usage-based pricing guide recommends.
How a pricing change plugs into the operating model
A pricing migration is a forecast change before it is a billing change. In a driver-based model, the move from seats to hybrid replaces one revenue driver (seats × price) with two (accounts × base, and accounts × usage × price), and each of those feeds the rest of the P&L:
- Revenue: the MRR waterfall gains a usage line, and expansion and contraction get split between base and usage.
- COGS: inference becomes a driver of usage (tasks × cost per task) rather than a fixed percentage of revenue.
- Headcount: if hybrid makes revenue less certain, hiring plans should key off the low case.
- Cash: monthly usage billed in arrears collects later than annual seat prepayments, which shortens runway even when revenue is unchanged. Model collections timing explicitly.
Run the migration as scenarios: seats unchanged, compression with usage growth, compression without it, and usage loss. In Adlega that means building the revenue drivers once and comparing scenarios side by side, with the AI CFO tracing any number back to the driver that produced it. Whatever tool you use, the deliverable for your board is the same: the dip, the month it recovers, and what it looks like if usage disappoints.
Common mistakes
- Designing for the average account. Revenue-neutral in aggregate can still mean a 72% increase for your best customers. Model every account against last year's actual usage first.
- Offering a free choice of plans. It invites adverse selection. If you offer a choice, time-limit it or cap the decrease.
- A base fee below the cost to serve. Light users become negative-margin accounts.
- Reporting monthly NRR on a usage book. Noise reads as churn. Use trailing averages and split base from usage.
- Ignoring the downside scenario. Hybrid swaps seat risk for activity risk. If customers move agent work to a competitor, revenue falls with no churn event to warn you.
- Quoting untraceable forecasts in a board deck. Several "70% by 2026" statistics in circulation have no primary source. The IDC 2028 forecast does.
FAQ
Is per-seat pricing dying?
Pure per-seat pricing is losing share, but seats are not disappearing. IDC forecasts that pure seat-based pricing will be obsolete by 2028, with 70% of vendors refactoring around new value metrics, and Growth Unhinged's 2026 survey found hybrid adoption rose from 25% to 37% in a year. Most vendors that changed kept a seat or platform fee and added a meter.
What is seat compression?
Seat compression is the fall in licensed seats when AI agents do work that people used to do. The customer gets the same or more output with fewer human users, so a vendor that charges per seat earns less even as its product delivers more.
What is a hybrid pricing model in SaaS?
A hybrid model combines a fixed recurring fee (per account, per seat or per platform) with a variable charge for usage, credits or outcomes. The fixed part gives predictability; the variable part ties revenue to the work the product does. See our pricing models guide for how it compares with the alternatives.
How do AI agents change SaaS pricing?
They decouple work from headcount, so vendors meter the work instead: per action (Salesforce Flex Credits at $0.10 per action), per resolution (Intercom Fin at $0.99 per outcome, HubSpot at $0.50 per resolved conversation) or per credit (Atlassian at $0.01 per Rovo credit above the allowance).
Is hybrid revenue less predictable than seat revenue?
Yes, for the usage portion. Only the base is contracted, and the committed share shrinks as usage grows. Annual usage commitments, prepaid credits and forecasting with p10, p50 and p90 bands restore most of the predictability.
How do you migrate from per-seat to usage pricing without losing revenue?
Model every account at the new price using its real usage, size the base fee to cover cost to serve, cap increases for the first term, avoid an open choice between plans, and migrate at renewal. Expect a temporary dip from accounts that pay less, and plan the month you expect usage growth to recover it.
What happens to NRR when you switch to hybrid pricing?
NRR becomes more volatile because usage creates expansion and contraction every month. It can also rise well above what seats allowed, since accounts expand with usage rather than headcount. Measure it on trailing averages and report base and usage retention separately.
Should AI agents count as seats?
Some vendors price agents as seats, but an agent's cost to you scales with the work it does, not with its existence. If agents drive inference cost, a metered component protects your margin better than a flat agent seat.