Per-Seat Pricing in the AI Era

By Brian Tr · Updated Oct 5, 2026

Per-seat pricing charges a fixed fee for every person who uses your product. It's the default way B2B software has been sold for two decades, and it rests on one quiet assumption: the more people a customer has using your product, the more value they get. AI puts pressure on that assumption. If your software does the work instead of helping a person do it, the customer may need fewer humans, and a revenue model tied to humans starts to shrink as your value grows.

That doesn't mean seats are finished. It means the seat is now one pricing choice among several, and founders need to know when it still fits and when it quietly works against them.

This article covers why seats won, the tension AI creates, where seats still make sense, the common adaptations, what research says vendors are actually doing, and a decision table you can use. If you want the classic mechanics of seat counting and expansion, our guide to seat-based pricing covers them. Here the focus is what AI changes.

Why per-seat became the SaaS default

Seats caught on because they matched how software created value. Bain & Company put it plainly in October 2025: per-seat pricing historically worked well because most business software made employees more productive, and it was easy to understand, forecast, and budget around.

Three properties made it stick.

It's simple to explain. "$X per person per month" needs no calculator. A buyer can multiply by headcount and get a number in ten seconds.

It's predictable on both sides. Finance teams can budget a seat count. Vendors can forecast recurring revenue, which is the figure investors and boards watch most closely (our primer on ARR and MRR explains why).

It grows with the customer. When a customer hires, you earn more without a sales conversation. Bain describes this as a natural growth path for vendors as customers add employees.

And there was a hidden fourth property that mattered more than anyone noticed: serving one more seat cost almost nothing. A user's activity didn't create a meaningful bill for the vendor, so a flat per-person fee was nearly pure margin. That's the property AI breaks first.

The structural tension AI creates

AI attacks seat pricing from two directions at once. Bain frames them as value and cost.

Value moves away from the human

Many AI features and agents run in the background and finish tasks with little or no human involvement. Bain's example is customer service automation that may replace entry-level support roles altogether, which means a price-per-human meter loses its meaning. If a customer needs fewer humans to operate the software, headcount-based pricing stops tracking the value delivered. Even where the seat count holds steady, Bain notes AI can deliver far more value per seat, or value unrelated to seats entirely.

Kyle Poyar's Growth Unhinged survey of more than 240 software companies, run in April and May 2025, states the logic in one line: if the AI works as intended, customers likely need fewer people and more AI.

Here's the trap in plain numbers. Say a customer has ten support agents on your helpdesk product at a seat price. Your new AI resolves half the tickets without a human. The customer, reasonably, reduces to six seats. You delivered more value, and your revenue from that account fell. The better your product works, the less you earn.

Cost moves toward the vendor

The second force is cost. Bessemer's AI pricing playbook makes the contrast directly: in classic SaaS, serving one more customer costs virtually nothing, while every AI query incurs a non-trivial expense, covering compute and inference plus "humans in the loop" support. Bain makes the same point: inference, fine-tuning, and AI-specific R&D all carry real costs, and vendors have to monetize AI features in a way that more than covers usage.

Under a flat seat price, a heavy user and a light user pay the same. In classic SaaS that was fine because both cost you roughly nothing. With AI, the heavy user may cost you many times more than the light one. You can end up with your best-engaged customers being your least profitable ones.

Put the two forces together and you get the core problem. Revenue is pegged to people. Value is drifting toward the machine. And cost is rising with machine usage, which the seat price doesn't see.

Where per-seat still works

The honest answer is that seats still fit a lot of products. A few conditions make them a good choice.

The human is still the one doing the work. In a collaboration tool, a design product, a CRM where reps live all day, or a project workspace, each person getting value is a person using the product. AI features in these products assist the user rather than replace them. A copilot that helps a professional work faster is the clearest case. Bessemer describes copilots as AI sidekicks that sit beside human users, typically priced per seat or consumption, much like SaaS.

AI costs per user are small and bounded. If the AI features add a modest, predictable cost per person, folding them into a seat price is safe. You can price the average and accept some variance.

Buyers value budget predictability. Many finance teams would rather approve a known monthly number than a metered bill that might spike. Growth Unhinged's survey listed predictability as one reason hybrid models are popular: by staying close to a traditional paradigm, buyers can estimate their costs and feel in control of their spend. Bessemer likewise lists bundling AI inside seat-based products as one way to keep predictability while managing inference costs.

Procurement is built for seats. Bain points out that customers' procurement teams are used to buying software by headcount, not value, and that selling a usage or outcome model means changing long-held norms and shifting budget lines from labor to software. If your buyers are conservative, a familiar seat price removes a sales obstacle.

If your product meets most of these conditions, don't tear up your pricing just because the market is talking about it.

When seats start to hurt

Watch for these signs that your seat model is misaligned:

• Customers shrink seat counts after adopting your AI features.
• Your highest-usage accounts have your lowest margins.
• Buyers describe your product as replacing a role, not helping one.
• A seat price high enough to cover AI costs makes the product too expensive for light users, while a price low enough for them loses money on heavy users.
• Prospects ask, "What do we pay if only two people log in but the agent does the work?"

Any one of these is a flag. Two or three together mean it's time to adapt.

Five ways to adapt without abandoning seats

Most companies don't jump from seats to outcomes. They adjust in steps. These are the common patterns.

1. AI bundled into higher seat tiers

The simplest move: put AI features into existing plans and raise the seat price, or reserve them for a premium tier. It's easy to sell and easy to operate. The risk is that you're pricing AI as a flat per-person cost, so heavy usage still isn't covered, and it does nothing about seat shrinkage.

2. A per-seat AI add-on

Customers pay a seat price for the base product plus an extra fee per user who gets the AI features. This keeps the seat logic and lets you charge only those who use the AI. It works for copilots. It struggles when the AI's value isn't tied to the user who is switched on.

3. Seat plus usage allowance

Each seat includes a bundle of AI usage, such as a monthly allowance of actions or credits, and customers pay for any overage or buy more. This is the most common hybrid shape. It protects your margin against heavy users while keeping a predictable base. Our guide to credit-based pricing goes deep on how allowances and credits are designed.

4. Digital worker seats

Some vendors price the AI itself as if it were a team member: a fixed monthly fee per agent, or per "role" filled. This keeps the familiar seat vocabulary, which helps a buyer who is comparing the cost against a salary. The weak point is the unit. An agent's capacity varies enormously, so "one agent" can mean very different amounts of work. If you use this model, define what one digital worker covers, such as a volume of tasks per month, so the fee maps to something real.

5. Platform fee plus usage

Drop per-user charges as the main meter. Charge a platform fee for access, support, and advanced features, then meter the AI usage. Poyar describes this as a model that locks customers in and works well when the pricing metric is getting commoditized or doesn't reflect the full value of the product. Some products go further and include unlimited seats under the platform fee, which removes the incentive for customers to ration who gets access.

For the wider family of options and how to combine them, see hybrid pricing for AI products and AI agent pricing models. The broader question of charging by consumption or results is covered in usage vs outcome-based pricing for AI.

What the research shows vendors are doing

Four data points give a grounded picture, along with their limits.

Established vendors are mostly hybrid, not fully converted. Bain analyzed more than 30 SaaS vendors that were adding generative AI, excluding AI-native companies. Roughly 35 percent had increased per-seat prices by bundling AI into existing tiers. About 65 percent had layered an AI meter, by usage or feature access, on top of seats. None now exclusively monetized AI as a separate add-on, and none had fully shifted to usage or outcome pricing. The sample is small and covers incumbents, so it describes how established vendors move, not what new startups do.

Seats lost ground quickly in a broader survey. In Growth Unhinged's April to May 2025 survey, seat-based pricing as a primary model fell from 21 to 15 percent over the previous twelve months, flat-fee subscriptions fell from 29 to 22 percent, and hybrid rose from 27 to 41 percent. Respondents were self-selected readers of one newsletter, with a typical company at one to 20 million dollars in ARR, so treat the direction as informative and the exact numbers as approximate.

Outcome pricing is still rare. In the same survey, 5 percent said their main model was outcome-based, while 25 percent expected it to be by 2028. Poyar's CAMP checklist explains the gap: outcome pricing needs consistent outcomes, attribution, measurability, and predictability, and few products have all four.

AI-native companies lean away from seats. Bessemer's playbook says AI-native companies are by and large abandoning seat-based SaaS pricing for usage-, output-, and outcome-based models. That's a venture investor's observation across its portfolio, not a census, but it matches the logic above: products that do the work have less reason to count people.

One more finding from Bain deserves a founder's attention: the transition is hard. Moving off seats takes product telemetry, billing and finance systems, retooled sales compensation, and customers willing to change how they buy. Bain's illustration is a vendor pitching a $40,000 AI agent that could eventually replace an $80,000 role. In the short term the customer needs both the employee and the agent, which raises their cost by 50 percent for an undefined period. Unless the vendor bridges that gap, with delayed payments, flexible contracts, or performance guarantees, the sale can stall.

Key Facts: Per-Seat Pricing in the AI Era

• Among 30-plus SaaS vendors adding generative AI (AI-native vendors excluded), about 35 percent raised seat prices by bundling AI and about 65 percent layered an AI meter on top of seats; none had fully moved to usage or outcome pricing (Bain, October 2025).
• Seat-based pricing as a primary model fell from 21 percent to 15 percent in a year, while hybrid rose from 27 percent to 41 percent, in a survey of more than 240 software companies (Growth Unhinged, April to May 2025).
• The 2026 edition, a survey of more than 230 software companies, found 37 percent on hybrid pricing, up from 25 percent of the same respondents twelve months earlier. The two samples differ, so compare the direction, not the exact figures.
• In the same survey, 5 percent used outcome-based pricing as their main model and 25 percent expected to by 2028.
• Classic SaaS serves an extra customer at near-zero cost, while every AI query incurs a non-trivial expense (Bessemer).
• Seats fit best when the human does the work, AI cost per user is small, and buyers want a predictable bill.
• The seat trap: if AI cuts a customer's headcount, your seat revenue can fall even as your value rises.

A decision table for founders

Use this as a starting point, not a rule. Match your product's situation to a likely fit.

Two rules of thumb sit behind the table. First, change the meter only as fast as your measurement allows. Bain's advice is to build telemetry before leaving seats, because you can't price on what you can't see. Second, keep customers able to move at their own pace. Bain recommends flexibility between seat-based and usage-based spending so the buyer can manage the transition.

A practical way to move

If your signals say seats are misaligned, here's a sequence that avoids a big-bang repricing.

1. Instrument first. Track AI usage per account and per user, along with the cost of that usage. You need to know your heaviest users' cost before you can price for it.
2. Add an allowance, not a rewrite. Include a bundle of AI usage in each seat and meter above it. Existing customers see little change, and you stop subsidizing the heaviest users.
3. Test a platform fee on new customers. Run the new structure on new deals, where nobody has an old price to compare against.
4. Keep a seat or admin anchor. Even in a usage-led model, a small per-seat or per-workspace component is useful for forecasting, and buyers understand it. For the broader range of choices, our overview of SaaS pricing models is a good reference.
5. Revisit when the product changes. As your AI takes on more of the work, the right meter moves with it. Pricing is something you review each year, not a decision you make once.

If you're still deciding whether pricing should differ by the kind of AI product you build, how AI startups make money maps the archetypes, and what an AI-native startup is explains why products built around AI face this question from day one. SaaS companies adding AI to an existing product face the same choice from the other side, as covered in the AI pricing question for SaaS.

The takeaway

Seats priced people because people were where software created value. As AI shifts that value to software itself, the seat becomes a weaker proxy. But it's still a strong one wherever humans do the work, and buyers still like how predictable it is. The practical path for most founders is incremental: keep what works, add a usage allowance to protect margin, and move the main meter off headcount only when your own data shows the link between seats and value has broken.

Related Reading

• Seat-Based Pricing
• SaaS Pricing Models
• Usage vs Outcome-Based Pricing for AI
• AI Startup Business Models Explained
• What Is an AI-Native Startup?