The End of Per-Seat Software: Why AI Employees Are Priced by Outcome, Not by Login

For thirty years, business software was sold by the seat. You counted your users, multiplied by a monthly fee, and that was the bill. The logic held because one login meant one person doing a bounded amount of work. More people, more value, more revenue. Every SaaS company on earth was built on that single assumption.

AI agents break it in one move. An AI employee does not log in, does not hold a licence, and does not do a bounded amount of work. It completes whole workflows autonomously, and one of them can do the work of many people. Charging per seat for an agent is like charging per parking space for a self-driving fleet - the unit you are metering has nothing to do with the value being created.

The market has already noticed. Gartner estimates that roughly 234 billion US dollars of enterprise application spend, about 20 percent of the total, will be repriced away from seats toward consumption and outcome models by 2030¹. This guide is for the CTO, CFO, or Geschaeftsfuehrer trying to make sense of that shift: what outcome-based pricing actually is, why it only works under specific conditions, and how to evaluate it as a buyer without getting burned.

The Per-Seat Model Is Collapsing

The per-seat subscription was the most successful pricing idea in software history. It is also the one AI is dismantling fastest. The numbers from the last two pricing cycles are not a gentle drift - they are a structural break, and every SaaS board is now staring at it.

• Seats are shrinking as the primary model - per-seat pricing dropped from roughly 21 percent to 15 percent of surveyed SaaS companies in a single year, and pure per-user pricing as the primary model fell from 64 percent to 57 percent¹³.
• Hybrid has become the default - hybrid pricing (subscription plus usage or outcome) adoption reached about 61 percent, up 12 percentage points year over year, and hybrid companies report the highest median growth rates⁹.
• Usage-based components are now normal - roughly 61 percent of SaaS companies now have a usage-based element, up from 34 percent in 2021, and around 83 percent of AI-native SaaS companies offer usage-based pricing³.
• Analysts see a one-off repricing, not a blip - Gartner puts about 234 billion US dollars of enterprise application spend at risk of shifting away from seats by 2030, roughly 20 percent of application SaaS spend¹.
• The refactor is broad - IDC forecasts that 70 percent of software vendors will refactor pricing away from pure per-seat models by 2028¹⁵.
• Incumbents are already moving - GitHub moved Copilot premium requests to usage-based billing, Zendesk rolled out pricing that charges for resolutions its agents deliver rather than seats, and Workday introduced consumption credits so customers draw down AI capability without adding per-user licences¹⁵.

What makes this different from earlier pricing fashions is the cause. Usage-based pricing spread in the 2010s because cloud infrastructure was metered. This shift is different: it is happening because the thing doing the work stopped being a person at a desk.

Why Seats Break When Agents Do the Work

To see why the seat is the wrong unit, you have to look at what the seat was quietly measuring all along. A seat was never really about a chair or a login - it was a proxy for human labour. AI removes the human from that equation, and the proxy stops tracking anything.

What a seat was actually pricing

• A seat priced a person’s throughput - one licensed user could process only so many tickets, invoices, or leads in a day, so the seat count tracked capacity.
• A seat priced access, not results - you paid for the right to use the tool, and whether the work got done was your problem, not the vendor’s.
• A seat scaled with headcount - growing output meant hiring more people and buying more seats, so revenue rose with the customer’s payroll.
• A seat was easy to count - the whole model survived because you could audit it with a login report, not because it reflected value.

An AI employee violates every one of those assumptions at once. It processes far more than a person, it does not need a login to create value, its output has no fixed relationship to headcount, and there is nothing to count in a seat report because no seat is occupied.

This is not a marketing preference. It is the reason the whole category is being forced to move, and the analysts covering it are unusually direct about the mechanism.

The self-driving fleet analogy, made concrete

Imagine a taxi company that charged you for every parking space its cars could sit in. When human drivers each needed their own car, parking spaces roughly tracked the size of the fleet. Now the fleet is autonomous, cars run nearly around the clock, and one vehicle serves far more trips. Paying per parking space would be absurd - you would pay for idle capacity that has nothing to do with the rides delivered.

• The seat is the parking space - it counted a resource that used to correlate with output and no longer does.
• The completed trip is the outcome - the thing you actually wanted and the only sane thing to pay for.
• Utilisation decouples from headcount - an agent, like an autonomous car, can run continuously, so capacity is no longer bounded by how many people you employ.
• Value concentrates in completion - a trip that never finishes is worthless, and so is an outcome that never completes, which is exactly what outcome pricing captures.

What Outcome-Based Pricing Actually Is

Outcome-based pricing is easy to say and easy to get wrong, because it sits next to two neighbours that look similar. Getting the definitions straight is the difference between a contract that aligns everyone and one that produces a nasty surprise at renewal.

The four models, side by side

The defining feature of outcome pricing is where the risk sits. Under per-seat and usage models, the buyer pays regardless of whether the work succeeded. Under outcome pricing, the vendor only gets paid when the result is delivered, so the vendor absorbs the cost of failed attempts.

What the real examples look like

• Intercom Fin - charges about 0.99 US dollars per resolution, where a resolution means the customer’s issue was fully solved without a human. Fin can run on top of existing helpdesks with no seats required⁶.
• Fin for Sales - Intercom extended the model with a lead qualification priced at about 9.99 US dollars, a higher-value outcome than a support resolution, showing that different outcomes carry different prices⁷.
• Decagon - offers per-conversation (usage) and per-resolution (outcome) options, and runs pilots that track deflection and CSAT before a customer commits to outcome pricing⁴.
• Sierra - built its commercial model around outcomes its agents resolve rather than seats or conversations⁸.
• Zendesk - shifted to charging for resolutions its AI agents actually deliver instead of per agent seat¹⁵.

a16z has been calling this shift since late 2024, and its framing is the one most enterprise leaders now repeat: when AI can handle the work itself, the natural pricing metric becomes successful outcomes, because per-seat is no longer the atomic unit of software⁴.

Why hybrid dominates in practice

• Pure outcome pricing is volatile - a slow month can starve a vendor of revenue even when the product works, so few vendors go outcome-only.
• A platform floor covers the base cost - a modest fixed fee keeps the lights on and the integration maintained regardless of volume.
• The outcome component aligns the upside - the variable portion still ties the vendor’s revenue to work actually completed.
• Buyers get predictability plus fairness - a known floor plus a per-outcome rate is easier to budget than a pure consumption meter.
• It is now the market norm - hybrid structures are the most common model in SaaS and report the strongest growth⁹.

Outcome Pricing Only Works When the Agent Owns the Outcome

Here is the part most pricing debates skip. Outcome-based pricing is not a billing choice you can bolt onto any AI product. It is only possible when the system can actually complete an end-to-end outcome on its own. If it cannot finish the work, there is no outcome to charge for - and no amount of clever contract language fixes that.

This is where a rebranded chatbot and a real AI employee part ways. A chatbot drafts a reply and hands off. An AI employee reads the request, does the work across your real systems, and finishes the task. Only the second one can be honestly priced per outcome, and three things have to be true for it to work.

1. It needs a Company Brain

To complete an outcome reliably, an agent has to know how your company does the work - not in general, but specifically. That knowledge is your rules, your exceptions, your approvers, your definitions, and it mostly lives in people’s heads rather than in any document.

• A Company Brain is durable memory - a living record of how your company actually operates that persists instead of resetting on every task.
• It survives turnover - when the person who “just knows how we do this” leaves, the knowledge stays in the system rather than walking out the door.
• It raises completion rates - an agent grounded in your context finishes far more tasks without escalating, which is precisely what makes outcome pricing viable.
• Without it, pricing per outcome fails - a generic model re-learns your context on every task, completion rates stay low, and neither side can price on a result that rarely arrives.

2. It needs write access to your real systems

An outcome is not a suggestion. A resolved ticket, a posted invoice, or a booked meeting only counts when something changes in a system of record. That requires the agent to write, not just read.

• Read-only means draft-only - an agent that can only read produces a recommendation a human still has to execute, so no outcome completes.
• Write access closes the loop - the agent updates the CRM, posts to the ERP, sends the email, or files the record, and the outcome is real and verifiable.
• The connectors are where value lives - integration into email, Teams, SharePoint, CRM, and ERP is the hard part, and it is what separates a demo from a deployed AI employee.
• Verifiability comes for free - because the agent writes to a system of record, the outcome leaves an audit trail that both sides can bill and dispute against.

3. It needs to learn from feedback

• Completion rates have to climb - outcome pricing only pays the vendor when work finishes, so the system must get better at finishing over time.
• Corrections feed the Brain - every human fix, override, and preference becomes durable knowledge the agent applies next time.
• A static bot never improves - if the system cannot learn, its completion rate is capped on day one and the outcome economics never work.
• The loop compounds into a moat - the longer an AI employee runs in your company, the more of your specific outcomes it can complete, and the harder it is to replace.

This is the Superkind argument in one line: leverage, not headcount. The point of an AI employee is more output per person, priced by the work done rather than the seats occupied. That only holds when the agent genuinely owns the outcome, which is why the Company Brain, the write-capable integrations, and the feedback loop are not features - they are the preconditions that make outcome pricing honest.

The Buyer’s Guide to Outcome-Based AI Pricing

Outcome pricing sounds like a free lunch for buyers: only pay when it works. In practice, a badly defined outcome can cost more than a seat licence and hide worse problems. Here is how to evaluate an outcome-based offer without getting caught out.

Step 1: Define what an outcome actually is

1. Insist on a discrete, verifiable unit - an outcome must be a single thing that either happened or did not, provable from a system record, such as a resolved ticket or a posted invoice.
2. Reject activity dressed up as outcome - messages sent, engagement, and time saved are activity metrics that reward motion rather than results.
3. Name the system of record - agree exactly which system proves the outcome, so nobody argues later about whether a ticket was really resolved.
4. Price different outcomes differently - a qualified lead is worth more than a deflected FAQ, and the contract should reflect that, as Intercom’s 0.99 versus 9.99 US dollar split shows⁷.

Step 2: Normalise every quote to cost per completed outcome

1. Convert seats to outcomes - take the seat quote and divide by the outcomes those seats would produce, so you compare like with like.
2. Anchor against the human baseline - compare cost per outcome to the fully loaded cost of a person doing the same task, not to the sticker price of a licence.
3. Model your real volume - run the maths at your actual monthly volume, because outcome pricing can beat or lose to seats depending on throughput.
4. Include the integration and memory layer - check whether connectors and the Company Brain are in the price or billed separately, since that is where most cost hides.

Step 3: Set the guardrails

Step 4: Watch for the failure modes

Run those four steps and outcome pricing becomes what it should be: a way to buy completed work at a price you can defend, from a vendor whose incentives point the same direction as yours.

How Superkind Fits

Superkind builds custom AI employees that own end-to-end outcomes inside your business. The approach is process-first, not tool-first - we start from the work you want completed, then build an AI employee grounded in a Company Brain and connected to the systems you already run. That is what makes pricing by outcome honest rather than aspirational.

• End-to-end ownership - an AI employee takes a workflow from request to completed result, not from request to draft, so there is a real outcome to point at.
• Company Brain - durable company memory captures how your team actually works, so completion rates are high enough to price on and survive staff turnover.
• Write-capable integrations - the agent acts across email, Teams, SharePoint, CRM, and ERP, closing the loop instead of leaving the last mile to a person.
• Learning from feedback - every correction feeds the Company Brain, so the agent completes more of your specific outcomes each week.
• Priced by the work done - engagements are scoped around measurable outcomes with a clear definition agreed before the build starts, not around seat counts.
• Leverage, not headcount - the goal is more output per person, so your team runs a set of AI employees rather than growing to keep up.
• Process-first discovery - we map the real workflow with the people who do it before writing code, so the outcome we price is the outcome you actually need.
• Human-in-the-loop by design - high-stakes decisions route to a person, and audit logs record every action for review and dispute.
• EU-ready - data ownership, processing location, DSGVO, and EU AI Act transparency are handled as part of the build, not bolted on afterward.

Decision Framework: Which Pricing Model Fits Which Work

Outcome pricing is not the right answer for everything. The model should follow the shape of the work, and matching them is a straightforward exercise once you know what to look for.

For most of the repetitive, high-volume, rule-based work an AI employee is suited to, outcome or hybrid pricing wins because the work naturally produces countable results. The analysts covering the shift agree on the underlying mechanism driving it.

Frequently Asked Questions

Related Reading

• Agent Washing - how to tell a real AI employee from a rebranded chatbot before you buy, with the tests that expose a system that cannot complete an outcome.
• Why 400,000 Copilot Agents Still Do Not Know Your Company - why a Company Brain beats generic copilots, the memory layer that makes outcome completion possible.
• The Integration Tax - why AI value lives in the write-capable connectors, not the model, and who absorbs the cost of reaching your real systems.
• The Best AI Customer Support Agents in 2026 - a close look at Decagon, Sierra, and Intercom Fin, the vendors pioneering per-resolution outcome pricing.
• Hiring Freeze, Not Layoffs - the leverage-not-headcount playbook for growing output while headcount stays flat.

Henri Jung

Co-founder of Superkind, where he helps SMEs and enterprises deploy custom AI employees that actually fit how their teams work. Henri is passionate about closing the gap between what AI can do and the value it creates in real companies. He believes the Mittelstand has everything it needs to lead in AI - it just needs the right approach, priced by the work done rather than the seats occupied.