From Tool Budget to Headcount Budget: How AI Agents Are Changing Software Pricing

In short: An AI feature gets paid for out of the software budget. An AI agent gets paid for out of the headcount budget. Price agents like features and you’re selling into the smaller pot. At the same time, every AI interaction creates compute costs the seat model never planned for. Most vendors are responding with hybrid pricing: a predictable base fee plus a variable component that grows with work, usage or outcomes.

Why did per-user pricing work for so long?

The seat wasn’t a mistake by the industry. For twenty years, it was a very good solution to a hard problem. It’s easy to understand: a buyer counts the people who work with the software, multiplies by the price and has a number they can budget. It’s predictable for both customer and vendor. And it grows with the customer without anyone needing a sales conversation.

Any new pricing model has to match those three qualities first. That’s why the shift is moving more slowly than many observers expect. But it is underway. According to the SaaS & AI Pricing Report 2027 by hy, OMR Reviews and Appinio, 80 percent of the 153 software and AI companies surveyed are planning or considering a pricing change in the next twelve months.

What’s the difference between an AI feature and an AI agent when it comes to pricing?

For pricing, I consider one distinction fundamental, worked out by the pricing consultants at hy. It concerns the budget a customer pays from and the benchmark they measure the price against.

OfferingWhat the customer buysBenchmarkBudget
AI featureEfficiency and qualityother softwareSoftware budget
AI agentwork donecost of human laborHeadcount budget
AI workflowend-to-end processcurrent process costs, such as service providersProcess budget

An assistant that turns a field technician’s voice memo into a report is a feature. An agent that schedules service calls on its own does work. The customer compares its price to salaries, day rates and tasks left undone. In most companies, the headcount budget is many times the size of the software budget. That is the real danger for vendors who stick with the seat, and the real opportunity for those who rebuild.

That leads to a practical rule. Even when features and agents live in the same product, they should be costed separately internally. They follow a different pricing logic, are measured against different benchmarks and drive different costs.

What pricing metrics are there for AI agents?

Four basic types have emerged in the market. They sit on a spectrum from “easy to measure but far from customer value” to “close to customer value but hard to attribute.”

ConsumptionBilling by compute, tokens or API calls. Protects the margin, but customers find it hard to understand and hard to predict.

TaskBilling per completed work step or executed workflow. Close to the work, but requires a clear definition of what counts as a task.

OutcomeBilling per result achieved, such as per resolved customer inquiry. Closest to customer value, with the most potential for conflict over what counts as a result.

Digital employeeA fixed monthly fee per agent, often with an allowance. Offers both sides the most predictability, but fits poorly with highly volatile volumes.

A metric has to pass three tests. It has to work for the customer, meaning it is fair, understandable and encourages usage. It has to work economically, meaning it delivers predictable revenue and scales with costs. And it has to work operationally, meaning it is clearly measurable and easy for sales to explain. One trap deserves special attention: per-task billing also rewards tasks that never needed doing. Good pricing models align the interests of vendor and customer.

Why are hybrid pricing models taking hold?

According to the pricing report, pure subscriptions are still the most common model at 40 percent, followed by hybrid models at 34 percent.

63%

of respondents expect hybrid models to be the most important for their company in two years.

The reason is predictability. B2B customers have annual budgets, procurement departments and controllers, and none of these institutions disappear because agents take over work. With pure consumption billing, users hesitate to use the AI at all so as not to waste credits. The consultants at hy call this the taxi meter effect. An agent nobody dares to use creates no value.

A hybrid model therefore combines a fixed base fee for what’s needed anyway, such as the system of record, with a variable component for the agents’ work. Proven mechanics include allowances with advance warnings before they run out and credits that translate very different services into a single, budgetable unit.

How is AI changing software company margins?

AI is bringing back variable costs that rise with usage, something the software industry hadn’t dealt with for a long time. Classic SaaS products scaled with almost no marginal costs. Every AI interaction, by contrast, creates real compute costs, and those vary widely with the difficulty of each case.

Even so, half of the vendors surveyed in the pricing report include AI features in the core product at no extra charge by default, and 22 percent don’t monetize their AI features at all. If you bundle AI for free into a package priced per user, you may be funding the very tools your customers use to cut users. That can be a deliberate choice to drive adoption. But it should be a choice, with a cap and a clear idea of when the investment will pay off.

Three tasks therefore belong on the executive team’s agenda and shouldn’t be left to engineering alone.

  1. Make costs visible: per customer, per feature, per agent. Without reliable measurement, there is no reliable billing.
  2. Contain costs: caps on steps and costs per case, handoff to a human when limits are exceeded, allowances in the contract.
  3. Choose models deliberately: Not every task needs the most expensive language model. Build agents so models can be swapped, and you benefit from falling prices and are less exposed to rising ones.

When does outcome-based pricing work?

Outcome-based pricing is the goal of many discussions, but the road there is longer than its advocates assume. The pricing report expects billing by task and workflow to catch on much faster than pure pay-for-performance. The latter only works where the outcome is clearly measurable and clearly attributable to the vendor.

With its support agent Fin, Intercom has shown that customers will pay for resolved inquiries, and at the same time that the conflict lies in what counts as resolved. Anyone who wants to bill for outcomes needs four safeguards. A minimum commitment protects the vendor, a cap protects the customer. Parallel billing covers services that can’t be measured as outcomes. And clear authority over definitions sets out who decides whether an outcome has been achieved. The most elegant outcome is one confirmed by a third party, such as equipment uptime reported by the machine itself.

What role does sales compensation play?

A pricing model your own sales team doesn’t want to sell stays on paper. As long as compensation is tied to newly sold licenses, licenses will get sold, even when the company needs something else. Compensation therefore has to shift to growth in a customer’s total recurring revenue, regardless of whether it comes from licenses, base fees, allowances or outcomes. That sounds simple, and in practice it requires new metrics, new billing systems and the sales team’s trust that they won’t come out worse under the new model.

What should executive teams review now?

  • Sort your AI offerings into features, agents and workflows, and check whether you cost them separately.
  • Ask to see the compute costs of your AI features per customer and the gross margin of your AI offerings, separate from the rest of the business.
  • Talk to your head of sales about what a compensation model that rewards customer revenue instead of licenses would look like.

Frequently asked questions

What is a hybrid pricing model in SaaS?

A hybrid pricing model combines a fixed base fee, which gives the customer predictability, with a variable component that grows with usage, completed tasks or outcomes. It’s considered the likely standard for AI in established software products.

Why is the seat model a problem for AI agents?

Because agents take over work that people with licenses used to do. The customer needs fewer licenses even though the software does more. At the same time, the agents generate compute costs that per-user pricing doesn’t reflect.

Should AI features cost extra?

Not necessarily. It can make sense to offer AI features at no extra charge to drive adoption. What matters is that the costs are known, a cap exists and agents that do work are billed according to their own pricing logic.

Further reading: Is SaaS dead? Why that’s the wrong question · Who owns the customer when the agent does the work? · Managing two curves at once: leading the transformation

Sources: SaaS & AI Pricing Report 2027 by hy, OMR Reviews and Appinio, summary at report at OMR Reviews · hy Consulting Group (2026): KI-Features bepreisen. Pricing für KI-Agenten im SaaS umsetzen (in German)

Similar Posts