Service-as-Software: Why AI Is Ending Per-Seat SaaS Pricing

Get in touch:

Thinking through a new idea or stuck with a challenge? Drop us a message—we'll listen, brainstorm, and help move things forward.

Service-as-Software, Defined

Service-as-software is a product that completes a job, where traditional SaaS gives people a tool to do the job themselves. The term covers a support agent closing a ticket, a coding agent opening a pull request and a finance agent reconciling an account. What the customer purchases is the completed task.

Joanne Chen and Jaya Gupta of Foundation Capital brought the phrase broad attention in their 2024 essay, "AI leads a service as software paradigm shift". They describe "a $4.6 trillion opportunity", made up of $2.3 trillion in global salaries and $2.3 trillion in outsourced services. The figure expresses a venture-capital view of the addressable market, not a prediction of software companies' actual earnings. In our view, it works best as an indication of direction.

For product leaders, the important part is what this means for pricing. Once software performs the work, login counts no longer represent delivered value well. In the firm's July 2026 release, Gartner analyst George Brocklehurst stated the point plainly: agentic AI "breaks the link between user growth and revenue growth."

Why Is Per-Seat Pricing Under Pressure Now?

Agents can increase a customer's output even when headcount, and therefore seat count, remains unchanged or declines. That puts per-seat pricing under strain: a vendor charging by user may earn less precisely as its product becomes more useful. This mismatch drives the change.

The revenue exposure

Gartner's July 1, 2026 press release estimates that up to $234 billion of enterprise application software spend, about 20% of enterprise SaaS spend, could be exposed to "agentic arbitrage" by 2030. Because this forecast gives an upper limit, it sizes potential risk rather than recording an actual loss. In our reading, customers may use agents to achieve the same output, then buy fewer licences for the applications those agents bypass.

That risk has already moved markets. After Anthropic launched a legal plugin for Claude Cowork, a $285 billion rout hit software, financial services and asset-management shares in early February 2026, Bloomberg reported. On February 3, Goldman Sachs' US software basket dropped 6%, and traders took to calling it the "SaaSpocalypse". A brief sell-off reveals investor anxiety more than customer behaviour, yet it also shows the scrutiny now applied to seat-based revenue.

The cost side

Vendor costs create another source of pressure. Previously, serving a seat cost about the same whether someone used the product all day or signed in weekly. A long, multi-step agent task consumes much more inference than a brief answer does. Consequently, customers sharing one plan can now have sharply different costs to serve.

Margins show the effect. Drawing on 20 startups, Bessemer's State of AI 2025 found average gross margins of about 25% for the fastest-growing AI "Supernovas", versus about 60% for "Shooting Stars". ICONIQ's State of AI 2026 puts AI product gross margins at 45% in 2025 and projects 53% in 2026. Both are low for a software business. When a flat seat price has to absorb variable inference costs, the busiest customers become the least profitable. That often provides the first reason for teams to reconsider pricing.

How Fast Are SaaS Pricing Models Changing?

Change is happening steadily, mainly through addition. Instead of removing an existing subscription, most businesses are attaching usage or outcome elements to it.

ICONIQ's State of AI 2026, published in July 2026 and based on roughly 300 executives, found consumption-based pricing increasing from 35% to 42% between Q4 2025 and Q2 2026. Outcome-based pricing also rose, from 18% to 23%. Subscription was still the most widely used model, and companies combined an average of 1.7 pricing models. The clearest message is the combination itself: several pricing units within one business.

Other research reaches a similar conclusion. Kyle Poyar's 2026 State of B2B Monetization on Growth Unhinged, published May 13, 2026, covered more than 230 companies. It found hybrid pricing in use at 37%, up from 25% a year before. When asked about investor preferences, respondents chose hybrid (35%), outcome-based (26%) and usage-based (24%); only 5% selected seat-based. Those figures reflect respondents' perceptions of investors, rather than demand from buyers, and a usage-billing vendor sponsored the survey.

Further ahead, Gartner expects at least 40% of enterprise SaaS expenditure to move towards usage-, agent- or outcome-based pricing by 2030, according to Deloitte's TMT Predictions 2026. This measures something different from the earlier $234 billion figure. It tracks spending that changes pricing model, rather than spending at risk of replacement. Deloitte likewise anticipates a transition away from seat subscriptions and towards hybrid outcome and usage models.

Retention provides another argument. The 342 companies covered by Aleph's summary of the Benchmarkit 2026 benchmarks had median 2025 net revenue retention of 108% under usage-based models, against 98% under seat-based models. That gap is correlational, not causal: usage-based companies can differ in other respects, and their revenue can contract as quickly as it expands. Even so, when customers can expand spend without adding headcount, revenue tends to follow the work. Pricing by itself will not defend an AI product; our article about the advantages competitors can't copy when the models are identical explains what can.

What Are Vendors Actually Charging For?

Major vendors have not settled on a single model. Instead, customers can choose among stacked options such as per-user licences, credit bundles and per-outcome charges. The following prices were listed in October 2026.

Pricing sources are Salesforce, Microsoft (Copilot and Copilot Studio), Intercom Fin and HubSpot.

There are two clear patterns. Outcome pricing has advanced furthest in customer support because resolved tickets are straightforward to define and count. In August 2024, Zendesk introduced outcome-based pricing for its AI agents, billing only when the AI resolves an issue independently. HubSpot made a revealing change in April 2026: charges moved from each conversation to each resolved conversation, removing customer payments for failed attempts. Elsewhere, outcomes are less precise, so most vendors measure activity through actions, credits or conversations.

Is Per-Seat Pricing Really Dead?

No. Seats are becoming less dominant for AI features and agent work. For core software, however, buyer preference is shifting toward fixed, budgetable pricing, and seats are the most common way vendors provide it.

Buyer data offers the strongest evidence. Published on April 10, 2026 and based on 830 IT decision-makers, Futurum's 1H 2026 Enterprise Software Decision Maker Survey shows preferences diverging according to product type. For core software, preference for consumption pricing fell 5.8 points to 30.1%. Meanwhile, fixed or perpetual pricing increased from 18.7% to 22.5%, and outcome-based pricing rose from 18.4% to 21.7%. GenAI features showed the reverse movement: consumption preference rose 5.3 points to 42.9%, while outcome-based preference slipped from 30.9% to 26.8%.

The decline in GenAI outcome preference deserves attention. Futurum's release offers no explanation. One reasonable interpretation is that agreeing on an outcome proves more difficult than expected, whereas buyers can observe and control usage.

Vendor behaviour points to the same conclusion. During ServiceNow's Q1 2026 earnings call on April 22, 2026, CEO Bill McDermott said "50% of net new business now comes from a non-seat-based pricing model, including tokens and other assets." Although substantial, that transition leaves the other half of net new business with different models, including seats. Salesforce places $125 per-user Agentforce add-ons beside Flex Credits; Microsoft pairs $30 Copilot seats with Copilot Studio credit packs. None of these companies has dropped seats; they have made seats one line on a longer price sheet.

How Do You Choose a Pricing Unit for AI Features?

For every AI capability you offer, ask three things before selecting its pricing unit. Can customers attribute the result to your product? How widely does your cost to serve differ between customers? How much spending certainty does the buyer require? Different product areas usually produce different answers, explaining why hybrid pricing is becoming standard.

Question 1: How attributable is the outcome?

For outcome pricing to work, both parties must agree on the meaning of a completed result and on your product having produced it. A support ticket resolved without human handoff is unambiguous. A "qualified lead" or an "improved forecast" is harder because people and other tools also play a part. Use a usage price when the contract cannot define the outcome in one sentence.

Question 2: How much does cost-to-serve vary?

A flat platform fee or seat is simple and suitable when inference costs remain low and broadly consistent between customers. If agent workload varies greatly, however, flat pricing lets the busiest users eat into margins. Credits or metered usage assign that variation to those generating it. Given the Bessemer and ICONIQ margins above, in our experience this is the question teams most often underweight.

Question 3: How much predictability does the buyer need?

Futurum's data shows buyer preference shifting toward predictable pricing for everyday software. An uncapped usage bill worries finance teams and delays sales. Annual minimums, committed bundles of credits and spending caps retain the underlying usage model while giving buyers a firm number for budgeting.

In our view, most software companies will land on a platform fee covering access and core functions, seats for people who continue working in the product, and an outcome or metered charge for agent work. Seats remain, but no longer provide the main engine of revenue growth. Pricing belongs within a larger AI plan. Our AI strategy guide for software companies connects it to product and go-to-market choices. For teams weighing whether to purchase or create the capability, our build vs buy AI decision guide goes through that choice.

The Metering and Billing to Build First

Build the billing and metering foundations before revising your prices. Outcome and usage models depend on both parties trusting the measurements. Missing infrastructure, more often than strategy, is what holds up a pricing change.

• Measure individual events. For each chargeable event (an agent action, a completed run, a resolved ticket), capture the customer, feature, timestamp and associated inference cost. Unless cost data is connected, unprofitable features and customers remain hidden.
• Show customers their usage ledger. Present an ongoing record of consumption and expected cost, using the same units as the bill. Allowing customers to reconcile it with invoices prevents most disagreements.
• Put outcome definitions in writing. State precisely what qualifies and what does not (escalations, reopened tickets, duplicates) for every outcome-priced item, including the period allowed for a challenge. Retain supporting evidence for every result billed.
• Add alerts and spending limits. Administrators should be able to choose monthly caps and alert thresholds. This directly addresses the need for predictability and helps usage pricing pass procurement much more easily.
• Connect the billing systems. Metered data must flow into the CRM, invoicing and revenue recognition without manual spreadsheets. Decide early between developing the connection and purchasing a usage-billing platform.
• Test prices at plan level. Separate pricing configuration from application code. Rates, caps and credit bundles can then change for each plan without requiring a release.

The work is not unusual, but it spans product, data and finance simultaneously, and accuracy matters from the very first invoice. When defining the scope, use our explanation of the cost to build a SaaS application to understand how administrative and billing functions affect the budget. Our SaaS application development team also builds this form of metering into products. The measurements you need depend on agents' behaviour under genuine workloads, which our guide to agentic AI workflows in enterprise operations examines.

What Should Buyers Ask Before Signing?

Buyers need to understand the bill's calculation, how to verify it and whether it can be capped. New models redistribute risk, so establish where that risk lands.

• Precisely what gets billed? Obtain a written definition for each action, credit, conversation or outcome.
• Is live usage visible? At a minimum, there should be a live dashboard that matches the invoice.
• Can spending be capped? Find out whether reaching the limit stops service, throttles it or allows charges to continue.
• Who covers unsuccessful work? Under outcome pricing, an agent's failure should not create a charge. Confirm that the contract says so.
• What happens to rates upon renewal? Bundle sizes and credit prices may change. Request protection against rate changes throughout the term.

In our view, buyers should compare metered agent expenditure with the seat costs being displaced. A resolution charge can cost less than extra licences, whereas heavy use may make a flat seat better value. For example, 2,000 resolved conversations each month total $1,980 at Intercom's listed $0.99 per outcome, or $1,000 at HubSpot's $0.50 per resolution. Compare those totals with the fully loaded expense of replaced agent seats and staff hours, using your actual volumes.

The Bottom Line

AI agents separate delivered value from the number of people using a product, something per-seat prices cannot reflect alone. Evidence suggests a gradual split: seats and platform fees provide certainty, while outcome-based or metered charges cover agent work. For each capability, select a unit according to attribution, variation in cost-to-serve and the buyer's need for predictability. Then establish billing and metering that both parties can trust.

If you need to set prices for AI features or add billing and metering to your product, talk to us. For help with the planning, read our AI strategy guide for software companies.