A strange paradox is sweeping enterprise SaaS. The better an AI tool gets at doing actual work, the fewer user seats a client needs to buy. An automated support platform resolving 80% of customer tickets without human assistance makes for a brilliant product. However, it’s also a product that works against its own per-user revenue model.
The problem is live and impacting revenue today. Enterprise AI agents carry out tasks in customer service, HR, finance and operations that once needed multiple software licenses. Intercom’s Fin AI Agent prices its service per resolved conversation, leaving behind the per-user subscription model. Salesforce has moved to conversation-based and credit-based pricing for Agentforce. McKinsey research shows that 40% of AI-native companies now use activity or consumption metrics instead of flat per-seat subscriptions.
The classic per-seat model assumed human headcount was a solid indicator of software value. Bigger support teams meant buying more helpdesk licenses. Hiring extra finance staff added ERP seats, while recruiting salespeople pushed up CRM counts. Autonomous AI agents defy that standard. Single operators now oversee an array of agents, leaving individual software tools to handle workloads once shared across whole teams. The customer gets more value, pays for fewer users and the vendor faces a revenue problem it may have actively created by building a product that works.
A Pricing Model That Punishes Effectiveness
Simon-Kucher pinpoints the core conflict. AI agent workflows let one human oversee an entire suite of automated tasks, meaning that per-seat pricing no longer maps to actual software usage, delivered value or support overhead. AI adds unpredictable compute, inference, memory and orchestration expenses that standard flat-rate plans weren’t designed to absorb. A vendor can find itself simultaneously undercharging high-usage customers and failing to build margins that reflect actual product costs.
The models gaining traction in response would likely include usage-based billing, charging for agent runs, actions, processed documents or tokens. Workflow-based pricing charges per completed invoice, resolved ticket or screened candidate. Outcome-based pricing ties the charge to a measurable business result such as a recovered payment or a closed case. Each model solves a particular piece of the puzzle while bringing distinct downsides. Outcome pricing aligns cleanly with customer value, raising attribution questions when metrics get contested. Usage pricing tracks expenses, delivering unpredictable bills that finance teams routinely fight.
The fix might turn out to be a hybrid setup. Companies pay a flat platform fee to cover security, integrations, management and governance, paired with metered rates for agent tasks or successful outcomes. McKinsey’s data suggests the market is diversifying, with current examples at Intercom and Salesforce both pointing toward layered models rather than a clean replacement for the subscription seat.
We asked SaaS founders and B2B product operators tackling this transition to share effective tactics, rejected approaches and realistic pricing models for software that cuts customer headcounts.
More from Business
- X Just Open-Sourced Its Algorithm And Added A Shadowban Checker – Real Transparency Or PR Theatre?
- Skan Lands $63 Million To Map How Employees Work – Is That Workflow Optimisation Or Workplace Surveillance?
- Businesses Keep Investing In AI Customer Service, But Customers Still Want Humans
- Inside The $3.3 Billion Funding Spike Transforming African Tech Infrastructure
- How Businesses Can Stay Connected During A Heatwave
- Top 10 Platforms To Hire Software Developers
- The Four Clocks That Start When A UAE Invoice Goes Overdue
- Trump Media’s Losses May Be A Warning For Investors In Political Brands
Our Experts
- Heath Squier, Founder and Chief AI Officer, EVKII
- Michal Piszczek, Chief Technology Officer, Archdesk
- Dmitrii Pshenin, Founder, Ligeniqo
- Viktor Bulanek, Founder, Penetrify
- Cache Merrill, Founder, Zibtek
Heath Squier, Founder and Chief AI Officer, EVKII
“Seat-based SaaS breaks when the buyer is paying for work done, not for humans logged in. If an agent removes three coordinator seats, the customer will not keep paying three coordinator licences out of habit. They will ask what outcome the agent owns, what happens when it is wrong, and who is still on the hook.”
“I price and buy AI like an operator, not a vendor. The question is not “how much per seat.” It is “what unit of work is being replaced, what review cost remains, and what is the kill switch.” A cheaper agent that creates rework is a headcount cut that bounced back as management time.”
Michal Piszczek, Chief Technology Officer, Archdesk
“The closer agents get to doing the actual work, the dumber per-seat pricing looks. What bothers me more is how usage-based pricing breaks in a direction nobody wants to admit.”
“Per-seat punishes you for succeeding. Automate well and the customer needs fewer people, so they buy fewer seats. Usage-based flips it and pays you for failing. Every retry is billable, so are dead tool calls, and an agent that needs three attempts bills three times. The customer funds your error rate; it shows up on the invoice as ordinary consumption.”
“I came at this from infrastructure, measuring cost and energy per task. That number is meaningless until you divide it by the work that survives verification. Systems that look cheap on raw throughput stop looking cheap the moment you do. Pricing is the same arithmetic, with dollars instead of watts.”
“Which brings me to the only unit worth selling: a verified outcome. In construction those can be specific: an approved change order, a cost report that reconciles. What the debate keeps skipping is the hard part – a verifier both sides trust. Without one you’ve rebranded usage and marked it up.”
“Failed attempts belong on the vendor’s ledger. Push them onto the customer and you get proof debt, that pile of AI work which looks finished and isn’t, with the buyer paying to find out which is which.”
“In practice we run both, often blended, depending on customer size and what they are really buying. Our working rule: if an outcome can’t be verified, it isn’t an outcome. It’s a claim.”
Dmitrii Pshenin, Founder, Ligeniqo
“The pricing paradox is real: if an AI agent lets one employee do the work that previously required five seats, a pure per-seat model can punish the vendor for creating measurable value. I wouldn’t jump straight from seat-based pricing to unlimited usage, though.”
“A more sustainable model is often hybrid: a platform fee for access, usage or workflow-based pricing for the agent work, and an outcome or value component where the result can be measured reliably. The key is to charge for the scarce value the product creates, not for the number of humans left clicking inside it. That could mean completed workflows, resolved cases, processed transactions or verified outcomes.”
“Vendors also need pricing floors so efficiency gains don’t reduce revenue faster than expansion into new workflows can replace it. For buyers, transparent measurement matters. If the vendor can’t show what the agent completed, what it replaced and what human review remains necessary, outcome-based pricing becomes difficult to evaluate.”
“The best model will likely combine predictable base revenue with upside tied to measurable business impact.”
Viktor Bulanek, Founder, Penetrify
“We never had seat revenue to lose, which maybe makes me the wrong person to ask or the right one.”
“We charge per scan. Plans run from one test a month up to a hundred, and seats are bundled in. The reason was simple and it wasn’t strategy. Every scan is a real AI agent running up to four hours on our infrastructure, burning compute and model tokens. It costs us actual money each time. Normal seat SaaS has close to zero marginal cost so seats work fine there. Ours doesn’t, so usage was the only honest metre.”
“We looked at outcome pricing and dropped it. If I bill per vulnerability found, my incentive becomes finding more of them, and you get noise. And a clean scan, which is what the customer is actually paying for, would bill zero. That maths never worked.”
“Seat contraction isn’t what worries me. Inference prices keep dropping, so the cost of a scan drops, and buyers notice. That deflation is the harder problem.”
“The thing nobody admits about usage pricing: revenue is lumpy and the customer is motivated to use less. Seat SaaS cashflow is predictable. This isn’t.”
Cache Merrill, Founder, Zibtek
“The per-seat model is going to get harder to defend as AI takes on more of the work. We’ve traditionally thought about software in terms of how many people are using it, but that’s changing when one AI agent can do work that used to involve several people. If a customer is getting more value from the product while needing fewer seats, charging them more simply because the old pricing model says so doesn’t make much sense.”
“That doesn’t mean every SaaS company needs to immediately move to usage-based pricing. Usage can work for some products, but it can also make costs harder for customers to predict. I’d rather see companies start by understanding what customers are actually getting from the product and then build the pricing around that value.”
“Sustainable pricing will increasingly be tied to the work the software enables, not simply the number of people who have access to it.”
