Most business software is priced by the seat. The model is so familiar it hardly seems a choice. It was, and it rests on an assumption that is about to stop being true.
Why the seat made sense
If a CRM's value comes from salespeople updating deals, more users means more value, and per-person pricing tracks it. It is simple and predictable. But the vendor is paid for access, not results, and the customer quietly absorbs the largest cost: the hours their people spend feeding the system.
What breaks when software does the work
An agent that sorts every incoming ticket creates value by the work it does, not the logins. If agents are good, a company may need fewer people in the system, and a seat-based vendor earns less for delivering more: a business model at war with its own product. Vendors respond with per-user AI surcharges or opaque credits. Neither is honest about what is being bought.
How other industries made the move
Early electricity was often sold by the lamp, a flat charge per bulb. Once reliable meters arrived, utilities charged for energy used, and customers could see what they paid for. Usage pricing works when the unit is understandable, the allowance covers normal use, and the customer controls the ceiling.
The case for keeping seats
Finance teams like predictable bills, and buyers have been burnt by usage pricing that looked cheap in month one and expensive by month six. That is why pure pay-per-action is also wrong. The answer is a hybrid: a familiar base price with a generous allowance, extra usage off by default and capped by the customer. Visible units and a well-sized allowance also blunt the temptation for vendors to make agents busy.
A vendor that earns less when its software works will, sooner or later, resist the software working.
A better unit
This is how piMonk prices. Every app has a familiar base price with a monthly allowance of agent actions included. Extra actions cost $0.01 each. They are off unless you switch them on, and they stop at the ceiling you set.1
What sits underneath matters too. By Kintoro's estimates, 1,000 agent actions cost about $1 on a small AI model trained for the job, about $2.30 on a mix of 80 per cent small and 20 per cent large general models, and about $13 on large general models alone.2 The trained small models are still being built and today's apps use a mix. But the choice of model is largely the choice of price.
Why this matters most in growing markets
Enterprise licences were priced for high-wage markets. Where salaries are a fraction of that, the same licence starts to compete with hiring the person. So companies buy fewer seats, share logins and keep critical work in spreadsheets. Pricing by work lets the whole team in without the bill multiplying. Pricing, not only technology, is part of how these economies can skip a stage, as we argue in The leapfrog economies.
What this means for a mid-market leader
What is the unit of value, and can I see it? If you cannot see how many actions you used, you cannot manage the cost.
Who controls the ceiling? You should, with extra usage off until you switch it on.
What does it cost to run, including my people's time? When Kintoro moved onto piMonk, our software run rate fell 83 per cent in three months once admin and maintenance time was counted.3
What happens to the price if the software works? If the vendor earns less, expect it to resist the outcome you are paying for.
The seat suited software people used. For software that does the work, it is the wrong unit.
From idea to practice
piMonk's insight on pricing agent work sets out the numbers: what an agent action is, what each app includes, and how ceilings work.
More perspectives
Sources
- piMonk list prices, US dollars, September 2026. See piMonk pricing.
- Kintoro estimates, September 2026. The trained small models are in development; today's apps use a mix of models.
- Kintoro internal figures, 2026. The 83 per cent reduction includes admin and maintenance staff time, not only licences.