Run the numbers on a hypothetical support team. Forty agents, $115 per helpdesk seat, $4,600 a month in licenses. Now an AI agent steps in, handles 61% of 9,300 monthly tickets, and charges $0.99 per resolution. That's 5,673 resolutions. The bill: $5,616.27. The vendor now earns more than it did with the whole team sitting in the chairs. And the customer, who just gained the equivalent of about 14 agents (assuming each closes around 400 tickets a month), still thinks it's a good deal. The numbers are illustrative, but the format is the one real contracts use. So who is charging wrong? Maybe no one. Maybe the problem is the word "resolution." ### <br>Who bet first Intercom was among the first to put a price on outcomes: Fin, its support agent, launched at $0.99 per resolved conversation, a figure the company has repeated since 2023. Salesforce introduced Agentforce in October 2024 at $2 per conversation and, in 2025, was already talking about flexible credits, around $0.10 per action executed. Zendesk and Sierra, the startup led by Bret Taylor, also argue for charging only for what the agent actually delivers. These are announcement prices. Tables like this change all the time, so check before putting them in a proposal. Notice Salesforce's move: it left the conversation and went to the action. It left the outcome and quietly went back to consumption. That shift says more than any of the prices. ### <br>Resolved according to whom? Ask three people what a resolved ticket is and you get four answers. The customer stopped writing after 72 hours: resolved? The customer clicked "this helped": resolved, but what if they came back two days later through another channel with the same problem? For the seller, every question turns into money once it's counted as resolved. For the buyer, every question counted that way raises the question of whether the customer was helped or just gave up in silence. Anyone who has watched an annoyed customer vanish and cancel the following month knows the difference. This conflict isn't settled by good faith. It's settled by a contract annex, an auditable log, and someone in finance willing to read both. On the vendor's side, the problem looks different. A software seat costs almost nothing extra per added user. A resolution costs tokens, and a refund ticket with five back-and-forths can consume several times what a question about opening hours does. The price is the same $0.99. Margin stops being a constant and becomes a statistical lottery. ### <br>Finance wants a ceiling Variable bills are a headache. If a product launch doubles ticket volume, the invoice doubles with it, in the very month the budget was already tight. That's why the most common design is a slightly dull hybrid: a flat platform fee, an allowance of resolutions or credits, per-unit overage and a monthly cap. Nobody posts that on LinkedIn with enthusiasm. It works. ### <br>Which unit survives My read is that seat pricing will shrink, but pure outcome pricing isn't the destination either. The unit that wins is the one buyer and vendor can count without arguing: an action executed, a task completed with a record, a credit consumed. These measures are boring and auditable, and that's exactly why they last. "Resolution" sells better on a pricing page and loses at signing time. There's also an irony almost no one mentions. If the vendor charges per resolution and the agent is good, the customer learns to prevent the problem before it becomes a ticket: improves the FAQ, fixes the billing flow, warns before the outage. Fewer tickets, smaller bill. Charging by outcome puts the vendor, in part, against its own prevention, and that crooked incentive only shows up in the second or third year of the contract. The strongest objection comes from the pure-outcome camp: the customer doesn't want to buy technology, they want the problem solved, and any intermediate unit is an excuse for the vendor to charge for effort, not value. It's a good argument. But it depends on a neutral way to measure value existing, and that doesn't exist yet.