The Resolution Trap.

Intercom Fin charges $0.99 per resolution. Make the AI better at resolving tickets and your bill goes up. Make it worse and your bill goes down. The pricing model is at war with the product it sells.

The maths

Intercom Advanced is $85 per seat per month as of May 2026. Their AI agent, Fin, is $0.99 per resolution on top.

A 10-agent team doing 1,000 Fin-resolved tickets a month: $10,200 a year in seats, plus $11,880 a year in Fin fees. About $22,080 a year. Cove costs £99 a month flat. Roughly $1,500 a year for the same team doing the same work.

But the headline number is not the interesting part. The interesting part is the slope.

Success is the bill

Per-seat pricing has one flaw: it penalises automation. Per-resolution pricing has a worse one. It taxes the outcome the vendor is asking you to buy.

You write a better help article. Fin resolves more tickets from it. Bill goes up. You train your agents to escalate less often. Fin closes more cases unattended. Bill goes up. You ship a fix that removes a class of tickets, so the remaining tickets are mostly the easy ones Fin handles fine. Bill goes up.

Every operational win flows straight into Intercom's revenue. The customer is on the wrong end of the meter.

What counts as a resolution

The vendor defines the meter. Intercom's documented logic counts a resolution when Fin sends a reply and the customer either confirms it solved their issue or does not reply within 24 hours. That second clause does a lot of quiet work.

A customer who reads Fin's reply, gives up, and goes to a competitor is billed as resolved. A customer who is satisfied enough to ignore the follow-up is billed as resolved. A customer who emails support directly because the chatbot's answer was useless is billed as resolved, because they didn't reply in-thread.

The unit is not "customer satisfied." The unit is "ticket marked closed by our rules." Those overlap a lot. They are not the same thing.

Rationing the AI

The second-order effect is worse. When AI is metered per outcome, lean teams ration it.

They gate Fin behind triage rules so it only sees the easiest tickets. They tighten the confidence threshold so it bails out earlier. They watch the dashboard at the end of every month and pull the lever back if outcomes spike. The instinct is sensible: a $990 monthly AI bill at 1,000 resolutions becomes $9,900 at 10,000. So you keep the AI small.

Which means the AI does less. Which means the tool is in tension with the workflow it is supposed to enable. You paid for the per-resolution model because you wanted the AI to handle more. The pricing nudges you to make it handle less.

The alternative

Cove drafts a reply on every ticket. Every one. There is no per-resolution meter, no outcome surcharge, no AI add-on. The £99 is flat and includes the drafting; you pay your model provider (Claude, ChatGPT) directly at cost with your own keys, with zero markup.

Hit 1,000 AI drafts this month: £0 extra. Hit 10,000: £0 extra. Improve your docs and the AI gets more first-touch wins: £0 extra. The cost shape rewards the team for making the AI work harder.

The full Cove vs Intercom comparison has the tier-by-tier maths, including the bit where Series, Campaigns, and Tours stay on Intercom if you need them while the support inbox moves over.

Where Fin wins

Fin is purpose-built for end-to-end autonomous resolution. It picks up a question, attempts a fix from your Help Center, and only escalates when it can't. If your goal is to let AI close tickets with no human in the loop and you accept the per-outcome unit cost for that autonomy, Fin earned its reputation. Cove's default is human-approved drafts, which is a different choice with a different cost curve.

If your support volume is huge and largely deflectable, the per-resolution model can still pencil out, especially if you would otherwise hire three more agents. That is the case Intercom built Fin for. It is a real case. It is not most lean teams.

The shift

Per-seat pricing assumed humans did the work. Per-resolution pricing assumes the AI did the work, then taxes it. Both miss the same thing: the cost of an AI draft is roughly the cost of running the model, and the model already meters itself. Anthropic charges per token. OpenAI charges per token. The provider you are already paying has already done the per-unit accounting.

A second meter on top of the first one is not a pricing model. It is a margin. Zendesk does this with the per-seat fee plus the Advanced AI surcharge. Intercom does it with the seat plus per-resolution. The shape is the same. The vendor sits between you and the model and charges rent.

If your tool charges more when the AI works, the tool is on the wrong side of the trade.