The price on the homepage is a seat. $29, $30, $55. You turn the AI on, it closes a conversation, and a second number shows up.
That second number is the business. The seat is how they get you through the door.
I'm not saying the invoice is hidden in a footer. The pages say $0.99, or $0.75, or "automated resolutions," which is a lovely way to describe a meter. I call it a scam because you get billed for the outcome they sold you. You get better at support. The bill goes up. You did the thing the onboarding nagged you to do.
The meter runs when you get good
Fin is $0.99 every time it closes a conversation without you. Help Scout is $0.75. Zendesk makes you buy Professional at $115, then Copilot at $50, then about $1.50 a resolution after ten free ones per agent. You pay three times before the meter even starts. I had to read that page twice, and I build this stuff.
Write the password-reset article. You were tired of pasting it. The agent starts closing those, which was the point of writing it, and that is the week the invoice moves. Same week. That's the trick.
Your best month of support is your most expensive month. The team that writes things down and lets the agent take them pays more than the team that leaves the queue to humans and goes to bed annoyed. I haven't seen a pricing page say that out loud.
"Resolution" is their word
Get them to finish one sentence. What event creates a charge? If the call moves on without it, you don't have a price. You have a vibe, and a contract.
On Fin and Help Scout it's usually this: the AI handled it, and nobody had to jump in. If the customer still needs a person, it often doesn't count. Sounds fair on the call. It also means you can't forecast from "we had 400 conversations." You forecast from a definition you don't control, a month later, after the month already happened to you.
A customer told us what that does to the agent. Theirs was Fin. It would work very hard to resolve a conversation it couldn't actually resolve, instead of handing it to a person. Resolving is how they get paid. Escalating isn't. So the customer sits there with a confident answer that isn't one, and the invoice gets the dollar anyway.
Help Scout lets you put a cap on it. Hit the cap and the agent turns off. People wait on you again. The safety feature is a limit on the product working. I'd rather leave it on and know the answer costs nothing.
Zendesk splits the same idea into three purchases. Copilot assists the human, and you can only buy it once you're on a plan that's allowed to sell it. The thing that actually closes conversations is a third meter, past an allowance that disappears the week the agent becomes worth having. Ten free resolutions per agent. You'll use those before lunch, on a bad Monday.
Why the demo felt like a deal
The trial costs nothing. The questions they show you are ones the help center already knows, or ones that don't hit the meter yet. Invoice's empty. You feel a bit clever. You sign.
Then real mail shows up. Failed payment. How to export. The bug you fixed in March and are still explaining in November. You write them down because you're sick of typing the same paragraph. The agent gets good, which is the whole game. The bill lands the month you finally took a weekend. That's the joke I can't get over. The month support stopped eating your evening is the month they invoice you for it.
Nobody put that in the budget. The number on the homepage was the seat. Of course it was. That's what homepages are for.
Transparent pricing
Fernand is free up to 100 conversations. After that, $29 for the people who can reply. The agent is on the plan. Plain bundles AI too, at $55 a seat, if you're a bigger technical team and that shape fits. Neither of us sends a second invoice because the help center did its job. That's the argument. The rest is taste.
On a demo call I ask one thing. What do 1,000 AI-handled conversations cost next month, in actual dollars, on this plan? If the answer depends on how many the AI "resolves," you're offering to pay for getting better at your own product.
On Fernand, Claude drafting those replies doesn't change the meter.