Deciding what not to charge for

We charge $0.50 per resolved ticket. Setting that number took an afternoon.

Deciding which tickets don't count took about 6 weeks and 3 arguments I remember clearly.

The question every buyer asks

Second or third call, always the same one, usually phrased politely: "so you get paid more the more tickets there are?"

They're not being difficult. They're doing the right thing. When you tell a company you charge per resolution, the first thing a decent operator does is look for where your incentives point away from theirs.

If you don't have a good answer to that question, your pricing model is a liability no matter how clean the headline number is.

What we ended up excluding

Spam doesn't count. Obvious, but it has to be stated, because spam volume is enormous and if you don't exclude it you're literally billing for junk mail.

Automated notifications don't count. Monitoring alerts, delivery receipts, out-of-office replies, the noise that lands in a support inbox and isn't a customer.

No-touch tickets don't count. If nobody did anything, human or AI, there's no charge. A customer emails, then replies "never mind, figured it out," and that's free.

Duplicates from the same person on the same issue collapse into one. Someone emails, then opens the chat widget, then replies to the confirmation. One problem, one charge.

Our own failures don't count. If the AI answers wrong and a human has to redo it, we don't bill twice. We bill once, for the human resolution, which is worth more to us than the 50 cents because it means we're not being paid for being bad at the job.

The arguments

The duplicate rule was contentious internally, and the objection was fair: it's real work either way. The system processed 3 conversations. Someone had to.

The counter that won was that a customer looking at their invoice doesn't see 3 units of work. They see one problem and three charges, and they feel taken. You can be technically right and still lose the account on a line item.

The failure rule was easier morally and harder commercially, because at volume it costs actual money. Nobody was excited about it. We shipped it anyway because I couldn't write the pricing page without it and have the page be honest.

The one we still don't have right is long-running tickets. A conversation that goes 14 messages across 3 weeks with 2 escalations is one resolution and one charge, and it costs us a lot more than 50 cents to serve. We eat it today. We'll probably have to do something about it and I don't yet know what.

Why the exclusions do more work than the price

The exclusions are what make the price legible.

$0.50 per resolution with no exclusions is a number a buyer has to defend against their own imagination. They picture the worst case, which is a bot resolving thousands of junk tickets and billing for all of it, and then they price in that fear.

$0.50 per resolution with a written, specific list of what doesn't count is a number they can take to finance. The list is the product of the pricing model. The number is just a number.

It also disciplines us. Once you've published that you don't bill for your own failures, the failure rate becomes a line item in your own P&L, and there's nothing like watching bad output cost you money to focus a roadmap conversation.

If you're designing usage pricing

The thing I'd tell someone starting this: write your exclusions before your prices, and write them as though a skeptical CFO is going to read them out loud on a call. Because one will.

Then check what each exclusion does to your own incentives. Every one of ours makes us slightly poorer and makes the product slightly better, which is the only direction that survives contact with customers over a few years.

We're 8 months into billing this way. Ask me again when someone's been on it for 3 years and has run the full analysis on us.