Per-resolution, per-conversation, or flat: which AI pricing actually wins

By Andrew Diaz · JUN 18, 2026 · 6 min read

Everyone obsesses over the sticker price of an AI assistant. The number that actually decides whether it pays off is the pricing model — how you’re billed, month after month, as you grow. Get the model wrong and a “cheap” bot becomes the expensive one. Here are the three you’ll be quoted, and what each one really costs you.

The three models you’ll be quoted

ModelHow you’re billedBest whenThe catch
Per resolutionA fee per “resolved” conversation (~$1–$6)Support-heavy, steady ticket volumeWho defines “resolved”? Cost climbs as you grow, and a human floor may be doing the resolving
Per conversationA fee every time someone engages — resolved or notLow, predictable message volumeYou pay for tire-kickers, duplicates, and retries; spam inflates the bill
Flat retainerOne predictable monthly fee, set per buildSales + support, variable or seasonal volumeYou need a partner who won’t cap usage or nickel-and-dime add-ons

The problem with paying per outcome you can’t audit

Per-resolution pricing sounds fair — pay for results. But “resolution” is a soft metric the vendor defines and counts. Was a one-line FAQ answer a resolution? A conversation the customer abandoned? You’re trusting someone else’s tally on the exact number that drives your bill, and the incentive runs the wrong way: the more they count as resolved, the more you pay.

Why per-conversation punishes the wrong thing

Per-conversation billing charges for engagement regardless of outcome. That means you pay for the spam, the wrong numbers, the person who said “hi” and left, and the same customer coming back three times. It taxes exactly the thing you want more of — people reaching out — and turns a busy week into a bigger invoice with nothing to show for it.

What a flat retainer changes

A flat monthly retainer decouples your cost from your volume. Your busiest month and your slowest month cost the same, so you can forecast it and you’re never penalized for growth. There’s no meter to dispute and no soft metric to audit. The vendor is paid to make the system work, not to rack up units — which is the alignment you actually want.

The question under the question

Underneath all three models is one question: are you buying software, or a metered service with a human floor? Some of the lowest per-unit prices are backed by a roster of human agents — you’re renting a contact center by the conversation, at contact-center margins. A flat retainer on software you own is a different thing entirely: the cost is the cost, and the system is yours.

Price the outcome, not the unit. The cheapest model on paper is rarely the cheapest one once your volume — and your success — start to climb.