$1 a conversation or $0.99 a resolution? Read what each AI support vendor counts before you compare
Two AI support quotes land on your desk. One says a dollar a conversation, the other says ninety-nine cents a resolution. They look like the same price, and they almost never produce the same invoice.

Per-conversation and per-resolution prices measure different things. KODIF bills every ticket its agent works, at $1 to $0.75 depending on annual volume. Intercom's Fin bills $0.99 per outcome, which includes a customer leaving without asking for more help. Before comparing, get each vendor's definition in writing, run last month's tickets through both, and ask how overages, minimums, hard limits and assumed resolutions are handled.
Two prices that look the same
AI support agents have settled into a handful of pricing models, and two of them are now sitting side by side on a lot of shortlists.
KODIF, which this weekend announced it is combining with the post-purchase platform Cavalry, prices by conversation. Its pricing page, as of November 8, describes one flat annual rate based on conversation volume: 24,000 conversations a year on Starter, 75,000 on Growth and 160,000 on Enterprise, with the per-conversation rate falling from $1 to $0.75 as volume rises. No seat fees and no setup fee.
Intercom prices its Fin AI agent by outcome. Its pricing page lists Fin at $0.99 per resolution, and the help article on outcomes explains how they're measured and charged.
A dollar and ninety-nine cents. It's natural to call that a tie and move on to the demo. Don't.
What each one actually counts
KODIF's page answers the question directly: a conversation is one customer ticket the agent works, start to finish, however many messages it takes, and follow-ups on the same issue don't count twice. If the agent works a ticket and hands it to a person, that's still a conversation. If you go over your annual volume, KODIF bills the overage at the same per-conversation rate and says it will flag it early.
Intercom's definition is narrower in one way and broader in another. A resolution is counted when, after Fin's last answer, the customer either confirms the answer was satisfactory or leaves the conversation without asking for more help. Intercom calls the second kind an assumed resolution. A reply to a greeting doesn't count, and you're charged for at most one outcome per conversation, even if Fin takes several actions. Intercom also offers usage reminders and hard limits.
So one model bills for effort, and the other bills for an outcome that includes silence.
The unit on the quote is a definition, not a number. Get the definition before you compare the numbers.
Decision guide
What each pricing unit counts
| Consideration | Per conversation (KODIF) | Per outcome (Intercom Fin) |
|---|---|---|
| Billed unit | Every ticket the agent works, start to finish | An outcome, counted once per conversation at most |
| Unresolved tickets | Still billed | Not billed unless an outcome is counted |
| Customer goes quiet | Billed as a conversation | Counts as an assumed resolution |
| Commitment | Flat annual package by volume; overage at the same rate | Usage-based; minimum monthly commitment mentioned for other helpdesks |
The unit on the quote is a definition, not a number. Get the definition before you compare the numbers.
Where each model can surprise you
Per conversation is predictable, and that's its strength. You know the annual number on the day you sign. The surprise is in what you pay for: every ticket the agent touches, including the ones it couldn't solve and passed to your team. If your volume is heavy with questions the agent will hand off, such as complex warranty claims or B2B orders, you're paying for those too.
Per outcome looks like it only charges for success, and in a sense it does. The surprise is in the assumed resolutions. A customer who gives up and emails instead, or who closes the chat window to call your phone line, can look the same as one who got what they needed. You aren't overcharged by the definition. You're charged for something that may not have been a resolution in any sense your support lead would recognize.
Both can change at renewal. Pricing for AI agents is still young, and the unit a vendor uses this year may not be the unit it uses next year. Ask what notice you get before the definition or the rate changes, and whether your contract locks the definition as well as the price. We wrote more about that moment in when the renewal email arrives, ask what AI has done to the price.
Commitments matter too. KODIF's is an annual package. Intercom's pricing FAQ mentions a minimum monthly commitment, such as 50 outcomes, when Fin runs alongside another helpdesk. Neither is unreasonable. Both change the math for a small store.
Put both prices on the same month
Here's a simple, hypothetical way to see it. Say a store handles 2,000 tickets in a month, which is roughly KODIF's Starter volume spread evenly over a year. Say half of those end with the customer confirming or simply leaving.
Per-conversation pricing bills the 2,000. Per-outcome pricing bills the 1,000. The per-outcome invoice is smaller, but you should read a sample of those 1,000 before you celebrate, because some of them are people who gave up.
Now say the agent is very good and 1,800 end in an outcome. Per-outcome pricing bills 1,800, and per-conversation still bills 2,000. The gap narrows as quality rises.
Those numbers are made up. Yours aren't. Pull last month's tickets, tag each one by what would have been billed under each definition, and price the same month both ways, minimums and overage included.
How the work moves
How to compare two quotes honestly
- Definitions
Get each vendor's billed unit defined in writing, with edge cases.
- Your tickets
Pull last month's tickets and tag which ones each definition would bill.
- Both invoices
Price the same month both ways, including minimums and overage.
- Quality check
Read a sample of assumed resolutions and unresolved conversations.
- Limits
Confirm spend alerts and hard caps before signing.
Questions to ask in the demo
Ask both vendors the same five questions and get the answers in writing.
What exactly is the billed unit, and what are the edge cases: handoffs to a person, reopened tickets, customers who come back the next day? How are assumed resolutions identified, and can we see them listed? What happens when we exceed the plan, and how early are we told? Can we set a hard monthly cap that stops billing, not just sends an email? And what does the agent do when it hits that cap: stop answering, or hand everything to a person?
You'll learn as much from how they answer as from the answers. A vendor who can show you a list of last week's billed units for a real customer, with reasons, is a vendor you can audit later.
Decide by the work, not the unit
Neither model is cheaper in general. Per conversation tends to suit stores with steady, predictable volume and a strong preference for a known annual number. Per outcome tends to suit stores with spiky volume and a support team willing to audit what counted as resolved.
Whichever you choose, keep one person responsible for reading a sample of billed units every month. That's how you find out whether you're paying for answers or for silence. Put the sample in front of the same person who reads customer complaints, because they'll recognize a false resolution faster than anyone looking at a dashboard.
For the product side of these tools, we've reviewed Intercom Fin's ecommerce connectors and Gorgias's AI agent on socials. If you want help pricing, piloting or building an AI support agent around your own tickets, that's our AI applications and agents practice. Talk to TWIMCO.
Related work: Sutton|Place
We build alongside the people doing the work.
Bring us the problem you’re thinking about.