Per-Agent vs Per-Resolution AI Pricing
Two ways support vendors bill for AI, how each behaves when volume spikes, and which model suits predictable budgets versus maximum deflection.
Short answer: support vendors bill AI in one of two ways. Per-resolution charges you each time the bot closes a conversation, so cost tracks value and also tracks your busiest week. Flat credits give you a monthly allowance, so budgeting is dull and predictable. Neither is wrong; picking without deciding deliberately is.
This is the single line item that most often surprises teams in their third month.
How the two models behave
Per-resolution is outcome pricing. The vendor is confident enough in the bot to be paid only when it works, which is a genuinely good sign about the product. Intercom’s Fin is the clearest example.
Flat credits are capacity pricing. Your plan includes an allowance, and whether you use 40% or 95% of it, the bill is identical. EasyChatDesk works this way, with credits allocated per plan per month.
The difference only becomes visible under stress. In a steady month they can cost about the same. In the month you have a shipping delay, an outage or a successful campaign, the per-resolution bill rises in step with the problem.
The scenario worth modelling
Take a store doing 2,000 support conversations in a normal month, where roughly half are repetitive enough for a bot to close.
In a flat-credit model you buy a tier that covers 2,000 and pay the same every month. In a per-resolution model you pay for about 1,000 closures.
Now Black Friday arrives and conversations hit 6,000. The bot closes about 3,000. Flat credits mean you either stay inside your allowance or move up one tier. Per-resolution means you pay three times the usual AI charge in the month your margins are already under pressure.
The bot did its job in both cases. Only one of them is easy to forecast.
Which to pick
Choose per-resolution when volume is steady and predictable, when you are funded and care more about maximum deflection than budget variance, and when you will actually measure the resolution rate you are paying for.
Choose flat credits when your volume is seasonal, when finance wants a fixed number, or when you are small enough that one unexpected invoice genuinely matters.
There is a third case worth naming: if your conversations are mostly complex and account-specific, your deflection rate will be low whichever model you pick, and the right move is to spend the money on agents instead of AI. Reading your last 200 conversations tells you this faster than any vendor demo. If the same five questions cover half of them, AI pays for itself. If every one is different, it will not.
Questions to ask before signing
- What counts as a resolution? Some vendors count a conversation the customer abandoned.
- What happens when I exceed my allowance — overage charges, throttling, or a forced upgrade?
- Can I cap spend, and what breaks when the cap is hit?
- Is the resolution rate reported in a way I can audit?
That first question matters more than it sounds. A generous definition of “resolved” inflates both the vendor’s success metric and your bill.
Where we sit
EasyChatDesk uses flat credits deliberately. The AI chatbot arrives on the Professional plan at $33 per agent per month billed yearly, with an allowance per plan per month, and a busy month costs what a quiet one costs. If you consistently run out, the signal is to move up a tier rather than to throttle the bot.
That is the right trade for teams who want a forecastable bill. It is the wrong trade if you want to pay only for outcomes and can absorb variance — in which case Intercom is the better buy and we would rather say so. Full detail on plans and pricing, and the wider cost picture in how much live chat software costs.
For the tools themselves, see best AI chatbots.
Frequently asked questions
What is per-resolution AI pricing?
You pay each time the AI closes a conversation without a human. Intercom's Fin is the best-known example. The logic is that you pay for outcomes rather than capacity, which is appealing until volume spikes and the bill rises with it.
What is flat AI credit pricing?
Your plan includes a monthly allowance of AI interactions, and a busy month costs the same as a quiet one. EasyChatDesk works this way. The trade is that running out means upgrading a tier rather than paying incrementally for the extra value delivered.
Which AI pricing model is cheaper?
At low, steady volume per-resolution is often cheaper because you only pay for what the bot actually closes. At high or spiky volume flat credits win on both cost and predictability. The crossover depends on your deflection rate, so model your own numbers rather than trusting a vendor example.
Why does per-resolution pricing worry finance teams?
Because the cost is correlated with your busiest periods. A product launch, an outage or a shipping delay triples conversations, the bot earns its keep, and the invoice triples in the same month. The value is real; the forecasting is hard.
How do I estimate my AI support costs?
Take your busiest month's conversation count, estimate what share are repetitive enough to deflect, and apply the vendor's per-resolution rate to that share. Compare it against the flat-credit tier that covers the same volume. Do it on the peak month, never the average.
Level up your customer support
Try EasyChatDesk free: live chat, help desk ticketing and an AI chatbot in one platform.
Start for freeRelated articles
Bad Customer Service Stories, and the Failures Behind Them
Six recurring bad customer service experiences, what actually causes each one operationally, and the specific fix — composites drawn from common support failures.
Read more
Client vs Customer: The Difference and Why It Matters
Client and customer are not synonyms. Here is the actual difference, when each word is correct, and how the distinction changes the way you run support.
Read more
Customer Success vs Customer Support: Key Differences
Customer success vs customer support — reactive versus proactive, different metrics, different skills, and how the two functions should hand off to each other.
Read more