Mike HeapMike is an experienced Product Manager who focuses on all the “non-development” areas of My AskAI, from finance and customer success to product design, copywriting, testing and more.
Table of Contents
- Per-conversation pricing, in more depth
- How does per-conversation pricing work, and what counts as a conversation?
- What's a "good" per-conversation rate?
- Common misconceptions about per-conversation pricing
- Misconception 1: per-conversation is the same as per-message pricing
- Misconception 2: per-conversation is the same as per-resolution pricing
- Misconception 3: a "conversation" means the same thing everywhere
- Misconception 4: per-conversation pricing balloons because chats spiral out of control
- What per-conversation pricing is NOT, and how it relates to adjacent terms
- How does My AskAI handle per-conversation pricing?
- FAQs
Created time
Jun 2, 2026 10:35 AM
Title length (<60)
Author
Ecomm?
Image
Publish date
Jun 3, 2026
Slug
what-is-per-conversation-pricing
Featured
Type
Article
Ready to Publish
Ready to Publish
Per-conversation pricing is a model where you pay a fixed fee each time a customer starts a conversation with your AI agent, no matter how many messages are exchanged or whether the issue is resolved.
The unit you're billed on is the conversation itself, regardless of message count or outcome. A customer opens a chat, the AI answers, and that counts as one charge whether the exchange runs to two messages or twenty, and whether the AI solves the problem or hands it to a human.
A "conversation" is almost always a session: a back-and-forth thread bounded by a time window. The CX Foundation describes the common version as "a back-and-forth interaction within a 24-hour window" (fun fact: that window is set by the vendor, not by any standard, so it moves around). Everything inside it is one billable conversation.
That single design choice is what separates per-conversation pricing from the two models it gets confused with. Per-message pricing charges for every turn the AI takes; per-resolution pricing charges only when the AI actually solves the issue. Per-conversation sits between them: one charge per thread, win or lose.
If you're reading this, I'd guess you're holding two vendor quotes right now. One priced "per conversation", one priced "per resolution", and you're trying to work out which is cheaper for your volume. Let's get into it.
Per-conversation pricing, in more depth
TL;DR: You're billed once per conversation opened, so the meter tracks how many conversations come in rather than how hard the AI works or whether it succeeds.
Per-conversation pricing ties your bill to inbound volume, regardless of how hard the AI works or how often it succeeds. You pay the same for a one-message deflection as you do for a twenty-message thread that ends in a failed escalation (yes, the ones the AI gets wrong cost you exactly as much as the wins). The meter doesn't care what happened inside the conversation, only that one started.
This is the model a lot of the early chatbot and live-chat tools used, before the category's center of gravity shifted toward per-resolution. It survives today in a handful of CX AI products and across the SMB chatbot market, where a flat "per conversation" rate is easy to understand and easy to put on a pricing page.
The whole model hinges on one definition: what counts as a conversation. A reopen the next morning, a customer who messages twice about two different things, a thread that goes quiet for an hour and restarts: each of those either is or isn't a new billable conversation, depending on how the vendor draws the line.
That's the detail I'd pin down before signing anything, and it's the one most pricing pages leave fuzzy.
How does per-conversation pricing work, and what counts as a conversation?
TL;DR: Monthly cost is your number of conversations times the per-conversation rate. The hard part is the definition of a conversation, since each vendor sets its own session window and counts reopens differently.
The maths is the simplest of any AI pricing model:
Monthly cost = number of conversations × price per conversation
The complication is the variable inside it. "Number of conversations" depends entirely on how the vendor defines a conversation, and vendors don't agree.
Here's how the unit differs across the AI customer service products buyers most often compare:
Vendor | Billable unit | What counts as one "conversation" | Headline rate |
Salesforce Agentforce | per conversation (at launch) | a session the agent engages within a 24-hour window | launched at $2/conversation |
Ada | conversation-based | every automated conversation, regardless of outcome | not public |
Assembled | per conversation | a back-and-forth within a 24-hour window | per published model |
Intercom Fin | per resolution (not per conversation) | billed once per conversation, but only when resolved | $0.99 per outcome, 50/mo minimum |
My AskAI | per ticket (usage-based) | every 2 AI replies, counted in your own helpdesk export | ~$0.10-0.15 per ticket |
Salesforce Agentforce is the example everyone reaches for, since it launched at $2 per conversation, but Salesforce has since moved off pure per-conversation pricing to a hybrid of usage credits and per-user licenses. If you're evaluating Agentforce today, re-derive the cost from a current quote rather than the launch rate.
I've also put Intercom Fin in the table, as the contrast. It's conversation-bounded (one charge per conversation, even if the customer asks several questions) but success-gated, which lands it firmly in the per-resolution camp.
The session window is the part that bites. Reopens inside the window are usually free; a new thread the next day is usually a new charge, so two vendors quoting the same headline rate can produce very different invoices depending on how generously they draw that window.
There's a quiet upside here, though. Because the unit is a conversation, you can usually count your own conversation volume straight out of your helpdesk before you sign (something per-resolution can't promise, since there the billable number depends on how good the AI turns out to be).
What's a "good" per-conversation rate?
TL;DR: Published rates run from about $0.50 to $2.00 per conversation, but the rate means nothing until you multiply it by your own volume.
Published per-conversation rates run from around $0.50 at the SMB chatbot end up to the $2.00 Salesforce launched Agentforce at. Most broader CX agents that use the model land somewhere in the $1.00 to $1.50 range, and the enterprise vendors who price this way (Ada, for instance) don't publish a number at all.
Tier | Rate per conversation | Usually means |
Budget / SMB | around $0.50 | SMB chatbot platforms, narrower scope |
Mid | ~$1.00-$1.50 | broader CX agents |
Premium | ~$2.00 | Salesforce Agentforce's launch rate |
Enterprise | custom / not public | Ada and enterprise-tier vendors |
The rate on its own tells you almost nothing, which is the trap. At 10,000 conversations a month, $2 per conversation is $20,000; a per-ticket model at roughly $0.10 is about $1,000 for the same volume. The headline number only becomes a cost once you multiply it by your volume, so the comparison that matters is rate times your conversations.
Per-conversation pricing has one structural advantage. Because you pay per conversation opened, your bill doesn't rise as your resolution rate climbs. Per-resolution works the other way around, since there it's getting better that makes the invoice grow (we walk through that maths in full in our guide on per-conversation versus per-resolution pricing).
Common misconceptions about per-conversation pricing
TL;DR: Per-conversation, per-message and per-resolution are three different meters, a "conversation" is defined differently by every vendor, and the fear of chats spiraling rarely shows up in real data.
Misconception 1: per-conversation is the same as per-message pricing
Per-message (or per-reply) pricing charges for each turn the AI takes; Aissist, for example, charges $0.09 per interaction with a $0.60 per-resolution cap. Per-conversation bundles every message in the thread into a single charge, so if your conversations run long, that bundling works in your favor.
Misconception 2: per-conversation is the same as per-resolution pricing
With per-conversation you pay whether or not the AI solves the issue. The CX Foundation puts it plainly: this model "bills successful and unsuccessful interactions at the same price." That's good news when your AI is good, and a cost you carry when it struggles.
Misconception 3: a "conversation" means the same thing everywhere
Session windows, reopens and multi-topic threads are all defined by the vendor, and the time window "may vary among different vendors." Two products at an identical headline rate can bill quite differently once you account for how each one counts.
Misconception 4: per-conversation pricing balloons because chats spiral out of control
This is the fear we hear most, and in our own rollouts we've never seen it happen. People worry they'll end up paying for endless back-and-forth threads, but a support agent locked to a company's own business isn't ChatGPT, so once a customer realizes it will only answer questions about the product, they get bored of testing it very quickly.
Roughly 0.5% of conversations ever run into the tens of messages. The vast majority are a few messages at most, partly because a good agent answers directly and skips the "hi, how are you?" padding.
What per-conversation pricing is NOT, and how it relates to adjacent terms
TL;DR: Per-conversation sits between per-message (finer) and per-seat (coarser), and its closest cousin is per-ticket. All of them are shapes of usage-based pricing.
Per-conversation pricing is one meter in a family of them. Here's how it sits against the terms it's most often confused with:
Term | What it charges for | How it differs from per-conversation |
Per-resolution pricing | only when the AI resolves the issue | per-conversation pays regardless of outcome; per-resolution is success-gated and rises as the AI improves |
Per-message / per-reply pricing | each AI turn | per-conversation bundles all turns in a thread into one charge |
Per-seat pricing | each human agent license | per-conversation scales with customer volume, not with headcount |
Per-ticket pricing | each support request | closely related; a ticket is tied to your helpdesk's own unit, which you can audit in your export |
Usage-based pricing | activity (tickets, conversations, messages) | per-conversation is one shape of usage-based pricing |
Outcome-based pricing | results delivered | per-conversation pays for the attempt, not the result |
The cleanest way to hold it in your head: per-conversation is a usage-based model (you pay for activity, meaning the conversations handled), it's coarser than per-message and finer than per-seat, and its closest cousin is per-ticket. The difference between those last two is simply who defines the unit.
How does My AskAI handle per-conversation pricing?
TL;DR: My AskAI bills per ticket on a usage-based model at about $0.10 per ticket, tied to a unit you can audit in your own helpdesk. It shares per-conversation's predictability without the "what counts as a conversation" ambiguity.
We don't bill per conversation. My AskAI is priced per ticket on a usage-based model, roughly $0.10 per ticket, and we deliberately steer clear of per-resolution or outcome-based pricing.
The billing unit is a credit: on helpdesk chat, every two AI replies is one credit; on email, the first reply is a credit and each follow-up is half a credit. On our own widget, an entire one-hour session counts as a single credit no matter how many messages it holds (that widget session is the closest we get to a per-conversation unit, but the model overall is per-ticket).
We picked per-ticket over per-conversation for one reason: you can audit a ticket in your own helpdesk export, whereas a "conversation" is a unit the vendor defines with a session window you don't control. Per-conversation and per-ticket are cousins, really: both are volume-metered, both are forecastable, and crucially neither one taxes you as your resolution rate climbs.
That last point is where both beat per-resolution. With usage-based pricing your cost per resolved ticket falls as the AI gets better, because most of what drives that improvement is your own work (the knowledge you add and the APIs you connect).
The objection we hear next is usually about volume: what happens if tickets spike one month? My answer is that a spike is the exact scenario AI is built for.
Would you rather have an AI absorbing that volume, or be scrambling to hire, onboard and train ten or twenty extra agents, or worse, leaving customers with a degraded experience because you couldn't? A pricing model that flexes with volume is doing exactly what you want when the volume arrives.
The proof sits in customer rollouts. TravelJoy went from 24% to 80% AI resolution after switching off a per-resolution-style native AI, and Kriptomat rejected Intercom Fin at $0.99 per resolution as uneconomical for their volume.
Edel Optics jumped from 25% to 79% resolution after their own team added a User Data API, a lift a per-resolution vendor would have billed them for, but which cost nothing extra under usage-based pricing. The boring-but-effective option here is the one where the upside of your own work stays with you.
FAQs
What is per-conversation pricing?
Per-conversation pricing charges a fixed fee each time a customer starts a conversation with your AI agent, regardless of how many messages are exchanged or whether the issue is resolved. A "conversation" is usually a session bounded by a time window, commonly 24 hours.
Which AI support tool gives the best value per conversation?
It depends on your conversation volume and the headline rate, but in practice a per-ticket model (around $0.10 per ticket) usually beats a per-conversation rate of $0.50 to $2.00 at real volume. Because you're billed per ticket rather than per session, the cost doesn't climb as the AI improves. I'd always tell a buyer to multiply the rate by their own volume before trusting the per-unit figure.
What counts as a "conversation" in per-conversation pricing?
Usually a back-and-forth session within a time window, commonly 24 hours. Messages inside the window are one conversation; a new thread outside it is typically a new charge. The exact window is vendor-defined, so I'd check it before signing.
Is per-conversation pricing the same as per-resolution pricing?
No. Per-conversation charges you whether or not the AI resolves the issue; per-resolution charges only on success. The practical difference is that per-resolution bills rise as your AI gets better, while per-conversation bills track how many conversations come in.
Is per-conversation pricing the same as per-message pricing?
No. Per-message (or per-reply) pricing charges for each turn the AI takes; per-conversation bundles every message in a thread into one charge. For the multi-turn conversations we see most often, per-conversation is usually the cheaper of the two.
Is per-conversation pricing cheaper than per-resolution?
It can be, and the gap depends on your resolution rate. Per-conversation doesn't carry per-resolution's "improvement tax", so your bill doesn't double when your resolution rate doubles. You do, though, pay for conversations the AI fails to resolve, which per-resolution wouldn't charge for.
Who charges per conversation for AI customer service?
Salesforce Agentforce launched at $2 per conversation (it has since shifted to a credits-plus-license model), Ada uses a conversation-based model, and Assembled bills per conversation. Plenty of SMB chatbot platforms price this way too. Intercom Fin, by contrast, is per-resolution.
How do I forecast my per-conversation AI costs?
Pull your conversation volume from your helpdesk for the last few months, multiply by the vendor's per-conversation rate, and stress-test it against your busiest month. Because the unit is a conversation rather than a resolution, the volume you're forecasting is something you already measure.
Does per-conversation pricing punish me as my AI gets better?
No, and this is its main advantage over per-resolution (it's the part we like most about a volume-metered model). You pay for each conversation that opens, so improving your resolution rate doesn't raise your bill. Under per-resolution pricing, that same improvement is exactly what makes the invoice grow.
What happens if a customer reopens a conversation, do I pay twice?
It depends on the vendor's session window. Reopens inside the window (often 24 hours) are usually free; a new thread started after the window typically counts as a new conversation. This is one of the details worth pinning down in writing before you commit.
What's the difference between per-conversation and per-ticket pricing?
They're close cousins, since both meter on volume and both are forecastable. The difference is who defines the unit: a ticket is defined by your helpdesk and visible in your own export, while a conversation is defined by the vendor's session window. We price per ticket for exactly that reason.
When is per-conversation pricing a bad deal?
When you have very high volumes of short, consumer-facing chats and the per-conversation rate is high, the per-unit cost can add up faster than a per-ticket or per-reply model would. The "conversations will spiral" fear is mostly overblown (in our experience around 0.5% run long), so the real risk is a high rate meeting high volume, well before runaway thread length ever becomes the problem.
How does My AskAI charge, per conversation or per ticket?
Per ticket, on a usage-based model at roughly $0.10 per ticket. We don't bill per conversation or per resolution. The one place we use a session-style unit is our own widget, where a one-hour session counts as a single credit regardless of message count.
Is per-conversation pricing usage-based or outcome-based?
Usage-based. You're paying for activity, meaning the conversations handled. Outcome-based pricing, including per-resolution, charges for the result instead, which is a fundamentally different model.
