The sticker price is never the price. I learned that the hard way, and most store owners I talk to learn it on their first invoice.
AI chatbot vendors price in different units on purpose. One charges per message, one per resolution, one per seat, one a flat monthly tier with overage once you pass a quota. Some quote a low platform fee and then bill the AI on top. The result is that you cannot put two quotes side by side and know which is cheaper. That is the point.
So let me decode the models, show you where the hidden costs hide, and give you a way to compare quotes that holds up.
The pricing models you will see
There are four common ways AI chatbots charge, and each one shifts cost onto a different part of your usage.
Per message
You pay for every message the bot sends or receives. It sounds cheap at a few cents a message until you remember that one real conversation is rarely one message. A shopper asking about sizing, shipping, and a discount can run ten messages. Now your "cheap" per-message rate is a dollar for one conversation, and a chatty customer costs more than a quiet one even when neither of them buys.
Per resolution
You pay when the bot resolves a query without a human. Better aligned to value, but the definition of "resolved" is the vendor's, not yours. Ask how it is measured. If a deflected question that the customer re-asks an hour later counts as two resolutions, you are paying twice for one unhappy shopper.
Per seat or monthly tier
You pay a fixed monthly fee per agent seat or per tier, with a conversation quota baked in. Predictable, until you grow past the quota and hit overage rates, or until you realize you are paying for seats in months where traffic is low. Tiers also tend to gate features: the analytics or the handoff you need sits one plan up.
Pay-as-you-go
You pay a flat rate per conversation, no monthly minimum, no seats. A conversation is a conversation whether it took two messages or twenty. You pay for what shoppers use and nothing in a slow month. This is the model Emporiqa uses, and I will be upfront that I built it this way because the other three burned me as a buyer.
Where the hidden costs hide
The headline rate is only part of the bill. Four line items show up later and decide what you end up paying.
- AI model fees on top. Some platforms quote a low subscription and then ask you to "bring your own key" for the language model, so OpenAI or Anthropic bills you separately. That second invoice is the one that hurts at volume, and it is the one buyers forget to ask about.
- Overage rates. The included quota looks generous until a good month pushes you over it, and the per-unit price above the quota is usually much higher than the price inside it.
- Annual lock-in. A discount for paying yearly is a discount for committing before you know if the tool works on your traffic. If you can only test for a month and then must sign for twelve, the discount is not for you.
- Feature gating. The number on the page covers the base tier. Conversion tracking, human handoff, or extra languages can sit one or two plans up, so the plan you need costs more than the plan you were quoted.
How to compare quotes that do not match
Here is the question that cuts through all of it. Ask each vendor, in writing:
"What is the all-in cost of 1,000 conversations next month, including AI model fees, with the features I need turned on?"
Pick a conversation count close to your real monthly support volume. Make them include the model cost and the features. A vendor that gives you a clean number is one you can compare. A vendor that cannot, or will not, is telling you something.
Then divide by 1,000. Now every quote is a single number, cost per conversation, all in, and you can finally put them next to each other.
What pay-as-you-go looks like in practice
Since I price Emporiqa this way, here is the whole thing with no asterisks.
- $0.25 per conversation. One flat rate, however many messages the conversation takes.
- AI model costs included. No second invoice, no key to manage. The language model is part of the $0.25.
- $25 of signup credit. Roughly 100 conversations to test on your real traffic. No card at signup.
- $59 per month default cap. A ceiling you set, adjustable anytime from the billing dashboard, so a traffic spike cannot surprise you.
- No monthly minimum, no annual contract, no per-seat fee. Every account includes unlimited team members, full analytics, conversion tracking, and human handoff.
For most stores that is the entire pricing page. Catalogs over 30,000 products move to Enterprise with custom rates, and that is the only other plan.
A worked example
Say your store handles 800 shopper conversations in a month.
On pay-as-you-go at $0.25, that is $200, model costs included. You would set your monthly cap above the $59 default to allow that volume, and the widget pauses only if you reach the ceiling you chose. On a per-message plan at $0.05 a message with an average of eight messages per conversation, the same 800 conversations is $320 before any separate AI fee. On a per-seat plan you are paying the seat fee whether those 800 conversations happen or not, plus overage if your quota was lower.
The numbers move with your assumptions, which is exactly why you should run them with your own traffic before you sign anything. Use conservative counts. If the case only works at best-case volume, it does not work.
When each model makes sense
Per-seat tiers fit teams that want one predictable line item and have steady, high volume. Per-resolution can fit support-heavy stores that trust the vendor's definition of resolved. Per-message rarely wins once you count messages honestly. Pay-as-you-go fits stores that want to pay for outcomes, test without commitment, and never pay for a quiet month.
There is no single right answer. There is only the right answer for your traffic, which you find by running your own numbers with model costs included.
The short version
Decode the unit. Add the hidden costs. Ask for the all-in price of a realistic conversation count in writing. Then compare one number to one number.
And pay for conversations that happen, not for seats, quotas, or a separate AI bill you did not see coming.
Want to see the per-conversation model on your own catalog? Create a free Emporiqa account with $25 of signup credit (about 100 conversations), no card required, and watch the cost per conversation with your real products. Prefer a look first? Try the live demo. The demo is a stocked electronics store, and the chat behaves the same on any catalog. Or read how to calculate the ROI once you know the cost.
