Store AI chatbots: the cheapest rate can cost the most

The short version

Sources: Fin pricing outcomes and Fin AI Agent outcomes, Gorgias AI Agent pricing, Tidio pricing, Zendesk pricing and the Shopify Inbox listing, all read September 2026.

You are buying a definition, not a rate

The category has moved from per-seat licensing to outcome billing, which sounds like a straightforwardly better deal: you pay when the robot works. The catch is that “works” is defined by the party sending the invoice, and the four definitions below were written independently of each other.

Read the two ends of that table against each other. Gorgias and Zendesk bill a success, and both say so explicitly: Gorgias states that “you are not charged an automated interaction for handed-over conversations”, and Zendesk that you “pay only for customer requests that were successfully resolved by the AI agent, without any escalation to a human agent”. Tidio bills an attempt. A Lyro conversation that fails completely, annoys the customer and lands on your desk is billed identically to one that solves the problem.

Neither approach is dishonest. But they are not the same product being sold at different prices, and every comparison that ranks these tools on their per-unit rate is adding up units that measure different things.

Tidio runs two meters side by side, which is easy to miss. Its “billable conversations” quota counts only conversations “that include a message from a human agent”, while Lyro AI conversations are bought as a separate block. AI and Flow replies do not consume the human quota, and vice versa.

The handoff you configured costs more than the one you did not

Intercom’s outcome list contains the sharpest detail we found in this category, and it is the one that should change how you set the tool up.

Both outcomes are identical from every angle that matters operationally. The conversation reaches a person. Your team does the same work. The customer experiences the same thing. The only difference is why the bot let go: because you wrote a rule telling it to, or because it ran out of road on its own.

The first one costs a dollar. The second is free.

Now hold that against standard implementation advice, which is to write explicit routing rules for the topics you do not want a language model improvising on. For a store that is exactly the high-stakes list: refunds, order changes, address edits before dispatch, chargebacks, anything touching payment. Route those deliberately and you have converted your free escalations into billable handoffs. Leave the bot to flounder until it gives up and you pay nothing, at the cost of a worse conversation.

This is not an argument for degrading the experience to dodge a dollar; one refund handled badly costs more than a year of handoff fees. The point is narrower: a well-configured Fin costs measurably more than a badly-configured one on identical traffic, so model the good configuration before signing, not the default.

Silence bills as a solved problem

Intercom splits resolutions into confirmed and assumed. Confirmed is what you would expect: the customer says something affirmative. Assumed is the interesting half — “if a customer disengages from the conversation for 24 hours after Fin’s last answer, it is considered an assumed resolution.”

For a software helpdesk that is a defensible proxy. Someone with a real problem usually comes back. For a storefront widget it is a much looser fit, because a large share of ecommerce chat is not support at all. It is a browser asking whether the jacket runs small, reading the answer, and closing the tab to think about it. Twenty-four hours later that bills as a resolution, at the same price as a genuine problem solved.

This is the published term, not a hidden charge. But it means the invoice cannot distinguish pre-sale browsing from post-sale support — and browsing scales with traffic rather than orders, so a seasonal spike raises the bill even if nothing goes wrong all month.

What that does to the bill

Take a store where the AI replies in 200 storefront chats a month and split them the way a store, rather than a SaaS helpdesk, tends to split: 110 where the customer does not come back, 50 routed to a human by a rule you wrote, and 40 where the bot gives up by itself. That mix is our illustration and yours will differ; what matters is that all three tools see the identical traffic.

The meter reading alone varies by 1.8x across the three before a price is applied. Then the ranking inverts twice over. Tidio’s effective entry-tier rate is 38% below Gorgias’s and produces a bill 31% higher. Intercom’s rate is 10% above Gorgias’s and produces a bill 60% higher. On this traffic the cheapest published rate in the set is the second most expensive tool to run.

And the single biggest lever is not the vendor. If all 90 human-bound chats reached your team as escalations rather than configured handoffs, the Intercom bill falls to $108.90 — the 50 configured handoffs are $49.50 of the total, about 31% of it, for no change whatsoever in what your customer or your team experiences.

Our arithmetic on published rates, not a vendor figure. The Tidio unit rate is the entry quota block, $32.50 for 50 conversations, divided out; its ladder above that tier did not render to us, so treat $0.65 as the entry-tier rate rather than the price at 200. Gorgias and Intercom also charge for the helpdesk underneath the AI layer, and Intercom applies a minimum monthly outcome commitment, so none of these is a total bill.

The free option is real, and it is not a helpdesk

Before any of the above, check whether you need a meter at all. Shopify lists Inbox as free, and it is not a stub: the listing describes an AI sales associate where “buyers can ask questions and get help finding the right products from your catalog”, which “learns from your catalog, policies, and store content while you control its tone, style, and rules”, and which can personalise replies from preferences and purchase history when a buyer is signed in.

That covers a real share of what a storefront widget is asked — and, pointedly, it covers the pre-sale browsing traffic that the assumed-resolution rule makes expensive elsewhere.

What it does not give you is a helpdesk. There is no ticketing, routing or multichannel inbox pulling email and social into one queue, which is the actual product Gorgias and Zendesk sell. So the upgrade trigger is not volume, it is shape: if your problem is answering product questions on the storefront, a meter buys you little. If your problem is that support is arriving through four channels and nothing is tracked, that is what you are paying to fix, and the AI rate is a secondary consideration.

This is the same test we applied to automation task pricing: metered tools are usually bought a tier too early, and the fixed platform cost underneath the meter is the part small operators underestimate. If you are still deciding on the store itself, start with our Shopify review and the real cost of Shopify transaction fees before adding a fourth subscription to the pile.

What we could not verify

Stated plainly rather than filled in with numbers we could not read at source.

• Zendesk’s price per automated resolution. Its pricing page publishes the definition and the seat prices — Support Team $19, Suite Team $55, Suite Professional $115 per agent per month billed yearly — but states only that AI agent pricing is “based on the successful outcomes they deliver” and links out. Third parties consistently report $1.50 committed and $2.00 pay-as-you-go; neither figure appears in our table and no Zendesk bill is calculated anywhere on this page.
• Gorgias plan prices and ticket allowances. The pricing page is rendered client-side and gave us only the model, verbatim: “AI Agent on every plan, pay only when it resolves a conversation” and “Helpdesk scales from 50 to 5,000 tickets a month, never priced per agent.” The per-resolution rates are first-party, from Gorgias’s own pricing explainer. The tier prices are not, so none appears here.
• Tidio’s Lyro rate above the entry tier. We read $32.50 for a 50-conversation block and the available quota steps up to 1,000, but not the prices at each step. Our $0.65 is the entry block divided out and is labelled that way wherever it is used.
• A correction we nearly published. A widely repeated third-party claim is that a Gorgias AI resolution also consumes a helpdesk ticket, so automation bills on both meters at once. Gorgias’s own documentation says the opposite: when the AI cannot resolve a conversation and hands it to a human, “that ticket shifts over to your regular Helpdesk plan… You’re not charged for both.” We have not tested an invoice, but the vendor’s published position is explicit and we report that rather than the roundups.
• Real-world resolution rates. The 110/50/40 split is our illustration chosen to make the mechanism visible, not a measurement of any store. Every vendor publishes deflection claims; we have not verified any of them and none is quoted here. Substitute your own mix before deciding anything.
• Enforcement and disputes. Everything above is published wording. We do not know how assumed resolutions are audited, whether miscounted outcomes can be challenged, or how often they are.

FAQ

What we would actually do

Ask the vendor one question before the demo: which events on my account will bill next month, and which will not? Every tool here answers it in writing, the answers differ materially, and the answer is worth more than the rate card. If a sales conversation resists that question, treat it as the finding.

Then start free. Shopify Inbox costs nothing, covers storefront product questions, and gives you a month of real data on what your customers actually ask — which is the input every quote above depends on and the one nobody has on day one. Buy a meter when the shape of the problem changes to routing and tracking, not when the volume merely rises.

And model the good configuration, not the default. The Intercom numbers here move by about a third on nothing but whether you wrote the routing rules that any competent setup would include.

Related reads: building a Shopify store with AI, Zapier vs Make vs n8n task costs, Shopify transaction fees in full, best AI tools for small business, and our Shopify review.