AI Chatbot Pricing Models Explained: Per Session, Per Conversation, Per Resolution and Per Seat
The short answer. AI chatbots and AI agents for ecommerce are priced by one of five units: seat, ticket, conversation, resolution or session. The unit decides how your bill moves when your team grows, when support spikes or when traffic jumps, and that matters more than the headline price. To compare quotes, turn each one into a monthly bill at your own volumes for a normal month and a peak month, and get the definition of the unit in writing. The worked example below shows why no unit costs less in all situations.
Disclosure: Rep AI sells AI agents for Shopify. Its core plans are priced by monthly website sessions, which is one of the five units below. This guide describes all five units in the same terms and quotes no vendor prices.
Prices change, so this guide explains units and method. It does not quote what any vendor charges.
Why the unit matters more than the price
Two quotes can look equal on a pricing page and behave very differently after you sign. One grows when your team grows. Another grows when shoppers write in. A third grows when traffic grows, whether or not anyone chats.
So the useful question is not "which plan is cheapest today?" It is "which number does this bill follow, and can I forecast it?" A bill that follows something you cannot control, such as a shipping delay that floods your inbox, needs a cap or a plan for the bad month.
The five units at a glance
| Unit | What is counted | The bill follows | Ask the vendor |
|---|---|---|---|
| Seat | Each person with a login to the tool | Team size | Do AI features cost extra on top of the seat? Do part-time staff or collaborators count? |
| Ticket | Each case created in a helpdesk | Number of cases | Does a chat the AI handles count as a ticket? Do merged, spam or reopened tickets count? |
| Conversation | Each exchange between a shopper and the AI | Volume of shopper contact | When does a conversation start and end? Does a returning shopper start a new one? |
| Resolution | Each conversation the AI closes without a person | How much the AI handles | Who decides a conversation is resolved, and what happens if the shopper comes back? |
| Session | Each visit to your website in a month | Your traffic | How are sessions counted, and what happens above your tier? |
Per seat
A seat is a person who can log in. It is built for helpdesks where human agents work through tickets. The bill does not rise when shoppers contact you more, although your team's workload does.
The main question is what the seat includes. Some products bill the AI as a second line, priced by usage, on top of the seats. Ask for a quote that shows both lines, and ask whether collaborators who just read or comment need a paid seat.
Per ticket
A ticket is a case in a helpdesk. Plans are often tiered by monthly ticket volume, with a charge for tickets above the tier. This is easy to model if your ticket count is steady.
The risk is definition. If the AI answers a chat and no human touches it, does that create a ticket? Does a reopened ticket count twice? Does a conversation that moves from chat to email count once or twice? Ask how the count treats AI-handled conversations.
Per conversation
A conversation is a thread between a shopper and the AI. Counting is simple until you ask where it ends. A shopper who asks about sizing in the morning and shipping in the evening may be one conversation or two, depending on the vendor's rule.
Ask about the time window, whether conversations the AI starts but the shopper ignores are counted, and whether conversations handed to a person are counted. These details decide whether the number on your invoice matches the number you would count yourself.
Per resolution
Pay when the AI resolves the question. This lines the vendor's fee up with a good outcome, so it can sound the fairest. Whether it is fair depends entirely on what "resolved" means.
Ask these questions in writing:
- Is a conversation resolved if the shopper simply stops replying?
- Is there a window after which a returning shopper reopens it, and is a reopened conversation billed twice?
- If the AI says it is not sure and passes the shopper to your team, is that a resolution?
- Can you see and sample the conversations you were billed for?
A fee that depends on a vendor's own judgment of success is easier to trust when you can read the conversations behind it.
Per session
A session is a visit to your website. It is counted whether or not the shopper talks to the AI, so the bill follows traffic, not ticket volume. You can set it against a traffic forecast, but overage rules and caps vary by plan, so a traffic spike can change the bill. The flip side is that you pay for traffic on days when nobody chats.
Rep AI's core plans (Sales, Support and Concierge) are priced by monthly website sessions. Its beta add-ons (helpdesk, email, social and WhatsApp AI) carry separate pricing. To see the tiers at your own traffic, use the pricing page.
Ask any vendor that prices by session:
- How is a session counted, and are bots and your own team's visits excluded?
- Does the figure match your analytics, and which analytics tool is the reference when they differ?
- What are the tiers, and what happens when traffic passes yours: a price step, a per-block charge, or the AI stopping?
- Is there a warning before you reach the limit?
Hybrids and extras
Many plans combine units: a base fee plus usage, a free starting tier, a minimum monthly spend, or an annual discount. Channels, onboarding and premium support are often add-ons. Ask for one page that lists each charge you would pay at your volume, including any minimum.
A worked example (invented numbers, for method only)
These figures are made up to show the arithmetic. They are not a market rate, not any vendor's price and not a benchmark. Each rate is set so that a normal month costs 1,000 dollars under each model. That removes price from the comparison and shows how each unit behaves when conditions change.
The store in a normal month: 200,000 website sessions, 4,000 shopper conversations with the AI, 2,000 of them resolved by the AI without a person, 2,000 tickets reaching the helpdesk, and 4 agent seats.
The invented rates: 250 dollars per seat; 0.50 dollars per ticket; 0.25 dollars per conversation; 0.50 dollars per resolution; 5 dollars per 1,000 sessions.
Now change one condition at a time.
| Scenario | Sessions | Conversations | AI-resolved | Tickets | Seats |
|---|---|---|---|---|---|
| A. Normal month | 200,000 | 4,000 | 2,000 | 2,000 | 4 |
| B. Support surge (a shipping delay) | 200,000 | 8,000 | 4,000 | 4,000 | 4 |
| C. Traffic campaign | 400,000 | 5,000 | 2,500 | 2,500 | 4 |
| D. The AI handles more | 200,000 | 4,000 | 3,000 | 1,000 | 4 |
The monthly bill under each model, in dollars:
| Scenario | Seat | Ticket | Conversation | Resolution | Session |
|---|---|---|---|---|---|
| A. Normal month | 1,000 | 1,000 | 1,000 | 1,000 | 1,000 |
| B. Support surge | 1,000 | 2,000 | 2,000 | 2,000 | 1,000 |
| C. Traffic campaign | 1,000 | 1,250 | 1,250 | 1,250 | 2,000 |
| D. The AI handles more | 1,000 | 500 | 1,000 | 1,500 | 1,000 |
What the table shows:
- In B, the usage units double and the seat and session bills do not. The seat bill hides the cost of the surge: your four agents carry twice the tickets.
- In C, the session bill jumps and the others move less. Traffic doubled while shopper contact rose by a quarter, so a bill that follows traffic moved more than one that follows contact.
- In D, the resolution bill rises by half while the ticket bill halves. The AI is doing more, and a per-resolution fee grows with that. The right check is whether each resolved conversation costs less than a person handling the same question would have, not whether the invoice went up.
- No column is lowest in all four rows. Each unit protects you from one risk and exposes you to another.
The example also answers a common question. Per-resolution and per-conversation pricing tie in scenario A because half the conversations are resolved and the price per resolution is twice the price per conversation. The break-even rule is simple: per-resolution costs less than per-conversation when the share of conversations the AI resolves is below the price per conversation divided by the price per resolution. Here that ratio is one half. At 75% resolved, as in D, per-resolution costs more.
How to compare real quotes
- Pull three months of your own numbers. Sessions, chat conversations, tickets, conversations the AI resolved if you have that figure, and seats. Include your busiest month.
- Get the unit definition in writing. Ask each vendor for the exact wording of what is counted and a list of all other charges: add-ons, channels, onboarding, minimums and overage.
- Build the bill three ways. Your normal month, your peak month and a growth case, for example traffic up by half. Put the results in one table like the one above.
- Ask what happens at a cap or a tier limit. Does the price step up, does service stop, and is there a warning beforehand?
- Test the definition on real conversations. Sample a batch the vendor would bill and check whether you agree with how they were counted.
- Compare the bill with the work it replaces or supports. A cheaper invoice that resolves less is not cheaper. Set the cost against the cost of a person handling the same questions, using your own loaded cost.
- Re-run the model at renewal and whenever your volumes change.
Which unit fits which store
These are tendencies to test against your own numbers, not rules.
- Steady support volume and a stable team: seat or ticket tiers give a bill you can predict.
- Seasonal or spiky support: usage units flex with demand. Set a cap and model the peak month.
- High traffic and a small share of shoppers who chat: model session pricing next to conversation pricing at your real ratio of conversations to sessions.
- Uncertain volumes: a free or low starting tier with a clear cap limits the downside while you learn.
For one vendor's pricing units worked through in detail, see Gorgias Pricing 2026: Plans, AI Agent Per-Resolution Costs, and Total Cost at Your Volume.
FAQ
How much does an AI chatbot cost for a Shopify store? It depends on the pricing unit and your volumes, so this guide does not quote a number. Start with your own sessions, conversations, tickets and seats for a normal and a peak month, then ask each vendor for a written quote, compare it with its pricing page, and compare the resulting bills. Some products offer a free starting tier.
Is per-resolution pricing cheaper than per-conversation pricing? It depends on two things: the two rates, and the share of conversations the AI resolves. Per-resolution costs less when that share is below the price per conversation divided by the price per resolution, and more when it is above. The worked example shows both cases.
What counts as a resolution in AI agent pricing? There is no common definition, so ask. Check whether silence counts as resolved, whether there is a reopen window, whether a handoff to your team counts, and whether you can read the conversations you were billed for.
Session-based or conversation-based pricing: which fits my store? Neither fits all stores. Session pricing follows your traffic, not ticket volume. Conversation pricing follows how many shoppers contact you. Compare them at your own ratio of conversations to sessions, in a normal month and a peak month.
Do I pay for seats as well as AI? Sometimes. Some products charge for seats and bill the AI separately by usage, and others price the AI alone. Ask for a quote that lists both.
More from REP AI
Shoppers Arriving From ChatGPT, Gemini and Perplexity: How to Convert Them Once They Land
A shopper who arrives from an AI assistant may already have asked a question and seen a short answer, so they may land with a specific expectation and have done part of the research. To convert them: find them in your analytics, make the landing page confirm what the assistant likely told them...
Proactive Chat Messages That Don't Annoy Shoppers: 25 Triggers and Examples for Ecommerce
A good proactive chat message is a short, specific question tied to the page the shopper is on, with one easy next step and a clean exit. Trigger it from a behavior that suggests a question, such as comparing two products or opening a size guide, and not from elapsed time alone. Show it once,...
AI Shopping Assistants for Beauty and Skincare Brands: From Skin Type to Routine
An AI shopping assistant can help customers choose skincare and beauty products by turning what the shopper says about themselves, such as a skin type they describe, a finish they like, a shade they wear or products they already use, into a short list from your own catalog. It should show each...
Want to see what your store would do with Rep?
Run your real Shopify catalog through the simulator before you commit to anything. No sales call required.
