Pay Per Sale AI Chatbot: Which Platforms Let You Pay Only When It Sells? (2026)

What pay per sale actually means

The phrase gets stretched, so it helps to pin it down. True pay per sale pricing has three properties:

1. The fee is calculated from revenue the chatbot produced, not from messages sent, tickets closed, or leads collected.
2. There is a defined attribution rule you can read, so you know exactly which orders count.
3. You can audit every charged order yourself, line by line.

Plenty of vendors advertise results. Very few will invoice on them. When a pricing page says "performance based" it usually means per resolution billing, which measures closed support conversations. That is a real improvement over flat subscriptions, and we compared all four industry models in our guide to performance based chatbot pricing. But a resolution is a ticket, not a sale. If your goal is revenue, per resolution pricing bills you for an outcome one step removed from the one you want.

Who offers what in 2026

Prices below were checked against each vendor's official pricing page or announcement in September 2026. The column that matters is the trigger: what event actually generates a charge.

The honest reading of this table: there is no pure commission chatbot on the market. Even SparkGPT keeps a $49 base fee, which covers hosting and unlimited conversations. What separates the last row is that the variable part of the bill, the part that can grow, only grows when verified orders come in. Everywhere else the variable part grows with usage, whether or not that usage sold anything.

Intercom deserves credit here for moving the industry toward outcome billing at all, and Fin is a strong product with real market share. Its own price list is also revealing: a qualified sales lead costs $10 while a support resolution costs $0.99. The biggest vendor in the category has priced a sales outcome at roughly ten times a support outcome. The step it declined to take is charging on the sale itself. In the blog post announcing Fin for Sales pricing, Intercom wrote that it considered a share of closing revenue and rejected it because attribution across a human sales team is hard and would demand deep CRM integration. For their market, that is fair reasoning. For an online store it does not apply, because there is no sales team between the chat and the checkout. The full story is in our performance based pricing guide.

How SparkGPT's pay per sale model works

The mechanics are short enough to state completely. When a visitor talks to your agent, that session gets tagged. If the same visitor places an order within 24 hours, the order counts as attributed and 5% of it is billed. If they return three days later and buy, that order is free, even though the conversation probably helped. Every attributed order appears individually in the ROI dashboard, so you can check each one against your own order history and dispute the ones that look wrong.

The rest of the plan removes the usual meters. Conversations are unlimited on the $49 Growth plan, so a long conversation that patiently walks one undecided shopper to checkout costs nothing extra unless that shopper buys. Billing runs on prepaid credits where one credit equals one dollar, and when the balance runs out the agent pauses rather than charging you again without asking. There is also a free plan for building and testing an agent against your real website, with no credit card and no expiry.

The one sentence version: you only pay when SparkGPT makes you a sale.

Does the bill become a growth trap?

One reasonable objection to outcome pricing, raised in SiteGPT's own pricing comparison, is that it creates a growth trap: the better the AI performs, the bigger your bill. For per resolution billing the point has teeth, because resolutions scale with traffic and ticket volume, not with revenue. A busy month full of "where is my order" questions costs you real money and earns you nothing.

Pay per sale pricing does not have this shape. The fee is a fixed fraction of new, verified revenue, so the bill cannot grow faster than the sales that justify it. A $3,000 attributed month costs $150 on top of the base fee, and you keep the other $2,850 from orders you can individually verify. The comparison every store owner already understands is marketplace economics: delivery platforms and marketplaces routinely take 15% to 30% of each order for delivering the customer. Five percent on a customer already standing in your own store is the same trade at a fraction of the rate, and only on orders the dashboard can show you.

Five questions to ask before signing any pay per sale agreement

Attribution is where a good idea can hide bad math, so inspect it like a contract.

1. What is the attribution window? A 24 hour window is strict and defensible. A 30 day window would let the vendor claim orders the chatbot barely touched. Shorter is more honest, even though it undercounts the vendor's real influence.
2. Is attribution per session or per person forever? The charge should follow a specific tagged conversation, not permanently mark a customer so every future order pays commission.
3. Can you see every charged order? If there is no per order view, you are trusting an invoice total you cannot check. Refuse that.
4. What does a zero month cost? On SparkGPT that answer is $49. Get the equivalent number in writing from any vendor, including what happens to unused prepaid balance.
5. Does the base fee cap conversations? A pay per sale model loses its point if a message meter underneath it punishes long selling conversations. Unlimited conversations should come standard.

Any vendor with an honest model answers all five quickly. Hesitation on question three is the loudest warning.

Which model fits your store

For a store under roughly $5,000 per month in revenue, flat subscriptions are hard to justify at all: $40 to $150 per month is a meaningful fixed cost against thin margins, and none of it is tied to results. A pay per sale model with a modest base fee means a worst case of $49 and an upside that pays for itself by definition.

For a store in the $5,000 to $50,000 range, the calculation is about verification rather than size. A subscription that happens to coincide with growing sales looks fine until you ask which orders it caused, and no subscription vendor can answer that. An attributed model answers it by default, which also tells you when to switch the agent off if it stops earning.

For support heavy businesses that mostly need ticket deflection rather than sales, per resolution pricing such as Intercom Fin remains the better structural fit, and it would be dishonest to claim otherwise. Pay per sale pricing is built for stores where the conversation and the purchase happen in the same place.

Frequently asked questions

Is there an AI chatbot that only charges when it makes a sale?

Not in pure form. As of September 2026 no major platform runs on commission alone. SparkGPT is the closest available structure: a $49 per month base fee plus 5% of orders placed within 24 hours of a conversation with the agent, with each attributed order listed in an auditable dashboard.

How is pay per sale different from per resolution pricing?

Per resolution billing, such as Intercom Fin at $0.99 per resolution, charges when the AI closes a support conversation. Pay per sale billing charges only when an attributed order is placed. A resolution can happen a thousand times without producing revenue, so the two models measure different things.

What attribution window should a pay per sale chatbot use?

Short windows are more trustworthy. SparkGPT uses 24 hours: an order counts only if the customer buys within a day of talking to the agent. A vendor claiming credit for purchases weeks after a chat is charging for influence it cannot demonstrate.

Does 5% of sales end up more expensive than a subscription?

In strong months, yes, and that is the point. A month with $3,000 in attributed orders costs $199 total on SparkGPT versus $150 on a typical mid tier subscription, but the $199 is backed by an order list you can verify. In a month where the agent sells nothing, the comparison flips: $49 versus the same $150.

Can I test a pay per sale chatbot before paying anything?

SparkGPT's free plan lets you build an agent from your website URL and test it with unlimited conversations, no credit card required. The 5% model only starts when you deploy the agent live on the Growth plan.

What should I ask a vendor before agreeing to revenue share pricing?

Five things: the exact attribution window, whether attribution is per session or permanent, whether you can audit every charged order, what a month with zero sales costs, and whether conversations are capped underneath the revenue fee.

See the model against your own store

The fastest way to judge pay per sale pricing is to watch it price your own traffic. Build a SparkGPT agent for free: paste your URL, let it learn your catalog and policies, and test it as long as you want without a credit card. Go live when you want the ROI dashboard doing the arguing for you. Start at https://www.sparkgpt.ai.