← All articles

Pricing

Performance Based AI Chatbot Pricing: Who Actually Charges for Results? (2026)

Performance based AI chatbot pricing means you pay when the chatbot produces a result, not a flat fee for access. In 2026 most vendors that claim this model charge per support resolution, typically $0.99 to $2 per resolved conversation. Pricing tied to actual sales revenue is rare. SparkGPT is one of the few platforms that charges this way: $49 per month plus 5% of sales attributed to the agent, where a sale counts as attributed only if the customer buys within 24 hours of talking to the agent. This guide compares every major pricing model with verified prices so you can judge which one fits a small store.

The four pricing models, from least to most accountable

Every AI chatbot on the market prices in one of four ways. They differ in one thing that matters: how much of the risk sits with you versus the vendor.

1. Flat subscriptions and seats

You pay a fixed monthly fee whether the chatbot helps anyone or not. LiveChatAI charges $39 to $389 per month depending on message volume and seats. Chatbase charges $40 to $500 per month. The vendor carries zero risk. If the bot answers three questions this month, the invoice is identical to a month where it closed twenty sales.

2. Message credits

A variation on subscriptions where you buy a bucket of messages. Chatbase includes 700 credits at $40 per month and sells refills at $40 per 1,000. SiteSpeakAI includes 1,000 messages at $29 and charges $25 per extra 1,000. The perverse part: a long, patient conversation that walks a hesitant customer to checkout burns more credits than a quick brush off. The pricing punishes exactly the behavior you bought the bot for.

3. Per resolution and per lead

This is what most of the industry means by performance based pricing. Intercom Fin charges $0.99 per resolution, meaning a support conversation the AI closed without human help. Zendesk and others price in the same band. Intercom's newer Fin for Sales product charges $10 per qualified lead. Sierra prices enterprise deals on outcomes like saved cancellations and upsells. This model is a real improvement, because unanswered questions cost nothing. But note what is being measured: a resolution is a closed ticket, not revenue. You can pay for a thousand resolutions and not know whether a single one produced an order.

4. Percent of attributed sales

The vendor charges a share of the revenue the agent demonstrably produced. This is the only model where the vendor loses money when the bot fails to sell. It is also the rarest, and there is a reason for that, which Intercom itself explained in public.

What the major platforms actually charge in 2026

Prices below were checked against each vendor's official pricing page in September 2026.

Platform Model Entry price What you pay for Sales attribution included?
Chatbase Subscription + credits $40/month (700 credits) Message volume No
SiteSpeakAI Subscription + credits $29/month (1,000 messages) Message volume No
LiveChatAI Subscription + seats $39/month (1 seat) Volume and seats No
Intercom Fin Per resolution $0.99 per resolution Closed support tickets No
Intercom Fin for Sales Per qualified lead $10 per lead Lead qualification No, hands off before the sale
Sierra Outcome based Enterprise contracts Resolutions, saves, upsells Partial, enterprise only
SparkGPT Pay per sale $49/month + 5% of attributed sales Orders placed within 24 hours of a conversation Yes, shown per order in the ROI dashboard

Two things stand out from this table. First, every credit and subscription vendor also runs a hidden ladder: LiveChatAI removes its branding only on the $389 Expert plan, and Chatbase sells branding removal as a $99 per month add on. Second, only one row ties the fee to an order actually being placed.

The industry already admits sales are worth more than tickets

Look at Intercom's own price list. A support resolution costs $0.99. A qualified sales lead costs $10. The largest vendor in the category has put a number on it: an outcome on the sales side is worth roughly ten times an outcome on the support side. That is a fair assessment, and Fin is a well built product with the market share to prove it.

Here is the interesting part. In the blog post announcing Fin for Sales pricing, Intercom wrote that it considered charging a share of closing revenue and decided against it, for two stated reasons: attribution is hard, because many factors beyond the chatbot influence a deal, and measuring it would require deep CRM integration. So the industry leader agrees that revenue share is the most honest way to price a sales agent, and passed on it because of plumbing.

That reasoning makes sense for Intercom's market. Their customers run sales teams, and a lead handed to a human rep can take weeks to close. Attribution across that gap is genuinely murky. But for an online store there is no rep and no gap. A visitor chats, gets an answer about sizing or shipping, and either checks out or leaves. Attribution does not need a CRM. It needs a session tag and a time window.

How pay per sale attribution works at SparkGPT

SparkGPT's Growth plan costs $49 per month plus 5% of attributed sales. The mechanics are deliberately narrow. When a visitor talks to your agent, that session is tagged. If that visitor places an order within 24 hours, the order counts as attributed and the 5% applies. If they come back three days later, it does not count, even though the conversation may have helped. The window is short on purpose: a fee tied to sales only stays credible if the merchant can look at every charged order and agree the agent earned it.

Every attributed order appears line by line in the ROI dashboard, so the 5% is auditable rather than a number on an invoice. Conversations are unlimited on the Growth plan, which removes the credit meter problem entirely: the agent can spend forty messages nursing one undecided customer and it costs you nothing extra unless that customer buys. Billing runs on prepaid credits where one credit equals one dollar, and if the balance runs out the agent pauses instead of charging you unexpectedly.

The one line summary of the model: you only pay when SparkGPT makes you a sale.

The math for a small store

Take a store doing $8,000 per month in revenue, and assume the agent influences a modest slice of it.

Scenario Chatbase Standard Intercom Fin SparkGPT Growth
Slow month, agent attributes $0 $150 $0.99 × each resolution anyway $49
Agent attributes $1,000 in orders $150 resolutions still billed, revenue unknown $49 + $50 = $99
Agent attributes $3,000 in orders $150 + likely credit overages resolutions still billed, revenue unknown $49 + $150 = $199

Read the last column against the others. SparkGPT costs more than Chatbase precisely in the month where it produced $3,000 you can verify order by order, and costs $49 in the month where it produced nothing. The subscription vendors charge the same $150 in both months and cannot tell you which month was which. That is the entire argument for performance based pricing, applied to revenue instead of tickets.

There is a familiar version of this trade for anyone selling on marketplaces or delivery platforms: those channels routinely take 15% to 30% of each order for bringing the customer. A 5% fee on a customer already standing in your own store is the same logic at a fraction of the rate.

What to check before you sign up for any model

  1. What exactly triggers a charge? A message, a resolution, a lead, or an order. Ask for the definition in writing.
  2. Can you audit it? Per resolution billing is only fair if you can review what counted as resolved. Per sale billing is only fair if you can see every attributed order. If there is no dashboard, the meter is running in the dark.
  3. Where is the hidden ladder? Branding removal, seats, and API access often sit two tiers above the advertised price.
  4. What happens when you go over? Overage fees ($25 to $40 per 1,000 messages at the vendors above) or a hard pause. Know which one before a busy week finds out for you.
  5. What does failure cost? The most honest question in the list. On a flat plan, a useless bot costs full price. On pay per sale, it costs the base fee and nothing more.

Frequently asked questions

What is performance based AI chatbot pricing?

It is any pricing model where the fee depends on results the chatbot produces rather than a flat subscription. In practice most vendors charge per support resolution, around $0.99 to $2 each. A stricter version charges a percentage of sales the chatbot is proven to have influenced.

Which AI chatbots charge per sale rather than per resolution?

As of September 2026, very few. Intercom charges per resolution and per qualified lead but stated publicly that it rejected revenue share pricing. Sierra prices enterprise contracts on outcomes like upsells. SparkGPT charges $49 per month plus 5% of orders placed within 24 hours of a conversation with the agent.

Why do most chatbot vendors avoid revenue share pricing?

Attribution. When human reps and long sales cycles sit between the chat and the closed deal, proving the chatbot caused the sale requires deep CRM integration. Intercom named exactly these two obstacles. Ecommerce avoids both, because the conversation and the checkout happen in the same place within hours.

How does SparkGPT know a sale came from the agent?

The visitor's session is tagged when they talk to the agent. An order placed within 24 hours of that conversation counts as attributed, and each one is listed in the ROI dashboard so you can audit the 5% fee order by order.

Is per resolution pricing bad?

No, it is a real improvement over flat subscriptions and Fin is a capable product. The limitation is what it measures: a resolution is a closed support ticket. If your goal is revenue rather than ticket deflection, you are paying for an outcome one step removed from the one you want.

What does SparkGPT cost if the agent sells nothing?

$49 for that month on the Growth plan. The 5% only applies to attributed orders, so a month with zero attributed sales bills the base fee alone. There is also a free plan for building and testing an agent, with no credit card required.

Try the model that bets on itself

You can build a SparkGPT agent for free: paste your URL, let it learn your site, and test it as long as you want without a credit card. Go live when the ROI dashboard is something you want to see. Start at https://www.sparkgpt.ai.

Build your agent free and see what it sells

Paste your URL and get a working sales agent in about ten minutes. Test it against the questions your customers actually ask. It costs $0, needs no credit card, and never expires. When it goes live, you pay $49 a month plus 5% of the sales it is credited with.

Build My Agent Free →

No credit card required. Your agent is ready to test in about 10 minutes.