AI Customer Service Cost Per Resolution in 2026

Why Everyone Suddenly Charges Per Answer

For fifteen years, support software was priced per seat. You paid for every person who logged in, and the software did not care how many customers they helped.

AI broke that. When a bot answers half your chats, you need fewer seats, and a vendor paid per seat watches its revenue shrink the better its own product gets. So through 2025 and 2026 most of the big names moved to outcome pricing. You now pay per answer, per resolution or per conversation, and the seat price stays on top for the humans.

That sounds fair, and partly it is. You pay when the AI does the work. But it also means your bill now grows with your traffic, and a busy month costs more whether or not your revenue did. So it is worth knowing exactly what each vendor counts, and what an answer really costs to produce.

Every Major Vendor's Price, Side By Side

Checked on each vendor's own pricing page on September 26, 2026. Prices are in US dollars and exclude the human seats most of them also require.

Two of these deserve a closer look. Agentforce's $2.00 is per conversation, so a chat where the bot gets confused and passes the customer to a human still costs the full $2.00. Tidio also bills per conversation Lyro handles. Intercom, Gorgias and Zendesk only charge when the AI actually resolves something, which is the fairer model even when the headline number is higher.

What That Means For Your Monthly Bill

Here is the AI part of the bill alone at three volumes, counting conversations the AI finishes on its own. For the two vendors that bill per conversation handled, the real figure is higher, because every chat the AI hands to a human is charged too.

At 2,000 AI answered conversations a month, which is a busy but ordinary online store, the per answer model costs somewhere between $1,800 and $4,000 a month before a single human seat. Over a year that is $21,600 to $48,000 for the AI alone.

What One AI Answer Actually Costs To Run

Vendors rarely talk about this number, so here is ours, worked out in the open.

Our AI agent runs on OpenAI's gpt-4o-mini for the writing and text-embedding-3-small for searching your help content. At OpenAI's standard list prices, that is $0.15 per million input tokens, $0.60 per million output tokens, and $0.02 per million tokens for embeddings. Each question a visitor asks goes through these steps:

Token counts are typical sizes for a small business knowledge base, rounded. A resolved chat usually takes one or two questions, so the real cost of a resolution lands between $0.002 and $0.004. The figure we plan our pricing with is $0.0032.

Put that next to the market. Intercom's $0.99 is about 300 times what one answer costs us to run, and Agentforce's $2.00 about 600 times. Vendors that run larger models pay more per answer than we do, so their own cost sits somewhere above ours. None of them publish how far above.

Is That Markup Fair?

Partly, and it is worth being straight about which part.

The model call is the cheap bit. What the big vendors also sell is everything around it: verifying that an answer really solved the problem (Zendesk uses a second AI model to check each resolution), integrations that let the bot look up orders or refunds, guardrails that stop it inventing policies, reporting, compliance paperwork, and a support team of their own. None of that is free to build.

But none of it scales with your volume the way a per answer price does. The engineering is paid once. The 2,000th answer this month costs the vendor no more to produce than the first. Per answer pricing is a business model choice, not a reflection of what each answer costs to make.

The honest summary: you are paying for software and for their margin, and that is normal. Just know that an answer like the ones you are billed for can be produced for well under a cent, as our own numbers above show, so compute alone does not explain a price near a dollar.

What Counts As A Resolution

This is where the fine print lives, and it changes your bill more than the headline price.

The window matters more than people think. If a customer comes back 3 hours later with the same problem on Zendesk messaging, that is a new conversation. If it was resolved the first time, you may have paid for a resolution that did not really resolve anything.

How To Estimate Your Own Bill

1. Count last month's chats. Your current tool reports this. Use conversations, not messages.
2. Estimate what share the AI could finish alone. For simple questions like shipping times, opening hours and returns, a well fed knowledge base handles a lot. Be conservative and start at a third.
3. Multiply by the vendor's unit. Per resolution vendors bill that number. Per conversation vendors bill every chat the AI touches, including the ones it passes on.
4. Add the seats. Intercom's Essential plan starts at $29 per seat a month on annual billing, and most of these vendors still need one seat per human.
5. Multiply by twelve and compare. A price that looks small per answer is often the biggest line in the annual budget.

Five Questions To Ask Before You Sign

• Do you charge per resolution or per conversation? The difference is every chat the AI fails.
• Who decides a conversation was resolved? A customer confirming it, a model grading it, or silence for a set time?
• Is there a monthly minimum? Intercom has one on standalone use. Others build it into tiers.
• What happens in a traffic spike? A sale, an outage or a viral post can triple a month's bill with no warning. Ask for a cap.
• Which model runs underneath? The answer tells you what each reply really costs to produce, and how much of the price is margin.

The Flat Price Option

Not every tool bills per answer. Some, including ours, include the AI agent in a flat monthly plan. TGLiveChat Pro is $19 a month with the AI agent included and no per resolution charge, because at $0.0032 an answer we can afford to include it and still make money. We are obviously not a neutral party here, so treat the tables above as the useful part and our plan as one option among several.

Flat pricing has trade offs too. You get less of the enterprise machinery: no LLM resolution auditing, fewer native integrations, no compliance team. For a large support operation that is worth paying for. For a small business answering a few hundred chats a month, paying $500 a month in resolution fees for questions about delivery times rarely is.

The Bottom Line

AI support in 2026 costs between $0.70 and $2.00 per answer at the major vendors, and a fraction of a cent to actually produce. The gap is the vendor's software, service and margin, and that is fine as long as you know it is there.

Before you pick a tool, check three things: whether you pay per resolution or per conversation, what counts as resolved, and what your bill does in your busiest month, not your average one.

Cite This Article

"In September 2026, AI support agents cost $0.70 to $2.00 per answer at the major vendors (Intercom Fin $0.99 per resolution, Gorgias $0.90, Tidio Lyro about $0.70 per conversation, Salesforce Agentforce $2.00 per conversation), while the model cost of one resolved chat on gpt-4o-mini is about $0.0032." TGLiveChat, AI Customer Service Cost Per Resolution in 2026, September 26, 2026.