How much does an AI customer service bot cost?

The AI support pricing pitch usually starts with a clean comparison.

An AI resolution costs around $0.99 to $1.50. A human ticket costs $8 to $12. The gap looks obvious.

But list price is not the same thing as cost per resolved customer problem. Once you include failed bot touches, integration work, knowledge-base upkeep, QA, and the cost of getting sensitive tickets wrong, the picture gets much less tidy.

This is how the math actually works.

The three pricing models you’ll see

Vendors price AI customer service in three main ways. Each one has a different incentive hiding inside it.

Per resolution. Intercom Fin lists $0.99 per resolved conversation. You pay only when the AI closed the ticket end-to-end. Sounds clean. The catch sits in the definition of “resolution,” which the vendor sets.

Per ticket touched. Gorgias AI and similar tools charge on tickets the AI handled, resolved or not. Lower per-unit cost, but every refund the bot tried and failed at still gets billed.

Per seat. Kustomer Copilot, Forethought, Front, and most of the agent-assist tools. Flat monthly rate per support rep using the AI. Decouples cost from volume, which is good when volume is uncertain and bad when volume is high.

The three models suit different shapes of work. Per-resolution makes sense when most of your tickets are simple lookups. Per-seat makes sense for agent-side AI where the AI is not trying to close tickets at all. Per-ticket is the trickiest, because the price is incurred whether or not the customer’s problem was solved.

What the list price doesn’t include

The list price is one line on the invoice. The actual cost of running AI in customer service has several more.

Integration cost. Connecting the AI to Shopify, your helpdesk, your returns platform, your CRM. Some vendors include it; most charge implementation fees that scale with stack complexity, ranging from low thousands for off-the-shelf integrations to tens of thousands for custom work.

Knowledge-base curation. AI is only as good as the policy and FAQ content it draws on. Mid-market deployments typically need a significant block of internal KB cleanup before the bot is reliable, and the KB drifts as the business changes. Budget recurring time, not just upfront.

Quality drift over time. Models change, intents change, your catalogue changes. Resolution rates measured in month one are rarely the same as month twelve. Without active QA, drift erodes the savings you booked.

Hidden cost on low-resolution categories. This is the one most pricing pages do not acknowledge. A bot priced at $1 per touch can be more expensive than a $10 human on categories the bot resolves at 8%, because eleven failed touches still get billed before the human takes the ticket cold.

Gartner now predicts that GenAI cost per resolution for customer service will exceed offshore human agent cost by 2030. The mechanism: inference cost flattens, while QA, integration upkeep, and per-resolution recovery on hard tickets grow. The bot-is-cheaper assumption holds today; the trend is closing the gap.

The right way to think about cost

Two questions make the math more useful.

What’s the cost per resolved ticket, by category? Not per touch, not blended across the inbox. Take the AI’s resolution rate on each ticket type, divide by its per-touch price, and compare to the same math on your human team. The picture changes per category.

On order status, the bot’s per-resolved-ticket cost is probably $1.50 to $2. The human’s is $8 to $12. AI wins.

On refunds at 8% resolution, the bot’s per-resolved-ticket cost is closer to $12: one touch billed every time the customer escalates, roughly twelve touches to actually close one. The human’s is still $8 to $12. The bot loses.

Across the inbox, the per-category ceilings determine the math. A blended cost-per-touch number hides the trade-off.

What’s the cost of getting a category wrong? The categories where AI doesn’t perform are also the categories with the highest customer-experience cost when the AI tries anyway. A failed refund ticket isn’t just an extra billed touch. It’s a Trustpilot review, a repeat contact, a churned customer, and sometimes a refund issued anyway out of frustration. Those costs don’t show up on the AI invoice. They show up on the next quarter’s churn report.

What this means in practice for a DTC brand

A rough sketch at three revenue brackets.

$5M brand, ~5,000 tickets/month. Per-resolution AI at $1 list, on the categories where it works (say 40% of inbox), runs around $2,000/month. The remaining 60% goes to your team. Per-seat agent assist for two reps on top: $200-$400/month. Total monthly AI spend: $2,200-$2,400.

$20M brand, ~20,000 tickets/month. Same math at four times the scale. Most $20M brands running deflection AI at list-price-equivalent land in the $6,000-$12,000/month range, with $1,500-$3,000/month of team-side spend on top. Total: roughly $9,000-$15,000/month, with annualised value-per-ticket-resolved that holds up against the all-human baseline.

$50M+ brand, 50,000+ tickets/month. Volume tips the calculation toward outcome-based or hybrid pricing. Real numbers depend on negotiated rates, but most $50M+ DTC brands running both deflection AI and agent assist land in the $25K-$60K/month total range, with payback measured against per-ticket-resolved cost, not list price.

These are sketches, not benchmarks. The point is simple: list price is the starting line, not the finish line. Per-resolved-ticket cost, by category, is the real number.

Closing

Cheapest per touch is rarely cheapest per resolution. The vendor that quotes you $0.99 is doing the math the customer never experiences. The vendor that breaks cost down by resolved ticket, by category, is doing the math the customer pays you for.

Sources

• Intercom Fin pricing. $0.99 per resolution list price.
• Gartner, January 2026. Prediction: GenAI cost-per-resolution for customer service will exceed offshore human agent cost by 2030.
• What’s the difference between deflection rate and resolution rate?. Why per-touch ≠ per-resolution.
• What’s the right deflection KPI?. Per-category ceilings driving the cost math.
• AI in customer service is two products. Per-seat vs per-resolution maps cleanly to agent assist vs deflection AI.