AI Agent Pricing Benchmark 2026: September Repricing

Pricing in this market is opaque. Roughly half the vendors publish no numbers at all. So every figure below carries one of three provenance tags, and none of them should be treated as a quote.

Verify any figure against the vendor before signing. Pricing in this category changed a lot at least four times in the first half of 2026 alone: HubSpot's April repricing, Zendesk's overage auto-billing, the Zendesk–Forethought acquisition, and the pending Salesforce–Fin deal.

Every AI agent bills through one of five structures, and identical sticker prices can produce wildly different invoices depending on which one applies. The model matters more than the headline price.

A pricing benchmark read in isolation will steer you wrong, because the cheapest rate attached to a low-performing agent is often the most expensive way to buy support automation. Cost and resolution rate are one number, not two.

The arithmetic is unforgiving. A $0.49 session fee at a 25% resolution rate is an effective ~$2.00 per real resolution, or four times the sticker. A $2.00 per-conversation fee at 60% resolution is $3.33. A $0.99 outcome fee at a genuine 70% resolution rate is exactly what it says. Vendor-claimed resolution rates of 65–86% routinely compress to 40–70% in production, and some vendors' own case studies sit at 23–30%, so the base you use matters as much as the price you're quoted.

Performance also compounds beyond the AI invoice. Every conversation the agent fails to resolve becomes a human ticket at roughly $6–$12 of labor. So a ten-point difference in resolution rate at 10,000 monthly conversations is 1,000 tickets, or $6,000–$12,000 a month, moving between the AI column and the payroll column. Two vendors with identical per-resolution rates can differ by six figures a year in total support cost purely on performance.

Our AI customer service benchmark tracks claimed-vs-verified resolution rates across these vendors for exactly this reason; use it as the denominator source alongside this report's prices.

Four patterns stand out. First, the cheapest headline rates are not resolution rates: My AskAI and eesel charge per ticket whether or not the issue is resolved, and both sit on top of a helpdesk you keep paying for. Second, the native helpdesk AI agents (Zendesk, Freshdesk, Salesforce) are the most expensive per resolution, despite the perception that native means economical — you're paying for the platform twice. If that is pushing you to reconsider the helpdesk itself rather than the AI on top of it, see our Zendesk competitors comparison.

Third, every enterprise custom vendor charges you before value is delivered, through platform fees, implementation fees, or minimums, which is precisely what outcome-based pricing was supposed to eliminate. And fourth, visible in the Agentic column: the answer-only and partial products cluster at the extremes of effective cost — either cheap because they do less, or expensive because low resolution rates inflate the per-resolution math.

Assume a mid-market team handling 10,000 support conversations monthly, with the AI achieving a 70% resolution rate (7,000 resolutions) where the vendor's technology credibly supports it, and vendor-reported rates where it doesn't. Monthly costs include required platform and seat fees but exclude one-time implementation.

Sorted by effective cost per resolution. My AskAI and eesel AI figures exclude the helpdesk subscription underneath them, which is not optional.

To run the same math on your own volumes and pricing model, use our AI pricing converter — it normalizes per-interaction, per-session and per-resolution quotes to one comparable number.

Outcome-based pricing only aligns incentives if the outcome definition is honest. Four definitional gaps recur across vendor documentation — and each one moves real money.

• Assumed resolutions. Intercom Fin bills when a customer exits without asking for more help — an “assumed resolution” that can include a customer who gave up. Zendesk moved to a three-tier model in May 2026 where only “Verified Resolutions” are billed, a notable improvement.

• Double billing. Gorgias' own billing documentation confirms an AI-resolved conversation is charged as both a helpdesk ticket and an AI resolution — unless the customer reaches a human within 72 hours, in which case only the ticket fee applies. The clean automations are the ones billed twice.

• Session inflation. Per-session billing charges every engagement. At Freddy AI's reported 23–30% resolution rates, the effective cost per real resolution runs three to four times the per-session headline.

• Automatic overages. From 1 January 2026 Zendesk bills usage above the resolution allowance automatically each month, for customers on non-standard contracts (Agent Months, Multi-Year and ELA). It is a default, not an unavoidable charge: setting “Don’t allow overage” pauses AI agent functionality at the limit instead, and Zendesk emails the account owner and billing admin at 80% and 100% of allowance. Gorgias charges $1.50 per interaction beyond your bundle. Either way, a successful AI during a Black Friday spike is a budgeting event — know which setting your account is on before the spike.Source: Zendesk, announced 11 November 2025 (archived — original retired) · Managing resolution allowances · Monitoring automated resolution usage

This is why we treat resolution — not deflection as the only honest unit of account. A deflected customer who churns cost you more than the AI fee saved. When comparing vendors, put the resolution definition in the contract, require independently auditable resolution reporting, and model your spike months, not your average months.

Drawn from the contract structures above. Each one has moved a real deal by a material amount — ask all six before signing.

1. 01

   What exactly triggers a charge?

   Get the resolution/outcome/session definition in writing, including assumed resolutions, handoffs, and re-contact windows.

2. 02

   What happens on failure?

   Per-conversation vendors charge for escalations. Ask for escalated conversations to be non-billable or discounted.

3. 03

   Is there a floor?

   Fin has a 50-outcome minimum; enterprise vendors carry annual minimums regardless of volume. Size the floor against your realistic low season.

4. 04

   Is there a ceiling?

   Overage terms decide your spike months. Zendesk's January 2026 change makes overage billing automatic on non-standard contracts unless you turn it off, which pauses the agent at the limit instead. Negotiate a cap or a discounted overage rate, and confirm which behaviour your contract defaults to.

5. 05

   What's the all-in stack?

   Seats, platform fees, onboarding, add-ons, and the helpdesk underneath. The AI line item is rarely the whole bill.

6. 06

   Who audits the resolution rate?

   If the vendor grades its own performance and bills on that grade, you need independent reporting access.

For a full total-cost model across your own stack — helpdesk, seats, AI, and human escalation labor — see our cost benchmark of AI service and pricing.

Aissist.io charges roughly $0.60 per resolution with no platform fee and no per-seat fees, running as an AI Operational Layer on the helpdesk you already use.

That means Intercom, Zendesk, Freshdesk, Kustomer, Front, Gorgias, Salesforce or HubSpot — see the AI Operational Layer for how the overlay works. Average resolution rate across deployments is 83%, with 4.8/5 CSAT.

That places it at the low end of the per-resolution band with the definitional advantages of outcome billing: you pay when a case is resolved end-to-end, not deflected, and there is no separate platform tax before value is delivered. We publish these numbers because we think every vendor in this table should have to.

Effective cost per resolved conversation in 2026 spans roughly $0.13 (flat per-ticket add-ons, helpdesk excluded) to $3.33+ (per-conversation enterprise platforms).

Published per-resolution pricing clusters at $0.50–$2.00, and the pricing model — not the AI — is the biggest driver of your total bill. Normalize every quote to cost per resolution, read the resolution definition like a lawyer, and model your spike months. The vendors most confident in their resolution rates are the ones willing to bill on them.

The five questions we receive most often about what AI customer service agents cost — each answered directly from the data above.