AI Customer Support Pricing Models in 2026: What to Pay For

The only pricing number that matters for AI customer support is cost per autonomous resolution. Formula:

Cost per autonomous resolution = (total monthly cost) ÷ (resolutions per month)

Run this on every vendor before you sign anything. It's the only way to compare apples to apples when one vendor charges per ticket, another per resolution, a third per seat plus AI usage. The headline price is noise; this number is signal.

Here's why it matters in one example. Vendor A: $400/month flat, resolves 800 tickets/month autonomously. Vendor B: $200/month flat, resolves 200 tickets. Vendor B looks half the price, and is twice as expensive per actual outcome ($1.00 per resolution vs $0.50). Sticker price will lie to you. The formula won't.

The rest of this post is what changes that calculation: which pricing model fits which kind of buyer, where the hidden costs hide, and how to spot the marketing patterns that disguise real prices.

The three pricing models in 2026

Per-ticket pricing

You pay for every ticket the AI sees, regardless of whether it resolves it.

Pros: predictable monthly cost (volume × rate), easy to budget for steady-state operations, vendor has a small incentive to lower per-ticket cost over time.

Cons: you pay even when the AI fails, bad alignment. Spike pricing on Black Friday or holiday surges hurts: volume goes up, your bill explodes. Encourages the vendor to claim "we saw the ticket" as victory, even if the customer escalated immediately.

Typical price: $0.25–0.75 per ticket seen.

When it makes sense: highly predictable steady-state volume with minimal seasonality, where you want budget predictability over performance alignment.

Per-resolution pricing

You pay only when the AI fully resolves a ticket without escalation.

Pros: aligned incentives, the vendor gets paid only when their tool works. This forces them to be honest about resolution rates because resolution rate is literally their revenue. Spike-resilient, volume up means resolutions up means proportional cost.

Cons: less predictable monthly cost in early months as your resolution rate stabilizes. Definitions of "resolution" vary (see below). Can incentivize vendors to be aggressive about claiming a resolution, watch CSAT to catch this.

Typical price: $0.50–1.50 per resolved ticket.

When it makes sense: most teams in 2026. This is the default for AI-native platforms and increasingly the standard.

Per-seat (legacy)

You pay per human agent who has access to the platform, plus AI usage on top.

Pros: familiar, the model help-desks have used for a decade. Predictable as long as your team size doesn't change.

Cons: makes less and less sense as AI absorbs volume, your seat count drops while your throughput grows. Vendors using this model are usually bolting AI onto an old pricing structure rather than redesigning. Often paired with separate AI fees that are themselves per-resolution or per-ticket, double-billing.

Typical price: $50–150 per agent per month, plus AI fees on top.

When it makes sense: large enterprise teams (50+ agents) where seat-based contracts are organizationally easier to budget for. Outside of that, find a per-resolution alternative.

Worked example: how to compare three vendor proposals

Same business: 5,000 tickets/month, expected 65% autonomous resolution rate (~3,250 resolutions). Three anonymized proposals on the table:

Vendor A is cheapest on cost-per-resolution at this volume but punishes you in spikes, a Black Friday week with 30% volume bump pushes you to $2,600 even if the AI doesn't resolve any more tickets. Vendor B is the most expensive but the cleanest price model. Vendor C is in the middle but has the per-seat tax that gets worse as you grow your AI side and shrink your human side.

The right answer depends on your volume profile and growth trajectory, not on which has the lowest sticker price.

Define "resolution" before you sign

The biggest gotcha in per-resolution pricing is the definition of "resolved." Vendors will define it differently to favor their billing:

• Strict: AI sent a reply, the customer didn't reply back within X hours. Suggests the customer was satisfied.
• Lenient: AI sent a reply, regardless of what happened next.
• Action-based: AI took an action (refund, label, etc.), counts as resolution even if the customer was confused by the action.
• CSAT-gated: AI's reply got a positive CSAT response, counts only if rated.

The right definition for most teams is: AI sent a reply or took an action, the customer didn't escalate or reopen within 48 hours, and CSAT (if collected) was 3+. This is honest. It doesn't credit the AI for sending a reply that left the customer hanging. It doesn't punish the AI for late replies the customer never came back to. Demand this definition in writing in your contract, not the vendor's.

"Unlimited" pricing: three patterns that mean limited

A few vendors market "unlimited" tiers, flat fee, all the resolutions you want. There's always a real cap somewhere. Three patterns to recognize:

Pattern 1: "Unlimited within fair use." Fair use is whatever the vendor says it is after you sign. The pattern in publicly reported B2B SaaS contracts is that "fair use" triggers renegotiation when a customer hits roughly 3x the vendor's average usage, a number rarely disclosed upfront. Ask for the specific volume threshold in writing or assume it's the lowest plausible number.

Pattern 2: "Unlimited tickets, $X per AI action." Unlimited everything except the part that delivers value. Tickets are free; refunds, label generation, subscription edits, address updates each cost extra. The "action" is the work; the ticket is just the wrapper.

Pattern 3: "Unlimited up to 10x industry average." The vendor sets industry average. As AI volume grows industry-wide, the threshold drifts to keep ~10% of customers paying overage. Useful for marketing, useless as a real ceiling.

If a vendor offers "unlimited," ask them in writing: "What is the maximum monthly resolution count beyond which you reserve the right to renegotiate or charge overage?" Their answer is the actual cap.

Hidden costs to ask about

Six items to ask vendors about explicitly before signing, most are missing from quote sheets:

1. Minimum monthly commitment. Some platforms require a $X/month floor regardless of usage. If your volume drops, you still pay. Ask: "What's the minimum monthly fee?"

2. Overage fees. If you exceed your committed tier, what's the per-unit rate? Often 2–3x the rate inside your tier, punitive. Ask: "What's the overage rate, and is there a cap?"

3. Channel fees. Some platforms charge separately for email, chat, SMS, social. The headline number is one channel; the real cost is several. Ask: "Does the price cover all the channels we need?"

4. Integration fees. Connecting Shopify or Stripe is usually free (these are widely supported). Connecting your custom backend? "Custom integration starting at $5,000." Ask: "What integrations are included? What's the cost for ones that aren't?"

5. Professional services. Some vendors won't let you self-serve setup, or charge an "implementation fee" of $5–25K. Ask: "Is implementation self-serve or is there a mandatory setup fee?"

6. Human seat fees on AI tiers. Some "AI-first" platforms still charge for human agent access on top of AI fees. Watch this. Ask: "Does the price include unlimited human agents, or are there per-seat fees?"

What to negotiate

Three levers consistently get you 10–25% off list pricing:

Multi-year commitment. A 12–24 month contract is usually worth 10–20% discount. Ask. Don't accept the standard month-to-month rate without checking.

Volume commitment. If your volume is predictable, commit to a floor in exchange for a better unit rate. Just make sure the contract has a sane overage clause, you don't want to commit to 10K resolutions/month and discover the overage rate is $3 each.

Competitive quote. Get written quotes from 2–3 vendors. Show them to each other (carefully, don't reveal proprietary terms). The presence of an alternative quote is the single most powerful negotiating lever in B2B SaaS.

A typical mid-market deal: list price $1.50 per resolution, negotiated to $1.10 with annual commitment plus competitive pressure. That's about $480 in annualized savings per 1,000 monthly resolutions.

A decision tree

Use this to pick a model:

1. Is your monthly volume highly predictable (±15%)?

   • Yes → per-ticket might work for budget predictability.
   • No → per-resolution; spike-resilient.

2. Are you scaling fast (>30% YoY ticket growth)?

   • Yes → per-resolution; per-ticket will punish you.
   • No → either model works on volume.

3. Do you have a large existing human team (20+) you're not reducing?

   • Yes → per-seat with AI add-on might fit your org structure.
   • No → per-resolution; per-seat is uneconomic at small scale.

4. Is implementation simplicity a priority?

   • Yes → AI-native platforms with per-resolution; faster to live, simpler contracts.
   • No → doesn't matter; pick on economics.

For most ecommerce brands in 2026, the answer is per-resolution from an AI-native platform. The exceptions are the ones above.

What to do next

Two practical steps:

1. Compute current cost per resolved ticket. Even if you're all-human now: total monthly support cost ÷ tickets resolved. This is your benchmark. Most teams find this number is $4–10/ticket.
2. Get quotes from 2–3 vendors using your real volume. Make them quote in their format, then convert each to cost-per-resolution using the formula above. Compare honestly.

Got a competitor quote in hand? Send it our way. We'll convert all the apples-to-oranges parts (per-ticket vs per-resolution vs per-seat) into a single cost-per-outcome number and tell you honestly whether ours actually wins. If it doesn't, we'll say so.

Sources

• Forrester, Predictions 2026: AI Gets Real For Customer Service, But It's Not Glamorous Work, context on enterprise AI deployment economics in 2026.
• Klarna, AI assistant handles two-thirds of customer service chats in its first month, public per-resolution outcomes from Klarna's deployment ($40M profit improvement, 2.3M conversations in month one).