You're Penalized for Your AI Getting Better

Per-resolution pricing is the most popular billing model for AI support tools in 2026. The pitch is simple and sounds fair: you only pay when the AI actually resolves a customer's issue. No resolution, no charge. What's not to like?

Here's what's not to like: the model means your cost goes up every time the AI improves.

When you first deploy an AI support tool, it resolves maybe 30-40% of incoming conversations. You're paying for those resolved ones, and humans handle the rest. Fine. Then you tune it. You improve your knowledge base. You add more training data. The AI starts resolving 50%, then 60%, then 70%.

Your per-resolution bill just doubled.

This is the only pricing model in software where success is directly punished. Imagine if your email provider charged you more when your delivery rate improved. Or your database charged per successful query. You'd call that absurd. But in AI support, it's the default.

The Math That Should Scare You

Let's make this concrete. A team gets 5,000 support conversations per month.

Month one with Intercom Fin at $0.99/resolution. AI resolves 30% (1,500 conversations). Cost: $1,485. Humans handle the other 3,500. Seems reasonable.

Month six. You've improved your help center, added custom answers, trained the AI on your specific use cases. AI now resolves 50% (2,500 conversations). Cost: $2,475. You've reduced human workload by 1,000 conversations. But your AI bill went up $990.

Month twelve. AI is crushing it at 70% resolution rate (3,500 conversations). Cost: $3,465. Your AI bill has increased 133% since launch. You're saving significant human time, but the savings are being eaten by the AI vendor.

Now add Intercom's seat fees. Even on the Essential plan at $29/seat/month (annual billing), a team of 5 agents pays $145/month. Total at 70% resolution: $3,610/month.

Same scenario with Supp at $0.30/resolution. Month one: 1,500 resolutions = $450. Month six: 2,500 resolutions = $750. Month twelve: 3,500 resolutions = $1,050. No seat fees. Total at 70% resolution: $1,050/month.

The difference at 70% automation: $2,560/month. That's $30,720/year.

Why Vendors Love This Model

Per-resolution pricing aligns the vendor's revenue with your AI's improvement. The better their product works for you, the more you pay them. It's brilliant from a business model perspective.

Intercom has a direct financial incentive to make Fin resolve more conversations. Every improvement to their AI engine generates more revenue per customer. They don't need to raise prices. They just need to make the product better.

This is a fundamentally different dynamic than traditional SaaS pricing. With seat-based pricing, the vendor's revenue grows when you hire more agents. With per-resolution pricing, revenue grows when you need fewer agents. The vendor profits from the same outcome that's supposed to save you money.

The Volume Trap

Small teams don't feel this. If you get 200 conversations a month and the AI resolves 70% of them, you're paying $138.60/month on Intercom Fin. That's fine. The pricing model works at low volume because the absolute numbers are small.

The trap springs at scale. A company handling 20,000 conversations per month at 70% resolution pays $13,860/month for Intercom Fin resolutions alone. At Supp's $0.30 rate, that's $4,200. The gap is $9,660 per month, or $115,920 per year.

At that scale, the per-resolution cost is your largest support expense after salaries. And unlike salaries, it goes up automatically every time your AI gets better.

What's the Alternative?

There are four pricing models in AI support right now.

Per-resolution (Intercom, Zendesk, Gorgias): you pay per solved conversation. Cost scales with AI quality.

Per-interaction/classification (Supp): you pay per processed message, whether it's classified, routed, or resolved with an action. Cost scales with volume, not AI quality. A classification costs $0.20. A resolution with an action costs $0.30.

Flat monthly fee (some smaller tools): fixed price regardless of volume. Great until you outgrow the tier and jump to the next one.

Enterprise custom (Ada, Forethought before acquisition): call sales, negotiate a contract, lock in for a year. Predictable but inflexible and usually expensive.

The per-interaction model has a structural advantage for growing teams. Your cost is tied to how many customers contact you, not how well the AI performs. If the AI improves from 40% to 70% resolution, your total cost barely changes because you're paying per classification either way. The savings from reduced human handling go to you, not the AI vendor.

How to Evaluate This for Your Team

Pull your last three months of support data. Count total conversations, not just the ones you think AI could handle. Multiply by the resolution rate the vendor promises (cut their number by 10-15%, they always overstate it). Multiply by the per-resolution price.

Now project forward. If AI resolution improves by 10 percentage points every quarter (a realistic trajectory for the first year), what does month 12 look like?

If the number makes you uncomfortable, the pricing model is wrong for your volume. A $0.30 per-resolution or $0.20 per-classification model scales linearly with your support volume. A $0.99 per-resolution model scales with volume AND AI quality, which means it compounds.

Success shouldn't be the most expensive part of your support stack.