How AI Vendors Price Their Products — and How to Compare Them

Every vendor prices differently on purpose

Two AI vendors solving the same problem will quote in units that can't be compared side by side — one per seat, one per resolution, one per thousand API requests. That's not an accident; it makes shopping harder. An AI vendor pricing comparison starts by normalising every quote to the same thing: cost per unit of work, at your volume, including peak.

The common pricing models

• Per seat. Flat per agent or per user. Predictable, but you pay for capacity, not usage — and it doesn't scale down when volume dips.
• Per resolution / per outcome. You pay when the AI resolves a ticket, closes a case, completes a call. Aligns cost with value, but the definition of "resolved" is the vendor's, and it climbs steeply in peak season.
• Per token or per API call. Common for model APIs and infra-style products. Scales smoothly with usage; a multi-step task is several calls, so cost the task.
• Per request plus per record. Search platforms often bill on both query volume and catalogue size. Two axes that both grow with the business.
• Tiered AI features. A base tier plus a higher-priced "AI" or "advanced relevance" tier — sometimes several times the base rate for the capability you actually came for. On our semantic-search engagement the managed platform's AI-relevance tier was a multiple of standard search pricing, which is part of why a custom stack won on total cost.

The clauses that move the real number

• Minimums. A monthly floor you pay regardless of usage.
• Overage rates. What a unit costs past your committed volume — often much more than the committed rate, and where a busy month hurts.
• Peak behaviour. Get the quote at your Black Friday or attack-wave volume, not your average.
• Ramp and lock-in. A low year-one rate that steps up, and an exit that costs you your data, prompts, or a fine-tuned model.

How to actually compare

1. Pick your unit of work — a resolved conversation, a scored transaction, a search, a monitored store.
2. Estimate units per month, and units in a peak month.
3. Get each vendor to quote against those exact numbers, including minimums and overage.
4. Add the run cost that's yours either way — integration, the escalation staffing that never goes to zero, monitoring.
5. Compare that total to the amortised cost of building the equivalent.

A worked normalisation

Vendor A quotes a flat rate for eight seats. Vendor B quotes per resolved conversation. You handle 12,000 conversations a month, 3,000 of them in a peak month. Convert both to cost per resolved conversation: divide Vendor A's monthly fee by 12,000, and take Vendor B's per-resolution rate — but check Vendor B's peak-month bill at 3,000, including any overage past the committed volume. Add the escalation staffing you'll pay either way. Then compare that total to the amortised cost of building the equivalent. The pipeline write-ups in our portfolio show builds scoped against exactly this comparison — the vendor quote at real and peak volume, lined up against build plus run.

Where this stops being right

• Low volume — the differences are small and the minimums dominate; pick on fit, not price.
• A capability only one vendor has — there's nothing to normalise against; the comparison is buy-or-build.
• Early-stage, unknown volume — buy the flexible model (per-usage) and revisit once you have real numbers.

FAQ

Which pricing model is best? The one whose unit matches your value and scales the way your business does. Per-usage is usually safest early; per-seat can win at stable high utilisation.

What's the number vendors hide? The overage rate and the peak-month bill. The committed rate looks fine until a busy month pushes you past it.

How do I compare a per-seat quote to a per-resolution one? Convert both to cost per resolved unit at your monthly and peak volume, and add the run cost that's yours either way — integration, monitoring, and the escalation staffing that never reaches zero. Until every quote is in the same unit with the same fixed costs added, you're comparing marketing, not price.

ISTRALLEN helps normalise AI vendor quotes to cost per unit of work at real volume, and compares that to building — see what we do.