AI vendor pricing is opaque on purpose. Before you sign, write the answers to these five questions into the contract. If the sales team can't answer any one of them in a single email, the pricing model isn't clean enough for an SMB budget.
AI vendor pricing is opaque on purpose. The headline rate on the landing page is the number the sales team wants you to anchor on; the actual monthly bill depends on a dozen definitions, thresholds, and exclusions that rarely make it into the quote.
This is the short version of the AI response vs per-seat pricing piece and the Intercom Fin cost breakdown. Five questions. Get the answers in writing. If sales can't respond in a single email, the pricing model isn't clean enough to sign.
"Conversation," "resolution," "response," "message," "session," "interaction" — these mean different things to different vendors, and the definition directly controls your bill.
Ask for:
Example red flag: "resolution" defined as "a conversation the AI handled without escalation, with no customer reply for 24 hours." Silent exits count. Most of your bill may come from customers who got nothing useful and bounced.
Just as important as what counts: what's explicitly excluded. Get the following in writing, too:
These exclusions add up fast. A vendor that bills auto-greetings can double your invoice in a busy month.
Per-resolution and per-response vendors often run on top of a per-seat platform. Intercom Fin requires an Intercom seat. HubSpot's AI features live behind Service Hub Pro. Salesforce Agentforce sits on Sales/Service Cloud.
Ask:
The real floor isn't "$X per resolution" — it's "(tier price × min seats) + usage." On a 5-seat minimum at $99/seat, that's $495/mo before any AI usage, whether you need five dashboard users or one.
If the AI is the product you're buying — not a feature layered on a CRM — per-response pricing avoids this trap entirely. Kazozo's pricing has no seat minimums.
Every AI plan has an included quota — a number of resolutions, responses, or sessions per month. What happens at the cap is the question.
Three common patterns:
Ask for the overage rate, whether it's tiered (first 500 overage units cheaper than the next 5,000), and whether hitting overages mid-contract triggers any fee reset. On Kazozo, overages above your included responses are flat per-response pricing — no hidden tier jumps.
This is the question that separates clean pricing from hidden pricing. When the AI can't resolve something and hands off to a human, three things can happen:
The third pattern is more common than you'd think in per-resolution pricing. Ask for the behavior in writing, and test it during a pilot if possible.
Paste this into your reply to any AI vendor quote, and don't sign until every answer is on paper:
Before we proceed, can you confirm the following in writing:
- Exact definition of the billing unit, including what triggers a billable event.
- Full list of excluded message types (auto-greetings, templates, internal testing, human-agent replies).
- Platform tier required for AI access, and minimum seat commitment at that tier.
- Overage pricing above the included volume, and whether any billing resets apply.
- Whether escalated conversations continue to count as billable resolutions after a human takes over.
If the answers are clean, the pricing model is clean. If any answer hedges, that's usually where your actual bill will come from.
AI pricing can be the cleanest SaaS billing you've ever signed, or the most opaque. The difference is almost entirely in the five definitions above. Per-response pricing — what Kazozo uses — is the most legible of the three common models because a response is an observable unit, not a definition.
If you want to see what your real traffic would cost, the ROI calculator models your monthly bill on Starter, Growth, and Pro. Or book a 20-minute demo — we'll walk through your current quote and flag anything that needs the five-question treatment.