Per seat: predictable, until you grow
The classic model charges per agent per month, and its appeal is that you can compute the bill on a napkin. Its problem appears exactly when things go well: every person you add to handle more conversations raises the bill, so the software's cost scales with your headcount rather than with the value it produces. Teams respond by sharing logins — which destroys conversation history and accountability — or by keeping the team artificially small.
Per-seat pricing also quietly punishes the AI era's main promise: if an AI handles half your volume, a per-seat vendor still charges you for every human, and often charges extra for the AI on top.
Per message, per contact, per token: the unpredictable family
Usage-based models sound fairer — pay for what you use — but the unit matters enormously. Billing per message or per monthly contact means a busy month produces a bill you didn't approve in advance. Billing AI per token is the least legible of all: nobody outside engineering knows what a token is, and the same customer question can cost wildly different amounts depending on how the model happened to answer.
The test for any usage-based plan is whether the unit is something you'd recognize as valuable on your own books. Messages and tokens are costs, not outcomes — you can spend a fortune on both while resolving nothing.
Per resolution: pricing the outcome instead
iNTELIGENCIA VIVA bills AI by the resolution: a conversation the AI closed completely, without a human stepping in. Pro at US$29/month includes 250 of them, Business at US$79/month includes 1,000; beyond that, each additional resolution costs US$0.35 or US$0.30. If the AI can't finish and hands off to your team, that conversation doesn't count — you pay when the work actually got done.
The incentive alignment is the point, not just the arithmetic: a vendor billing per resolution profits from an AI that's honestly good and hands off when unsure, not from one that maximizes messages sent. Your human team, meanwhile, is never metered — plans don't charge per seat at all.
The hidden costs no pricing page lists
Whatever vendor you evaluate, ask about the costs that live outside the plan table: WhatsApp Business API conversations are billed by Meta separately, on every platform — a vendor who implies otherwise is hiding it in a markup. Setup and onboarding fees, charges for removing the vendor's branding, exporting your own conversation history when you leave, and "integrations" that are actually a higher tier are the classic four.
And check the free layer's fine print: if the widget that made the tool free stops working the day you cancel the paid plan, the free layer was a hostage, not a feature. A tool worth building on keeps its free promises independent of the paid relationship.
A quick way to estimate your real monthly cost
Count your conversations in a normal month, estimate the share that's repetitive (order status, hours, availability — usually well over half), and price that share as AI resolutions; what's left is your team's actual workload. For most small businesses that arithmetic lands at US$29–79/month all-in — see the pricing page for the current plan table — and, importantly, the entry point costs nothing: the free Directory has no meter attached at all.