1. Message credits
Who uses it: Chatbase, SiteSpeakAI, SiteGPT. How it works: you buy a monthly bucket of credits; each reply spends some. The trap: "some" is not "one". Premium models consume multiple credits per reply, so a 500-credit plan may be 150 real answers. Vendors also cap indexed pages or training sources separately, which means a content-heavy site pays for a higher tier without any extra traffic.
2. Per-resolution or per-outcome
Who uses it: Intercom Fin ($0.99 per outcome), Helply ($1 per ticket with a $250/month and $3,000/year minimum). How it works: you pay when the AI is judged to have resolved something. The trap: the vendor defines the judgement. Intercom counts an outcome when the customer confirms, when the customer simply doesn't come back, or when Fin completes a workflow or handoff — so silence is billable. It aligns incentives well at enterprise scale and is genuinely fair-feeling, but you cannot forecast it precisely, and minimum commitments can dwarf a small team's actual usage.
3. Seats plus bundled AI credits
Who uses it: Crisp ($45–$295/month by seat count, with $5–$75 of AI credits bundled), Intercom's underlying plans. How it works: you buy human capacity, and AI rides along. The trap: you pay for people who watch an inbox, which is exactly the cost automation was supposed to remove. Adding a colleague to look at analytics costs $10 a month; the AI meter still runs underneath.
4. Conversation caps
Who uses it: Tidio (billable conversations, Lyro AI conversations and Flows visitors — three meters at once). How it works: a conversation is a unit regardless of length. The trap: overlapping meters. A single visitor can consume a billable conversation, a Lyro conversation and a Flows visitor slot, and each tops up separately.
5. Per-task
Who uses it: eesel AI ($0.40 per ticket or chat session, $4 for heavy generation tasks, $50 free to start, spend caps you set). How it works: one session equals one task, however many messages it takes. The trap: few, honestly — it is one of the cleaner meters in the market. It assumes you already run a helpdesk for it to sit on top of.
6. Per-answered-message (what Flocci Chat does)
One credit is charged when a visitor's message is actually answered. Nothing else is metered: crawling, uploads, re-indexing, ticket handling, operator replies, analytics, unanswered turns and failed turns all cost zero. Signup includes 500 free credits, so you can measure your real volume before spending anything. Because the same credit wallet works across every Flocci app, and platform plans (₹499 and ₹2,999 a month) cover the whole suite, support automation stops being a separate vendor line item.
Worked example: 1,000 answered questions in a month
Take a site that answers 1,000 real support questions in a month, with two colleagues watching the inbox. Under message credits you need the tier above 1,000 and you must guess the model multiplier. Under $0.99 outcomes that is up to $990 if most conversations end without a follow-up. Under $1-per-ticket outcome pricing it is $1,000 against a $250 minimum you were paying anyway. Under seats plus credits you buy two seats and top up AI credits. Under per-task billing at $0.40 it is $400. Under per-answered-message billing it is exactly 1,000 credits — the first 500 free in your first month — with no seats to buy. The point is not that one number is always smallest; it is that only one of these can be computed in advance.
Questions to ask any vendor before you sign
- Does a single reply ever cost more than one unit? Under what conditions?
- Are unanswered, clarifying or failed turns billed?
- Is there a cap on indexed pages, training sources, agents or seats — and what happens at the cap?
- Is there a monthly or annual minimum commitment?
- What does removing vendor branding cost per year?
- When credits run out mid-month, does the assistant stop talking to customers or degrade gracefully?