How AI Support Is Priced: Credits vs Per-Conversation Billing

Ask a vendor “how much a month” before you sign, and a lot of them can’t actually answer — because the answer isn’t a number, it’s a formula. AI support pricing comes in several shapes: per seat, per conversation, per resolution, or a separate AI usage allowance bolted on top. The names all sound similar. The invoices can differ by multiples.

This isn’t a price comparison — line items across vendors aren’t directly comparable anyway. This is about the structure of the pricing model itself, so you can tell what you’re actually agreeing to, and which models are quietly built to bite you during peak season.

The common billing units

Here are the pricing units you’ll most likely recognize on a quote:

These aren’t mutually exclusive — many platforms stack them: a seat fee as the base, plus a per-conversation or per-resolution layer on top of the AI portion. Before you sign, run every quote through this table and ask which part is fixed and which part floats.

Per-conversation billing: reasonable on paper, painful in practice

Per-conversation pricing sounds fair on the surface — use more, pay more. But cross-border support has an unusual conversation shape: one customer, across time zones and language barriers and shipping anxiety, might ask the same question three separate times. And a single peak-season spike can double or triple your monthly volume.

That’s where per-conversation pricing starts to hurt: you can’t estimate at signing time how many conversations you’ll actually generate in a month, so your bill becomes a floating number — and it floats hardest exactly during the peak-season month when your cash flow is already tight. Worse, to keep the bill in check, support leads instinctively start rationing AI — manually routing conversations to a human that the AI could have handled, purely because per-conversation pricing makes them nervous. That’s the opposite of why you bought AI support in the first place.

Per-resolution billing: the better the AI, the more you pay

Per-resolution pricing is another common model. Zendesk’s publicly documented pricing describes exactly this kind of per-resolution billing¹. The logic is “you only pay when the AI actually resolves it” — which sounds fairer to the buyer.

But look one layer deeper and there’s a counter-intuitive side effect: the better the AI performs and the higher its resolution rate, the more you pay. The more accurately it learns and answers, the bigger your bill — which directly contradicts your incentive to improve it. You end up hesitating over whether to train the AI to be smarter, because a smarter AI means a worse-looking bill next month. That’s not an incentive structure that belongs in a tool sold on cost savings.

Intercom’s Fin bills per outcome (successfully resolved conversations), which is structurally the same category of problem as per-resolution billing².

YundaDesk’s model: AI credits included in every plan

YundaDesk takes a third approach: every plan includes a fixed AI credit allowance, with no per-conversation or per-resolution surcharge on top. Within your allowance, usage is priced the same regardless of volume, and your monthly bill is a fixed number you know the day you sign.

The reasoning is straightforward: we don’t want a support lead to hesitate for even a second over whether to let the AI take a given conversation. The AI credit allowance is sized to cover normal business volume, and when peak-season traffic doubles, you still pay the same plan price — the bill doesn’t move just because inquiry volume did³. If you do go over your allowance, the overage terms are a fixed, disclosed rate — not a floating black box.

This lines up with how AI support/en/blog/what-is-ai-customer-service/ is supposed to work in the first place — catching as many inquiries as possible so humans can focus on judgment calls. If the pricing model quietly rewards you for using the AI less, the product and the invoice end up working against each other.

Three questions to ask before you sign

Whichever vendor you pick, pin down these three questions — they’ll save you from most billing surprises:

• What’s the actual billing unit — seats, conversations, or resolutions? Are they stacked on top of each other?
• If peak-season volume doubles, how does my bill move? Is there a cap or an allowance?
• Is there a perverse incentive where “the better the AI performs, the more I pay”? If so, will that make my team instinctively use the AI less?

Get the answers written into the contract or the quote’s fine print — don’t rely on a sales rep saying “that usually doesn’t happen.”

A predictable bill is what lets you actually trust the AI with volume

Pricing sounds like a finance question, but it actually decides whether you dare hand real volume over to the AI. If every AI interaction runs a silent cost calculation in your head, AI support stays a small pilot forever — it never becomes the real first line of defense that actually absorbs most of your inquiries.

Pricing structure deserves at least as much scrutiny as the feature list — a full feature set means nothing if the bill spirals out of control. That’s also why “is the bill predictable” is the first question in our evaluation framework at /en/blog/choosing-ai-support-platform/.

There’s no single “correct” pricing model, but there’s one line worth holding before you sign: your bill shouldn’t rise and fall with how well the AI performs. A predictable bill is what lets you actually hand real volume to the AI — and that’s the entire point of buying AI support to begin with.