A merchant emailed me a screenshot last month. His support tool's plan page said $300/month. His actual invoice was $640. He wanted to know if OrderWise would do the same thing to him.

The gap wasn't a billing error. It was per-resolution AI fees stacked on top of the base plan. He'd had a good month — more orders, more questions, more tickets the AI handled — and every one of those auto-resolved tickets carried a charge he hadn't modeled. He grew, and his support bill grew faster than his revenue did.

That's the trap. Not that AI support is expensive in the abstract, but that the pricing model on most of these tools is designed to scale with your success in a way that quietly eats your margin. Here's how the three main pricing models actually work, and how to put real numbers on what you'll pay before you commit.

The three pricing models

Almost every AI support tool you'll compare falls into one of three buckets. They're not equally honest about which one they are.

Flat seat pricing. You pay per human agent seat, per month. This is the classic helpdesk model — Zendesk's Suite plans start around $55/agent/month, Help Scout sits in a similar range. The cost is predictable and tied to headcount, not volume. The catch: AI features are usually a separate add-on, and the seat model assumes you have human agents in the first place. If you're a one-person store, you're paying for a multi-agent architecture you don't use.

Per-resolution pricing. You pay a fee every time the AI fully resolves a ticket without a human. Gorgias's AI Agent runs at roughly $0.90 per automated resolution on top of your base plan. Intercom's Fin charges $0.99 per resolution. The pitch is "you only pay for outcomes." The reality is that a resolution is exactly the thing you want more of, so the better the tool works, the more you pay.

Per-AI-reply or message add-ons. You buy a bucket of AI replies or "interactions" and pay overage when you blow past it. Tidio's Lyro bills per AI conversation in tiered packs; several smaller apps sell credits. This sits between the other two — less brutal than per-resolution at scale, but the overage pricing is where the surprises live.

I'm not going to pretend one of these is universally evil. Flat seat pricing is genuinely fair for a 10-agent team. The problem is what happens when you grow.

Why per-resolution punishes you exactly when you grow

Here's the part that doesn't show up on the pricing page.

Per-resolution pricing is sold as alignment: you pay for value delivered. But the cost curve runs in the wrong direction for a growing store. Every new order is roughly a new chance at a "where is my order?" ticket. WISMO scales linearly with order volume. So as your orders go up, your auto-resolutions go up, and your per-resolution bill goes up — on top of the base plan you're already paying.

The cruel detail is that these are your cheapest tickets to resolve. A WISMO lookup is an order-status query and a templated answer. The marginal cost to actually serve it is fractions of a cent. But you're being charged $0.90 for it because the pricing is decoupled from the real cost of the work. You're paying a premium on the easiest, highest-volume category in your inbox — the one category where automation should be nearly free.

I wrote a whole post on why 78% of WISMO tickets are fully auto-resolvable. The flip side of that number: if you're on per-resolution pricing, 78% of your inbox is also your fastest-growing line item.

How to model your own monthly cost

Stop reading the plan price. Model your actual cost. You need three numbers:

C = conversations per month (roughly your ticket volume)
A = the share the AI auto-resolves, as a decimal
The pricing terms of each tool you're comparing

Let's run a real store: 1,000 conversations/month, 70% auto-resolved. That's 700 AI resolutions and 300 that touch a human. This is a normal mid-size Shopify store, not a hypothetical.

Per-resolution tool (base ~$300/mo + $0.90/resolution):

300 + (700 × 0.90) = 300 + 630 = $930/month

That's the merchant from the intro. The base plan looked like $300. The real number was three times that, and it climbs every month his orders do.

Per-AI-reply tool (say $50/mo for the first 500 AI conversations, $0.50 each after):

50 + (700 − 500) × 0.50 = 50 + 100 = $150/month

Cheaper here — but watch the overage rate, because at 2,000 conversations the same math gives you 50 + (1,500 × 0.50) = $800.

Flat tool (one price regardless of resolution count, say $99/mo):

99/month, full stop

The point isn't that flat always wins on a single month. The point is what the line looks like over a year of growth. Per-resolution and per-AI-reply both slope upward with volume. Flat is a horizontal line. Plug in your own C and A, then project them forward to where you actually want your store to be, not where it is today. The tool that's cheapest at 200 tickets is often the most expensive at 2,000.

The hidden cost: you pay even when the AI is wrong

This is the line item nobody puts on the pricing page.

On most per-resolution models, a "resolution" is counted when the AI closes the conversation, not when the customer's problem is actually solved. If the AI gives a confident wrong answer, the customer accepts it and leaves, the ticket gets marked resolved — and you get billed. The customer comes back angry two days later, opens a new ticket, a human fixes it, and you've now paid a resolution fee plus the human's time for one underlying problem.

Some tools have refined this with "no resolution charge if the customer reopens within X hours," and credit to them for it. But the window is short, the definition of resolution is theirs not yours, and you're trusting the vendor's own meter to decide what you owe them. The incentive structure is backwards: the tool earns more by closing tickets, not by solving them. A hallucinated tracking number that the customer believes for a day is, on the meter, a successful resolution.

If you want to see how we think about whether an answer was actually correct rather than just closed, our quality report breaks down resolution accuracy separately from resolution count. They are not the same metric and they should never be billed as if they were.

Where OrderWise sits

I'll be direct about our model so you can drop it into the same spreadsheet.

OrderWise is flat. You pay a monthly price for your plan and there is no per-resolution fee and no per-AI-reply charge. If the AI resolves 700 tickets or 7,000, the price on your invoice is the same. We made that choice specifically because of the merchant in the intro — the pricing model shouldn't penalize a good month.

Two things we added because I'd been the person staring at a surprise invoice:

Per-conversation cost visibility. You can see what each conversation actually costs to run, in the dashboard, not buried in a quarterly export. The number is small and we'd rather you see it than wonder about it.

Monthly cost caps. You set a ceiling. The system tells you as you approach it instead of letting the meter run. No tool should be able to hand you a 2x invoice without warning you first.

I'm not claiming flat is right for every business — a large team with predictable, low automation might genuinely pay less on a seat model. But for a growing Shopify store where WISMO volume tracks order volume, a price that doesn't move when you succeed is the one I'd want. If you're weighing the specific trade-offs against a per-resolution incumbent, I put the honest comparison in OrderWise vs Gorgias, and the plan numbers are on our pricing page with no asterisks.

Before you sign anything

Do the math with your own two numbers. Take your real conversation volume, your honest auto-resolution rate, and run all three formulas above. Then run them again at double your current volume. The tool you pick should still make sense at the size you're trying to become, not just the size you are.

If you want to sanity-check your numbers against ours, the free plan covers 50 conversations a month with full Shopify order context and no card required. Run a month of your real traffic through it and compare the invoice to the screenshot you're worried about.

What AI customer support actually costs (and the per-resolution trap nobody warns you about)