You can usually guess what a normal software subscription will cost next month. It's the same as last month.
AI tools are changing that. A growing number charge for what the AI actually does: per conversation, per action, per problem solved. That can be fair. It can also mean a surprise on the bill the month you get busy.
This post explains the three pricing models, shows real published prices as of November 2026, works through an example with made-up volumes, and lists ways to keep the bill in check. Prices change often, so check the vendor's page before you buy. For the full cost picture beyond tools, see what AI implementation costs a small business.
The three ways AI gets billed
1. Per seat
You pay a flat amount per person per month. It's predictable, and it's what most owners are used to.
Team chat assistants mostly work this way. Third-party roundups put ChatGPT Business and Claude Team at around $20 per user per month billed annually, after an April 2026 price cut, with Microsoft 365 Copilot Business around $21 per user on top of a Microsoft 365 license. We couldn't open the vendors' own pages for those, so treat them as rough and check the roundup against the vendor.
How much do small businesses actually spend? A JPMorganChase Institute study of Chase transaction data found median entry spending of $20 a month for firms that started in 2024, rising to $29 in 2025. That's paid tools only, and it's a median, so some spend far more.
2. Per use (minutes, conversations or actions)
You pay for how much you use. Phone answering tools often bill by minutes. Rosie's pricing page, read in November 2026, lists $49 a month for 250 minutes, $149 for 1,000 and $299 for 2,000. Its website texting add-on is $50 a month with 25 conversations included and $1 each after. The page didn't state what happens when you go over on minutes, which is exactly the kind of thing to ask.
Salesforce's Agentforce pricing lists $2 per conversation, or "Flex Credits" at $500 per 100,000 credits with 20 credits per action. That works out to about $0.10 per action. Salesforce notes prices are subject to change. This is a large-company product, but it shows where pricing is heading.
3. Per resolution (or "outcome")
You pay only when the AI solves the problem without a person stepping in.
- Intercom's pricing page lists $0.99 per outcome for its Fin agent, on top of helpdesk seats at $19, $85 and $132 per seat per month depending on the plan.
- Zendesk's pricing page lists $1.50 per automated resolution on a committed plan and $2.00 pay-as-you-go. It says you're billed only when the AI resolves the issue without a human. Seat plans run $19 to $115 per agent per month billed yearly.
All of these are as of November 2026, from vendor pages. They're the best public numbers we found. They're list prices, so what you'd pay depends on your plan.
Why it matters: the bill moves with volume
Per-seat pricing is steady. Usage pricing follows your business, and that has two sides.
The good side. You pay for work that happened. If the AI solves nothing, a per-resolution plan costs little. That lines up cost with value.
The bad side. A busy month, a promo that brings 300 texts, a storm that doubles your calls, a bot stuck in a loop: the bill goes up with no warning. According to a vendor survey by BetterCloud, nearly 40% of AI apps are billed on usage, and two-thirds of IT pros worry about accidental data loss from unapproved AI more than about slow adoption. Take that as a vendor survey of IT professionals, reported in BetterCloud's 2026 State of SaaS report, not a measure of what owners pay.
A worked example (made-up numbers)
Illustrative only, not a quote. Say a shop's AI agent resolves 200 customer questions in a normal month with no human help.
- At $0.99 per resolution: 200 x $0.99 = $198.
- At $2.00 per resolution: 200 x $2.00 = $400.
- Now a busy month hits and it's 600 resolutions. That's $594 at $0.99, or $1,200 at $2.00.
Same business, same tool, three times the volume. That's why a ceiling matters.
Two more things the example hides. First, what counts as a "resolution" is the vendor's definition, so ask for it in writing. Second, a low per-unit price isn't automatically cheap. Entry plans like $49 a month look small until your call volume outgrows the included minutes.
How to cap the bill
None of this is hard. It's about asking before you sign, and checking after.
- Ask for a monthly ceiling. Can you set a spending limit, or a hard stop at a number of conversations? If the vendor can't say, assume it can't.
- Ask what happens at the limit. Does the tool stop, hand off to a person, or keep going and bill you?
- Get the definition of the billable unit. What's a "resolution," a "conversation," an "action"? Does a conversation that ends with "talk to a person" count?
- Ask for overage rates in writing. Rosie's page didn't state one for minutes. Don't guess.
- Turn on usage alerts, if the tool has them. A text at 50% and 80% of budget beats a bill.
- Review the bill monthly for the first three months. Write down what you spent and what you got for it. If you set one number to measure when you started, check it next to the bill. See whether AI saves small businesses money.
- Watch for loops and spam. A bot that talks to another bot, or a form flooded by junk, can run up usage. A person glancing at the daily count catches that.
- Know the exit. Month to month is safer than annual while you're learning.
Per-seat plans have their own trap: paying for ten seats when two people use them. Check who actually logs in each quarter.
Which model is better for a small business?
It depends on the task.
- Steady, team-wide use (drafting, summaries): per seat is usually simpler.
- Lumpy or unpredictable volume (phones, chat): per use or per resolution lines up with value, but only with a cap.
- A task where "solved" is clear: per resolution can be fair, if you agree on the definition.
For phone answering specifically, see what an AI receptionist costs.
Frequently asked questions
What does per-resolution pricing mean?
It means you pay only when the AI solves the customer's problem without a person stepping in. Intercom lists $0.99 per outcome and Zendesk lists $1.50 to $2.00 per automated resolution, as of November 2026. The catch is the vendor decides what counts as a resolution, so get the definition in writing before you buy.
How do I stop an AI bill from running up?
Ask for a monthly ceiling and what happens when you hit it. Turn on usage alerts. Get overage rates in writing. Review the bill monthly for the first three months. A flat monthly price with a stated usage ceiling is the simplest protection. See our cost guide.
Is per-seat or per-use pricing cheaper?
Neither is always cheaper. Per seat is steadier for team tools. Per use can be cheap at low volume and expensive at high volume. Use your own expected volume to do the math, as in the example above. Then add 50% for a busy month and see if you can live with that number.
How much do small businesses spend on AI tools?
The best public number we found is from a JPMorganChase Institute study of transaction data: median entry spend of $20 a month for 2024 starters, $29 in 2025, for paid tools only. Most of what owners spend beyond that is setup, training and upkeep, not software. See what an AI receptionist costs for one example.
How Di-Hy fits
The tool price is the small part. The rest is setting it up right, training your team, and watching the bill and the results. Keep It Running includes a check on usage so a busy month doesn't surprise you. If you'd like a plan before you spend, Get a Plan maps what to buy and what to skip, and the free audit is a good first look. We're in Ybor City and work across Tampa Bay.
Next step: send us the AI bills you've got now. We'll tell you what you're paying per unit and where a cap belongs. See also our guide to AI implementation for small business.
Prices are published list prices from vendor pages read in November 2026, and they change. Examples with made-up numbers are labeled. Not financial advice.


