Per-seat pricing is usually cheaper when a few people use AI heavily and unpredictably every day. Usage-based pricing is usually cheaper when the work is a repeatable, countable job (a message, a ticket, a document) run at volume by software. To decide, price both at your busiest month: if usage still costs less than the seats, choose usage.
Most owners compare the wrong things. They put an average month's usage bill next to the seat price and conclude usage is a bargain, then meet the peak month. Or they buy seats for a job that software could do on its own, paying for people to copy and paste. The two models also buy different things: a seat buys a person open-ended access, while usage buys a specific unit of work. Once you see that, most decisions make themselves.
How each pricing model charges, in plain terms
There are really three models on the market, and most tools now mix two of them.
| Model | What you pay for | Real examples (list prices, September 2026) |
|---|---|---|
| Per seat | Each named person, per month, whatever they use | ChatGPT Business $25 per seat monthly or $20 annual; Claude Team the same; Microsoft 365 Copilot Business $21 annual |
| Usage: tokens | The amount of text a model reads and writes; a token is a chunk of a word, and a million tokens is roughly 750,000 words | OpenAI gpt-5.6-luna $0.20 per million input tokens and $1.20 per million output; Claude Haiku 4.5 $1 and $5 |
| Usage: tasks or credits | Each step or action the software completes | Zapier Professional 750 tasks for $19.99 annual; Make from about $9 for 5,000 credits; Notion credits $10 per 1,000 |
| Usage: outcomes | Each conversation the AI resolves or completes | Intercom's Fin $0.99 per outcome; HubSpot Customer Agent about $0.50 per resolved conversation |
| Usage: sessions | Each AI conversation, resolved or not | Freshdesk's Freddy AI agent $49 per 100 sessions after a one-off 500 |
| Hybrid | Seats plus an included pool of credits, then pay for extra | monday.com (1,000 to 3,000 credits a month by plan); ClickUp Brain AI (1,500 credits per user); Asana (5 AI requests per user, capped per account) |
One distinction matters more than the others: whether the unit is charged on success. Fin and Zendesk charge only for resolutions; Freshdesk charges for every session; token pricing charges for every call, including the ones that produce rubbish. Outcome-based AI pricing goes deeper into how "resolved" gets defined.
The break-even sum, step by step
You need three numbers: how many people would need a seat, how many units of work a month, and what one unit costs. Then compare at peak.
Seat cost per month = people x seat price
Usage cost per month = units in your BUSIEST month x price per unit
+ any base plan the usage runs on
Add 25% to usage for surprises (a launch, a complaint spike).
If usage (with buffer) < seat cost -> usage-based is cheaper
If usage (with buffer) > seat cost -> seats are cheaper
If they're within 20% of each other -> pick the one you can predictThe last line is deliberate. When costs are close, predictability is worth more than a few dollars, especially for a small business where a surprise bill lands on the owner's card.
A bicycle repair shop prices the same job both ways
A worked illustration: a three-person bicycle repair shop (the owner and two mechanics) handles about 350 repairs a month, rising to 600 in the spring rush. Every job needs two customer messages: a booking-in confirmation with the estimate, and a collection message summarising the work from the mechanic's notes. At the moment the owner writes them all.
Option A: give everyone a seat
Three ChatGPT Business seats cost $75 a month on monthly billing, or $60 on annual. The mechanics paste their notes in and get a tidy message back in about two minutes. Cost is fixed at $60 to $75 whether it's January or April, and everyone can also use the assistant for other things: supplier emails, parts research, social posts.
Option B: automate the messages and pay per use
The shop's booking system sends each job's notes to an automation that calls a small model through the API, then drafts the message for a quick check. The sums for a normal month:
Messages: 350 jobs x 2 = 700 messages
Input tokens: 700 x 1,100 tokens = 770,000 -> x $0.20/M = $0.15
Output tokens: 700 x 150 tokens = 105,000 -> x $1.20/M = $0.13
Model cost: about $0.28
Zapier tasks: 700 runs x 2 action steps = 1,400 tasks
Zapier plan: Professional, 1,500 tasks = $39 annual ($58.50 monthly)
Total, normal month: about $39.30 on annual billingThe model costs 28 cents. The plumbing costs $39. That's the most common surprise in usage pricing for small businesses: the AI is nearly free at this scale, and the automation platform is where the money goes.
In the spring peak, 600 jobs mean 2,400 tasks, which overruns the 1,500-task plan by 900. Zapier's pay-per-task billing on an annual plan charges extra tasks at 1.25 times the plan's rate, about 3.25 cents each, so the peak month costs roughly $39 + $29 + $0.48 of model = about $69. On a monthly plan the multiplier is 2.5 times, and the same overrun would cost about $88 on top of $58.50.
What the shop chose
Option B, plus one ChatGPT Plus subscription for the owner at $20, for a typical total of about $59 a month and roughly $89 in the peak. That's close to Option A in money, so predictability alone wouldn't decide it. What tipped it was time: the automation removes about 700 two-minute copy-and-paste jobs a month, around 23 hours, and the mechanics no longer need to use an assistant at all. When usage pricing and seats cost about the same, ask which one removes work rather than just speeding it up. The API side of the sum is explained further in what the ChatGPT API costs for a business automation.
Per-session vs per-outcome support pricing: a break-even rate
Support chatbots are where the choice of unit changes the answer most. Compare Freshdesk's per-session charge with Intercom's per-outcome charge for the same shop's online parts store, which gets 500 chat conversations a month.
- Freshdesk: $49 per 100 sessions after the one-off free 500, so 500 sessions cost $245 whether the bot resolves them or not.
- Intercom's Fin: $0.99 per outcome, where an outcome includes a confirmed resolution, a customer leaving without asking for more help, or a completed handoff procedure.
The break-even is a single division: $0.49 per session divided by $0.99 per outcome is about 49 per cent. If more than roughly half of conversations end as billable outcomes, the per-session price is cheaper per result; below that, per-outcome is. At a 60 per cent outcome rate, Fin costs 300 × $0.99 = $297 and Freshdesk's $245 works out at $0.82 per resolved chat. At a 30 per cent rate, Fin costs $148.50 and Freshdesk's cost per resolved chat rises to $1.63.
So the question to answer before choosing is not "which is cheaper?" but "what share of our chats could a bot actually close?". For a parts store with lots of "where's my order?" questions, that share can be high. For a repair workshop whose chats are mostly "can you look at my wheel?", it's low, and paying per session means paying for many conversations that end with a human anyway. Note the seat costs of each helpdesk sit on top; the Freshdesk vs Zendesk comparison adds those in.
Hybrid plans: seats plus a pool of credits
Most work-management and knowledge tools now sell seats with an included credit allowance, then charge for extra. The seat price is the floor; the credits are where the bill can move.
Take a four-person bakery on monday.com Standard, used here as an illustration. Seats come in buckets, so it buys five at $12 each on annual billing: $60 a month, with 2,000 AI credits included. Sidekick messages cost between 1 and 8 credits depending on complexity since monday.com started charging them in July 2026. If three staff send 10 messages a working day averaging 4 credits, that's 3 × 10 × 22 × 4 = 2,640 credits, 640 over the allowance. At about $0.01 a credit on yearly billing, the extra is around $6.40. Hardly a worry. The risk comes from automations: an AI step that runs on every one of 1,200 wholesale orders a month uses credits on every run, and that's a volume no person would ever generate by chatting.
Notion publishes rough per-run costs for its Custom Agents, which make the same point. A daily-brief agent runs at about $0.10 to $0.30 per run, so every weekday for a month is at most $6.60. A mail-triage agent at $0.04 to $0.10 per run, handling 30 emails a day, is 660 runs and $26 to $66 a month. Same credit price; very different bills, because one runs once a day and the other runs on every email.
Where usage pricing catches people out
- What counts as a unit. Fin counts a customer who simply stops replying as an outcome. Freshdesk counts every chat within 24 hours as one session, and every email thread within 72 hours as one. Read the definition before modelling costs.
- Multipliers inside the unit. An AI by Zapier step uses 1, 3 or 5 tasks per run depending on the model tier chosen. A realistic mistake: an owner builds a lead-sorting zap on the top tier "to be safe". At 900 leads a month that's 4,500 tasks instead of 900, enough to push a small plan into overage. Test on the cheapest tier first; sorting and extracting rarely need more.
- Overage rates. Zapier charges 1.25 times your plan's per-task rate for overage tasks once you opt in to pay-per-task billing, and pauses Zaps at three times the plan limit. Zendesk says its pay-as-you-go overage costs more per resolution than buying a committed block.
- Automatic top-ups and upgrades. Freshdesk session packs auto-recharge when you fall below a threshold, 50 sessions by default. Tidio can auto-upgrade at 95 per cent of a quota. Both are useful, and both mean the bill can rise without anyone deciding it should.
- One-off allowances that look monthly. Freshdesk's 500 free sessions are once per account, not per month. Tidio's free Lyro conversations are also one-off.
Where seat pricing catches people out
- Minimums. ChatGPT Business and Claude Team both need two seats, so a sole trader pays for someone who doesn't exist.
- Workspace-wide add-ons. ClickUp's AI add-on is billed across the paid members of a workspace, so adding it for two keen users in an eight-person team bills eight.
- Buckets. monday.com sells 3, 5, 10 and so on, so a team of six pays for ten.
- Idle seats. A seat nobody uses costs the same as a busy one. Microsoft 365's admin centre shows Copilot usage over 7, 28, 90 and 180 days; most business plans have something similar. Anyone inactive for 28 days is a candidate for a cheaper plan. Standard vs premium AI seats covers who needs the larger seat at all.
Questions to put to a vendor before you model anything
Pricing pages rarely answer the questions that decide the bill. Send these by email and keep the replies, because a written answer is easier to hold a vendor to than a sales call.
- What exactly is one billable unit, and is it charged when the AI fails or hands over to a person?
- Is any free allowance monthly or one-off per account?
- What happens when the allowance runs out: does the feature stop, top up automatically, or bill overage, and at what rate?
- Can we set a hard monthly cap, and who gets alerted before we reach it?
- Are add-ons billed per user who uses them or across every paid member?
- Do credits roll over, and does annual billing change the credit price?
A short version you can paste:
Before we choose a plan, could you confirm in writing:
1) the definition of a billable [resolution/session/credit/task],
including whether handoffs to staff are charged;
2) whether included units reset monthly or are one-off;
3) the overage rate and whether we can set a hard monthly cap;
4) whether the AI add-on is billed for every member or only
for those we assign it to.
We expect around [X] units a month, peaking at [Y] in [month].Including your expected volume and peak month is deliberate: it prompts the vendor to tell you which tier fits, and it gives you a record of the advice if the first invoice looks different.
Matching the model to your situation
| Your situation | Usually cheaper | Why |
|---|---|---|
| 2 to 5 people using an assistant all day for varied work | Per seat | Open-ended use would be expensive and hard to predict on usage |
| A single repeated job run by software (messages, tagging, summaries) | Usage | Model costs are tiny per item; no seats needed for the machine |
| Website chat with mostly simple, closable questions | Per session, if over about half close | Cheaper per resolution at high resolution rates |
| Website chat with mostly complex questions | Per outcome | You pay only when the AI actually finishes something |
| Strongly seasonal volume | Seats, or usage with a hard cap | Protects cash in the peak month |
| One person, occasional use | Individual plan (a seat of one) | Business-plan minimums and base fees outweigh usage savings |
The general method for reading these pricing pages, including where the unit definitions hide, is in how to read AI software pricing.
Checking you picked right after 60 days
Whichever model you choose, look again after two full billing cycles. Four numbers tell you if it was the right call:
- Cost per unit of work. Total AI spend divided by the jobs done (messages sent, tickets closed, documents produced). For the bike shop's automation, about 6 cents per customer message.
- Peak-to-normal ratio. Your most expensive month divided by a typical one. Above about 1.5 on a usage plan, check whether a cap, a committed block or a bigger tier would be cheaper over the year.
- Seat activity. Share of paid seats active in the last 28 days. Below 70 per cent, you're overpaying for seats.
- Overage share. The part of the bill that came from overage or top-ups. If it's regularly over a fifth, you're on the wrong tier; a bigger plan or annual billing is usually cheaper than paying multipliers.
For the illustrative bike shop, the 60-day check showed about 6 cents per message in both March and the April rush, with the April bill about 1.75 times a normal month. The owner compared two options: the 2,000-task tier all year ($49 a month annual, $588 a year) or staying on the 1,500-task tier and paying overage in the two peak months (about $468 plus $58, roughly $526). Staying put was cheaper, so the only change was a calendar reminder to check task use in mid-March. That's the pattern to aim for: a model that fits your normal month, with a known, costed plan for the busy one.
More on choosing an AI pricing model
Can I set a spending cap on usage-based AI tools?
Often, yes, but check how it works. Some tools stop the feature when the pool runs out, some top up automatically, and some cap overage at a multiple of your plan. Zapier stops at three times your task limit, Notion pauses Custom Agents when credits run out, and Tidio can upgrade automatically at 95 per cent of a quota unless you switch that off.
Is usage-based pricing riskier for a seasonal business?
It moves with your volume, which cuts both ways. In a quiet month you pay less than a seat would cost; in the peak month you pay the most exactly when cash is also busiest. Price your busiest month, not your average, and keep a buffer of around a quarter for surprises such as a product launch or a complaint that triggers a spike.
Why do so many AI tools now charge seats and credits together?
Because AI features cost vendors real money every time they run, while seats alone don't track that cost. A seat covers the person and the everyday features; credits cover the heavy work such as agents, long research or automated runs. For buyers it means two numbers to watch, so ask what an average task in your business uses.
Further reads
- Per-Seat SaaS Fees vs One Custom Tool: Five-Year Costs Compared — The long-run view: renting seats versus owning one custom tool.
- Annual or Monthly Billing for AI Tools: Which Saves You More? — The billing-period choice that changes both pricing models.
- How to Negotiate AI Software Prices and Seat Numbers — Get seat counts and credit pools right at renewal.
- How Much Does Customer Service Software With AI Cost per Agent? — Seat and resolution costs combined for support teams.
- AI Subscription Prices Compared for Small Businesses (2026) — Every major tool's seat and usage prices in one place.
- HubSpot AI Agents and Breeze: What Small Teams Should Switch On — How HubSpot's credit-based agents fit alongside its seats.
- How Much Does AI Cost a Small Business in 2026? — Itemised 2026 AI budgets for a sole trader, a six-person office and a 20-person firm, with the charges that grow after month three.
- AI Ordering vs Delivery Apps: What a Takeaway Actually Pays — Delivery app commission against AI phone and web ordering, line by line, with a break-even worksheet a takeaway can fill in with its own numbers.
- How Photographers Cut Editing Time With AI and Keep Their Style — Train an AI editing profile on your own finished work, test it on a job you've delivered, and review by scene so the look stays yours.
- Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — When a legal AI seat beats paying by use, how to work out who in the firm needs one, three monthly budgets and the costs quotes leave out.
- Best AI Help Desk Tools for Small MSPs (2026) — Eight help desk options for small MSPs compared on the AI that matters: summaries, triage, drafted replies, resolution help and pricing model.
- AI Estimating Software for Small Builders: Six Options Compared — Six AI estimating options for small building firms compared on what the AI does, list price, best fit and the catch, plus a way to test them on a real job.
- Can a Small Courier Firm Use AI to Plan Delivery Routes? — What route-optimisation software really does for a small courier firm, what it costs per stop or per driver, and a four-van worked example.
- Best AI Lead Generation Tools for Small Businesses (2026) — Ten lead generation tools sorted by the job they do, with verified 2026 prices, a worked example for each, a costed wholesaler stack and a trial scorecard.
- How Much Does an AI Receptionist Cost for a Small Business? — Four ways AI receptionists charge, real list prices, an itemised cost table, one clinic's month priced four ways, and the hidden costs that change the answer.
- Best AI Business Intelligence Tools for Small Businesses (2026) — Ten reporting and analytics tools compared on what their AI really does, list prices, the tier each AI feature needs and which small businesses each suits.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI API pricing page and ChatGPT Business pricing; Zapier pricing page and pay-per-task billing help article; Intercom pricing page; Freshworks Freshdesk pricing and Freddy AI add-on help pages; Zendesk automated resolutions help article; Notion credits help pages; monday.com AI credits pricing; HubSpot credits documentation; checked September 2026.