Read AI software pricing by finding the unit you pay for (seats, credits, tasks, tokens or resolutions), how many units your plan includes, what extra units cost, and the terms around them: minimum seats, billing period, whether credits expire, and when promotions end. Then multiply by your own expected monthly volume to get a real figure.
The headline number on a pricing page is usually the smallest sensible purchase, at the cheapest billing option, sometimes on promotion. AI products increasingly charge by usage as well as, or instead of, by user, so your real cost depends on your volume. Two tools that both say "from $25 a month" can end up far apart once your numbers go in.
The six pricing units, decoded
| Unit | What it counts | Real examples (list prices) | The question to ask |
|---|---|---|---|
| Seat | Named users, whatever they do | ChatGPT Business Standard, $25 a seat monthly or $20 billed annually; Microsoft 365 Copilot Business, $21 a user on annual terms | Is there a minimum, and can seats be reduced mid-term? |
| Usage tier | A multiple of an unpublished baseline | ChatGPT Pro at $100 (5x Plus usage) or $200 (20x; new sign-ups paused since 10 September 2026); Claude Max from $100 (5x) and $200 (20x) | Five times what, and does anyone on my team hit the limit? |
| Credits | A platform's own currency, spent by features | HubSpot credits at $10 per 1,000; Make credits; Microsoft Copilot Credits in packs of 25,000 for $200 a month | How many credits does one of my jobs use, and do unused ones expire? |
| Tasks or executions | Steps or runs of an automation | Zapier's Professional plan, 750 tasks for $19.99 a month on annual billing; n8n, which bills in euros, counts each whole workflow run as one execution | What counts as one, and what happens above the limit? |
| Tokens | Text processed by an AI model | Claude Haiku 4.5 at $1 per million input tokens and $5 per million output; Meta's Business Agent at $2 per million tokens | How many tokens does my typical job use? |
| Outcomes | Results, such as resolved conversations | Intercom Fin at $0.99 per resolved outcome; Help Scout's AI Answers add-on at $0.75 per resolution; Jobber's Receptionist at $29 for 30 conversations, then $0.79 each | How does the vendor define the outcome? |
Many products combine two or three units: seats for the people plus credits or outcomes for the AI. Seats versus usage is the classic trade-off, and per-seat vs usage-based AI pricing works through which costs less at different volumes. The rest of this tutorial is about reading any combination accurately.
Seats, and the tiers that multiply them
Seat pricing looks simple: users times price. The traps are in the edges.
- Minimums. Two seats is the smallest purchase on ChatGPT Business and Claude Team, which is why many sole traders choose an individual plan instead. Help Scout's Pro plan lists a minimum of ten users.
- Reductions. On annual terms, check whether seats can be removed mid-term or only at renewal. Adding seats is usually easy; removing them often isn't, so start small.
- Seat types. Some plans mix seat tiers. ChatGPT Business Premium seats are $125 a month ($100 annually) for five times the usage of Standard, with no five-hour usage limit. Claude Team's Standard seats give roughly 1.25 times Pro's per-session usage and Premium seats roughly 6.25 times.
Usage multipliers are the hardest part of any AI pricing page to read, because the baseline is never published as a fixed number of messages. "5x Plus usage" tells you the ratio, not the quantity, and limits can change. The practical way to decide is observation: run a month on the standard tier, note who actually hits limits, and upgrade only those people. Standard vs premium AI seats covers that decision in detail.
Credits: a currency you need to convert
Credits let a vendor price many different features with one number, which is convenient for them and confusing for you until you convert them into your own jobs.
- HubSpot: a thousand credits cost $10. One resolved Customer Agent conversation spends 50 of them (about 50 cents); each lead the Prospecting Agent recommends for outreach spends 100 (about $1). Professional plans include about 3,000 credits a month, which covers roughly 60 Customer Agent resolutions before extra credits are needed. Unused credits don't roll over.
- Make: most actions cost a single credit, but its built-in AI modules are metered on the tokens they process; its help pages give about 18,080 input tokens per credit on the small model tier and 3,616 on the large one. Unused credits lapse when the billing term ends, and a scheduled trigger spends credits on every check, busy or not.
- Microsoft: Copilot Studio runs on Copilot Credits, sold in capacity packs of 25,000 for $200 a month. Power Automate's AI Builder credits are giving way to them too: licences bought or renewed from 1 November 2026 no longer come with seeded AI Builder credits.
- Canva: each plan comes with a monthly AI allowance that its AI features draw on.
The conversion to do every time: credits per job, times jobs per month, compared with the credits included. If you can't find how many credits one of your jobs uses, ask the vendor for a worked example at your volume, in writing.
Tasks, executions and tokens: pricing by volume of work
Automation platforms and AI models charge for work done. Three rules make their pages readable.
Know what counts. On Zapier, a task is one successful action; the trigger, filters, Paths and Formatter cost nothing, while an AI by Zapier step is charged at one, three or five tasks according to the model tier selected. n8n counts a whole workflow run as one execution, however many steps it has. The same workflow can be 4 units on one platform and 1 on another.
Know what happens above the limit. Zapier bills extra tasks at a premium over the plan's own per-task price (a quarter more on annual plans, two and a half times the price on monthly ones), switches this on by default for accounts created after January 2024, and halts Zaps once usage hits triple the plan's allowance.
Turn tokens into words. A useful rule of thumb: a million tokens is roughly 750,000 English words. A 250-word customer email plus a page of instructions is perhaps 1,000 to 1,500 tokens. At Claude Haiku 4.5's $1 per million input tokens, a thousand such emails cost around $1.50 to read; OpenAI's gpt-6-luna, at $0.10 per million input tokens, around 15 cents. A ChatGPT or Claude subscription doesn't cover any of this: API usage is billed separately.
Outcome pricing: read the definition before the price
Paying per resolution sounds like paying only for success, and it can be good value. But "resolved" means what each vendor says it means, and the definitions differ:
- Intercom: if the customer sends nothing for 24 hours after Fin's final reply, the conversation is billed as an "assumed resolution".
- Zendesk: a conversation closes after two hours of messaging inactivity by default (72 hours on email and web forms), then an AI check sorts it into a tier; since 18 May 2026 only verified resolutions are billed, while assisted escalations and unverified contained resolutions are free.
- Help Scout: a resolution is a conversation resolved by AI Answers without human help, billed only when the customer doesn't escalate, search the knowledge base, ask more questions or say they need more help.
A customer who gives up and phones you instead may count as resolved under a silence-based rule. Before comparing $0.75 with $0.99, compare the rules, ask each vendor whether you can review the conversations it bills as resolved, and during a trial check a sample of "resolved" conversations yourself. The pricing model as a whole is examined in outcome-based AI pricing.
A sports shop turns three pricing pages into one number
A sports equipment shop with four customer-service staff handles about 600 email and web-chat conversations a month, plus about 250 conversations in Instagram and WhatsApp messages. The owner has three options open, each priced in a different unit.
- Seats: ChatGPT Business for drafting. Four Standard seats at $25 a month on monthly billing is $100, or $80 billed annually. Staff paste in a question, get a draft, and send it themselves. The cost is flat, whatever the volume.
- Seats plus outcomes: a helpdesk with an AI agent. Help Scout's Standard plan at $25 a user a month for four users is $100, plus the AI Answers add-on at $0.75 per resolution. In a trial, the agent fully resolved about 30% of conversations; at 600 a month that's 180 resolutions, $135. Total about $235.
- Tokens: Meta's Business Agent for messages. Meta charges $2 per million tokens, 4 to 5 cents a message by its own estimate, so 250 conversations averaging six AI messages each come to about $60 to $75 a month, with no free allowance.
Then the owner does the step most buyers skip: re-running the sums for the shop's busy season, when volume roughly doubles.
| Option | Normal month | Busy month (double volume) | Cost behaviour |
|---|---|---|---|
| ChatGPT Business, 4 seats | $100 | $100 | Flat; the extra work lands on staff time |
| Helpdesk with AI Answers | About $235 | About $370 | Rises with resolutions |
| Business Agent in messages | About $60 to $75 | About $120 to $150 | Rises with every message |
It also helps to convert each option into a cost per conversation and set it beside the staff cost it replaces. At an illustrative $22 an hour, five minutes of a person's time is about $1.83. The seats work out at roughly 12 cents per conversation across all 850, but staff still spend most of those five minutes. A helpdesk resolution costs $0.75 and, when genuine, saves the whole five minutes. A message-channel conversation costs about 24 to 30 cents in AI fees. Seen per conversation, the dearest monthly option can be the cheapest way to get work off people's desks, and the cheapest monthly option can be the one that saves least.
None of these is automatically the winner. Seats are cheapest in a busy month but save the least time, because people still write and send every reply. The helpdesk agent costs most, but each resolution at $0.75 compares well with five minutes of staff time, provided the resolutions are genuine. The shop's decision: the helpdesk agent for web chat, where questions are repetitive (sizing, stock, delivery times), with a monthly check of 20 "resolved" conversations; and drafting help for email, where questions are more varied. The sums didn't make the decision; they made it an informed one.
The fine print that changes the total
- Billing period. Annual billing is cheaper: Zapier's Professional plan is $19.99 a month annually against $29.99 monthly; Claude Pro is $17 a month billed annually against $20; Help Scout shows 16% off for annual billing. But annual usually locks your seat count until renewal. Annual or monthly billing weighs up the trade.
- Promotions. HubSpot's pricing page often shows promotional prices; Starter's undiscounted price is $20 per seat per month. Microsoft 365 Copilot Business is $18 on a promotion for purchases between 1 July and 31 December 2026, against a $21 list price. Budget year two at list.
- Included versus add-on. Microsoft bundles Copilot Chat into its business plans free of charge; the full Copilot licence costs extra. The Business Standard plus Copilot bundle ($23.50 on annual terms) costs less than Business Standard ($14) plus the add-on ($21) bought separately.
- Feature gates. Customer Agent only comes with Service Hub on Professional or Enterprise; the Free and Starter tiers get Breeze Assistant for writing and summaries and nothing more agentic. The cheapest tier that shows the feature on a comparison chart may not be the cheapest tier that includes it.
- "From" and "contact sales". "From" is the entry tier. "Contact sales" usually means the price depends on volume and negotiation, which is also an opening: see how to negotiate AI software prices and seat numbers.
Reading a pricing page with an AI assistant, then checking the maths
An assistant can extract a pricing structure quickly if you give it the page text and your numbers, and ask for the working rather than just the answer:
Below is the text of a software pricing page. Extract: every plan,
its price, the billing period each price assumes, the unit charged
(seat, credit, task, token, resolution), what is included, overage
rates, minimums, promotions and their end dates.
Then calculate my monthly cost on MONTHLY billing for: [number]
users, [number] conversations a month, of which about [percent]
would be handled by the AI. Show every step of the sum. Quote the
page for each figure you use; write "not stated" where the page
doesn't say.
[paste pricing page text]For the sports shop's helpdesk option, the first answer (illustrative) was:
"Standard plan: $25 per user per month. 4 users = $100. AI Answers: $0.75 per resolution x 600 conversations = $450. Monthly total: $550."
Two mistakes, both common. It multiplied the resolution price by all 600 conversations instead of the 180 the AI actually resolves, overstating the bill by $315. And it assumed $25 was the monthly-billing price, when the page's toggle between monthly and annual billing means you should confirm which price is showing before relying on it. Asking for every step of the sum makes the first error visible at a glance; checking the billing toggle yourself catches the second.
Ten questions to answer from any AI pricing page
- What unit am I paying for, and how many will I use in a normal month?
- How many units does the plan include?
- What does each extra unit cost, and does overage switch on by itself?
- Is there a hard ceiling where the service stops?
- Is there a minimum number of seats or a minimum term?
- Is the displayed price for monthly or annual billing?
- Is any price promotional, and what is it after the promotion?
- Do unused credits or units roll over or expire?
- Which features need a higher tier than the one I'm looking at?
- For outcome pricing: how exactly is the outcome defined?
Answer all ten for each tool, multiply by your normal and busy-month volumes, and the comparison stops being between headline prices and becomes a comparison between real monthly bills.
Further reads
- AI Subscription Prices Compared for Small Businesses (2026) — Current list prices for the main AI subscriptions.
- How Much Does Customer Service Software With AI Cost per Agent? — Customer service tools with AI, priced per agent.
- AI Software Contracts: Auto-Renewals, Price Rises, Notice Periods — The contract terms behind the pricing page.
- What Does the ChatGPT API Cost for a Business Automation? — Token pricing worked through for a real automation.
- AI Washing: How to Spot Software That Oversells Its AI — Check the AI is real before you study its price.
- Is HubSpot Worth Paying For Just for Its AI Features? — A credit-priced product weighed against its benefits.
- 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.
- What AI Costs a Small Travel Agency and What It Saves — The monthly AI bill for a one-to-three adviser travel agency, set against a task-by-task ledger of the hours it can give back.
- 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.
- How Much Does AI Cost a Small Accounting Firm? — An itemised AI budget for three-, eight- and fifteen-person accounting firms: list prices, per-client costs, internal time and the hidden extras.
- Choosing a Mortgage CRM With AI Built In: What Actually Matters — A demo-ready checklist for mortgage CRMs with AI: document extraction, chasing, client-bank alerts, compliance logs, integrations and exit terms.
- Best AI Tools for Mortgage Brokers at Each Stage of a Case — Eight stages of a mortgage case, the AI tool worth using at each, what it costs at list price, a real-looking example of it working, and when to skip it.
- How Staffing Agencies Automate Candidate Follow-Up With AI — The seven moments where candidates go quiet, which to automate first, message templates, AI reply sorting and the consent rules that keep texts welcome.
- 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.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: vendor pricing pages and help documentation for OpenAI, Anthropic, Google, Microsoft (Copilot Studio), HubSpot, Zapier, Make, n8n, Intercom, Help Scout (checked September 2026), Zendesk, Meta and Jobber.