Automating one workflow with AI usually costs $0 to $100 a month in software and AI usage, plus a one-off build. Built yourself, it takes 10 to 30 hours of your time; hired out, Contra's hiring page puts simple two- or three-app automations at $500 to $1,500 and complex multi-step workflows at $2,000 to $10,000 or more.
The AI model is rarely the big number. For most small-business workflows it costs cents or a few dollars a month. The money goes on the automation platform's pricing tier, the build, and the staff time spent checking outputs. Get those three right and the workflow can pay back within months; get them wrong and an automation that looked like $30 a month ends up needing a plan several times that price.
Every cost line in one AI workflow
Here is the full list, split the way your accounts will see it. The platform prices are list prices in USD from the vendors' own pricing pages.
| Cost line | Type | Typical range | Notes |
|---|---|---|---|
| Mapping the process and choosing the tools | One-off, internal time | 2 to 6 hours | Watching the job done, counting volumes |
| Building the workflow | One-off | 10 to 30 hours of your time, or a freelancer's or agency's fee | Depends on the number of apps and branches |
| Preparing test examples and testing | One-off, internal time | 3 to 8 hours | 30 to 50 real cases, scored |
| Automation platform | Monthly | Zapier Professional from $19.99 a month billed annually or $29.99 monthly for 750 tasks; Make from $9 a month for 5,000 credits; n8n, which bills in euros, from 20 a month billed annually for 2,500 executions | Priced by tasks, credits or executions |
| AI model usage | Monthly | Cents to a few dollars at small-business volumes | Paid per token, or as extra tasks or credits on the platform |
| Checking outputs | Monthly, internal time | Often 1 minute per item where a person approves | Usually the largest ongoing cost |
| Fixes when connected apps change | Occasional | A few hours a year for a simple workflow | Or a support arrangement with the builder |
The platforms count usage differently, which matters more than their headline prices. Zapier bills per task, meaning each successful action step; triggers, filters, Paths and Formatter steps are free. An AI by Zapier step counts as 1, 3 or 5 tasks per run depending on the model tier you pick (1 with your own AI provider key), and it needs the Professional plan or above. Make bills in credits: most actions use one, while its own AI modules use a variable number based on tokens, and if you connect your own AI key, Make charges per operation and your AI provider bills the tokens. n8n counts one execution per whole workflow run, however many steps it has. A side-by-side on features is in Zapier vs Make vs n8n for AI automation.
A homeware brand's returns workflow, costed line by line
An online homeware brand gets about 250 returns-related emails a month. Each one takes a member of staff around six minutes: find the order, check it's within the 30-day window, decide whether the item is returnable, write the reply, log it. The workflow to automate: when a returns email arrives, an AI step pulls out the order number, item and reason; the workflow looks up the order date in the store; a date check marks it in or out of the window; a draft reply is saved for staff to approve; and the case is logged in a sheet.
Running it on Zapier
Per email: the trigger is free; the AI step, using the company's own AI key, is 1 task; finding the order in the store is 1; the date check (Formatter) and the in-or-out branch (Paths) are free; creating the draft is 1; logging the row is 1. Four tasks per email, 1,000 a month. That's over the 750 in Zapier's entry tier, so the choice is the 1,500-task Professional tier at $58.50 a month on monthly billing, or staying on the 750-task tier and paying for overage. Zapier's own plans page puts overage tasks at 1.25 times your plan's per-task rate once you opt in to pay-per-task billing, which for 250 extra tasks on the 750-task tier comes to about $12.50 on top of $29.99. Staying put and paying the overage, roughly $42.50 a month, beats the $58.50 upgrade until volume climbs further.
Running it on Make
Per email: five modules at one credit each with the brand's own AI key, so 1,250 credits a month. Add the inbox trigger: a scheduled check uses a credit every time it runs, even when nothing has arrived, so checking every 15 minutes costs about 2,880 credits a month on its own. The total, about 4,130, still fits inside the 5,000 credits of Make's $9 plan. The same workflow costs about a sixth as much to run, in exchange for a builder who is comfortable with Make's more technical editor.
The AI usage itself
Each email plus instructions and the returns policy is about 2,000 input tokens, with about 300 tokens of output. For 250 emails: 500,000 input and 75,000 output tokens a month. On Claude Haiku 4.5 ($1 per million input tokens, $5 per million output) that is about 88 cents; on OpenAI's gpt-6-luna ($0.10 and $0.50) about 9 cents. The model is the smallest line on the bill by a distance. Note that a ChatGPT or Claude subscription doesn't cover this: API usage is billed separately, per token, on the developer platform. The per-token sums in more detail are in what the ChatGPT API costs for a business automation.
The prompt inside the AI step
The AI step does one narrow job, which is what keeps it cheap and checkable:
Extract from this customer email: order_number (digits only, or
"none"), item (as the customer describes it), reason (one of:
damaged, wrong item, changed mind, doesn't fit the space, other),
and wants (refund, exchange, unclear). Return JSON only.
If more than one order or item is mentioned, set needs_human to
true. Email: [email text]Given the email "Hi, the blue vase from order 48213 arrived chipped, can I swap it for the green one?", it returned this illustrative result:
{"order_number": "48213", "item": "blue vase", "reason": "damaged",
"wants": "refund", "needs_human": false}The model read "swap" as a refund. Adding "swap, change or replace means exchange" to the instructions fixed it, and the 40-email test set caught the problem before any customer saw a wrong reply. Small wording fixes like this are where most build hours go, and they are why testing time belongs in the budget alongside the build.
Build, testing and review time
The brand's operations lead built it on Make in about 18 hours spread over three weeks, plus 3 hours mapping and 5 hours testing on 40 held-back emails. Once live, staff approve each draft in about a minute: roughly 4 hours a month instead of 25. At an illustrative staff cost of $25 an hour, that's about $525 a month of time back, against running costs of about $10 on Make. The 26 hours of set-up time, worth about $650 at the same rate, are recovered in the second month.
Three budgets for the same workflow
The same returns workflow can be delivered three ways. The hired figures sit inside the ranges on Contra's hiring page for Zapier freelancers ($500 to $1,500 for simple automations, $2,000 to $10,000 or more for complex multi-step ones, $1,000 to $5,000 a month for retainers); they are illustrations, not quotes.
| Built yourself on Make | Freelancer builds on Zapier | Agency build plus retainer | |
|---|---|---|---|
| One-off build | About 26 hours of your team's time | $1,500 | $4,000 |
| Platform and AI usage per month | About $10 | About $60 | About $60 |
| Support per month | Your own time when it breaks | Ad hoc, paid by the hour | $1,000 retainer |
| First-year cash cost | About $120 | About $2,220 | About $16,720 |
The retainer column is the one to question. For a single, stable workflow, $1,000 a month buys far more support than it needs; retainers make sense when there are several workflows, frequent changes or customer-facing AI that needs watching. For one workflow, a clear handover plus paying for fixes when needed is usually enough. Whether the spend is justified at all comes down to the return, and calculating the ROI of an automation before you build it gives the method.
Building it yourself or paying someone: the break-even sum
The do-it-yourself column looks cheapest because your own time never arrives as an invoice. Put a value on it. The homeware brand's 26 hours of set-up at an illustrative $25 an hour is $650 of staff time. Had the owner built it, valuing their own time at $60 an hour, the same 26 hours would be worth $1,560, slightly more than the freelancer's illustrative $1,500 build, before counting the learning curve. Three questions usually settle it:
- Will you build a second and third workflow? If so, learning the platform pays back across all of them.
- Is the workflow customer-facing or tied to money? Mistakes cost more there, and experience is worth paying for.
- Who fixes it at 7am when it breaks? If the answer is "whoever built it", make sure that person will still be around.
A furniture maker, for instance, might sensibly pay for its first quote-drafting workflow, then build the simpler follow-up reminders itself, using the first one as a pattern.
Where the monthly bill creeps up
- Premium AI tiers multiply tasks. An AI by Zapier step on the top model tier counts 5 tasks per run. At 1,000 runs, that's 5,000 tasks for one step.
- Overage on monthly plans is expensive. At 2.5 times the base rate, regular overage is a sign you need the next tier, or annual billing. Hit triple the plan's task allowance and Zapier stops running Zaps altogether.
- Frequent polling eats credits. On Make, polling every 15 minutes is roughly 2,880 checks a month, and each check costs a credit: close to three times the free plan's entire allowance.
- Premium connectors. In Power Automate, many popular connectors are premium and need a higher licence, which can change the cost of an otherwise "included" automation.
- AI features priced per outcome. If the workflow hands off to a helpdesk or CRM AI feature, check its unit: Intercom Fin, for instance, charges $0.99 per resolved outcome, and HubSpot's AI uses credits at $10 per 1,000.
- Review time that never shrinks. If staff still rewrite half the drafts after a month, the automation is costing more time than planned. Fix the prompt, or narrow what it handles.
A realistic mistake shows several of these at once. A subscription box company built a Zap that ran an AI classification step on every email reaching the support inbox, using the top model tier. Newsletters, delivery notifications and spam all went through it: about 2,400 emails a month at 5 tasks each, before any other step. The account hit its limit in the second week and started billing overage. The fix cost nothing: a Filter step before the AI step (filters use no tasks) so only emails from customers reach it, and a cheaper model tier once testing showed it sorted just as accurately. Task use fell by more than three-quarters.
When the AI model does become the big cost
Model costs grow with the amount of text, so long documents, images and high volumes change the picture. Say a business wants to extract figures from 2,000 supplier PDFs a month, each around 10,000 tokens once read in. That's 20 million input tokens. On Claude Opus 5.5 ($4 per million input tokens) the input alone is $80 a month; on Claude Sonnet 5 ($2), $40; on Claude Haiku 4.5 ($1), $20. Anthropic's Batch API halves those prices for work that doesn't need an instant answer, which suits overnight document runs.
The rule that follows: test the cheapest model tier first, on your own held-back examples, and only move up if it fails. For sorting, extracting and drafting routine messages, the smaller models are usually enough, and the difference between tiers can be a factor of ten or more.
A quick estimate worksheet
Fill this in before you ask anyone for a quote. Here it is completed for a food truck business that wants office lunch pre-orders from a web form confirmed automatically, with an AI step that turns free-text orders ("2 x chicken wrap, one no chilli") into a clean order line.
Runs per month: 120 pre-orders
Steps per run (paid ones): 4 (AI tidy-up, add to sheet,
confirmation email, kitchen list)
Platform units per month: 120 x 4 = 480 credits on Make
Trigger checks per month: hourly checks = about 720 credits
Total: about 1,200 credits
Free plan limit: 1,000 credits -> over by ~200
Options: check every 2 hours (about 360)
-> about 840 credits, fits free
or Make's $9 plan
AI tokens per month: 120 x (800 in + 100 out)
= 96,000 in, 12,000 out
-> well under $1 on a small model
Build: about 8 hours of own time
Review: owner glances at the kitchen list
each morning, 5 minutes
Monthly cash cost: $0 to $9 plus cents of AI usageThe worksheet also exposes the trade-off: checking every two hours keeps it free, but an order placed at 10:05 might not be confirmed until nearly noon. For lunch orders, the $9 plan with more frequent checks is the better buy. That is the kind of decision the arithmetic should make visible before anyone builds anything.
Keeping the running cost down without cutting corners
- Put free steps first. Filters, branching and formatting steps use no Zapier tasks, so screen out irrelevant items before any paid or AI step runs.
- Use the smallest model that passes your test set. Keep the held-back examples so you can re-test cheaply if you switch.
- Match polling to need. A trigger that checks every 15 minutes for a job that can wait an hour is paying for impatience.
- Move to annual billing once the workflow is stable. Zapier's annual Professional price is a third lower than monthly; don't commit before the workflow has settled.
- Keep accounts in the business's name. It avoids paying twice when someone leaves and the workflow has to be rebuilt.
- Budget for upkeep honestly. Connected apps change and workflows need occasional fixes; what you pay after an automation goes live sets out realistic numbers.
If you're building on Zapier yourself, how to add AI steps to Zapier shows the classify, summarise and draft steps used in the returns example.
Further reads
- When Does Zapier Get Too Expensive? Finding the Tipping Point — Spot the volume at which your platform choice needs revisiting.
- Hidden Costs of AI Implementation Most Small Businesses Miss — The costs beyond one workflow that owners tend to miss.
- What Connecting Two Business Apps Really Costs: Four Options — What plain app-to-app connections cost without AI.
- How to Hire an AI Automation Freelancer on Upwork or Fiverr — How to hire someone to build the workflow for you.
- AI Workflow Automation Examples: 20 Processes to Automate First — Twenty candidate workflows to cost with this method.
- How Much Does Automated Invoice Processing Cost per Invoice? — The same sums applied to invoice processing.
- 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.
- How to Brief a Developer on a Custom AI Workflow You Need — The eight parts of a developer brief for an AI workflow, a copyable requirements template, and a worked example from a dog-grooming salon.
- AI vs Rule-Based Automation: Which Does Your Task Need? — A simple test for choosing rules, AI or both, a side-by-side comparison, the hybrid pattern that keeps AI small, and a dry cleaner's jobs sorted.
- How Much Does an AI Consultant Cost for a Small Business? — Published AI consultant rates from three marketplaces, the costs that sit outside the fee, and three first-year budgets for a small brewery.
- How Soon Should AI Pay for Itself? Payback Periods by Project — Payback targets for seven kinds of AI project, the sum with ramp-up months left in, and a toy shop testing four ideas against them.
- How Much Does a Fractional Operations Manager Cost? — What published rate guides say fractional COOs charge, how a part-time operations manager or project compares, and a yoga studio's three priced scenarios.
- How to Audit Your Workflows for AI Opportunities Yourself — A six-stage AI opportunity audit you can run yourself in about eight hours: inventory, timing, sorting, scoring, desk tests and a ranked register.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Zapier pricing page and help on pay-per-task billing; Make pricing page and credits help; n8n pricing; Anthropic and OpenAI API pricing pages; Contra hiring page for Zapier freelancers (checked September 2026).