Estimate the yearly benefit (hours saved × what an hour costs you, plus errors avoided and extra revenue you can evidence), subtract the build cost and a year of running costs, and divide by the cost. Base it on a one-week time log, not a guess, and build only if the cautious scenario pays back within about 12 months.
Two mistakes inflate most pre-build estimates. The first is counting the whole task as saved, when an automation usually removes part of it: someone still reviews drafts, handles exceptions and fixes the odd failure. The second is counting freed hours as cash. An hour saved is only money if it turns into billable work, an avoided hire or less overtime. Keep those separate and the sum gets honest quickly.
Step 1: time the task for one real week
Estimates from memory are usually wrong in both directions: people forget the small interruptions and overstate the big jobs. For one normal week, whoever does the task logs each instance on a simple sheet. It takes seconds per entry.
Date Start End Task Notes
Mon 08:10 08:22 Quote follow-up (x2) one needed revised price
Mon 16:40 16:48 Quote follow-up customer asked for call
Tue 08:05 08:35 Quote follow-ups (x4) looked up 2 in old emails
Wed -- -- none done on site all day
Thu 17:30 18:05 Quote follow-ups (x5) 3 were over a week oldThis illustrative log from a roofing contractor's owner already tells you three things: follow-ups take about 6-8 minutes each, they bunch up at the ends of days, and on busy days they don't happen at all. The last point matters as much as the minutes, because a missed follow-up may be a lost job.
Step 2: count the volume and the exceptions
Multiply up from the log carefully. If the week was unusual (school holidays, a quiet spell), use last month's records instead: quotes sent, invoices raised, enquiries received. Then count the exceptions, the cases an automation can't handle alone. For quote follow-ups, those are quotes where the customer already replied, quotes that were revised, and commercial clients who expect a call. If exceptions are more than about a quarter of the volume, the automation will save less than it appears to.
For the roofing contractor: 45 quotes a month, each followed up twice in an ideal world (day 3 and day 10), so 90 follow-ups. In practice about 40% of quotes are never followed up at all. Around 10% are exceptions that need a personal call.
Step 3: put a price on an hour
For staff, start with the hourly wage and add what an hour really costs beyond it: employer contributions, holiday, pension, equipment. Your accountant can give you the figure; without it, adding 25% to the wage is a reasonable placeholder. For the owner, use what the freed hour will actually be used for. If the roofing contractor's owner will spend it pricing more jobs, it's worth more than a wage. I've used $45 an hour in the example below.
Write down where the freed time goes before you count it. "Two more site surveys a week" is a benefit. "It frees up time" is not, because freed time with no plan tends to disappear into other small tasks.
Step 4: estimate what the automation really removes
Describe the automated version step by step and time each part that still needs a person. For the roofing contractor, the design is: each day, find quotes sent three or ten days ago with no reply; an AI step drafts a short follow-up using the job details; the draft waits in the owner's inbox for one-click approval; approved emails send and the CRM is updated.
- Manual today: 90 follow-ups a month × 7 minutes = 10.5 hours, if they were all done. Actually done: about 60% of that, 6.3 hours.
- Automated: 90 approvals × 1 minute = 1.5 hours, plus 9 exceptions × 7 minutes = about 1 hour. Total 2.5 hours.
Measured against the hours actually spent today, the saving is 6.3 − 2.5 = 3.8 hours a month, about 46 hours a year. At $45 an hour, that's roughly $2,050 of time. That is smaller than the "10 hours a month" the owner guessed, and it's the right number to use, because the automation also does the 40% of follow-ups that currently don't happen. That extra work is a revenue question, not a time saving (step 7).
Step 5: cost the build
Include everything up to the day it runs on its own:
- Design and build: your own hours at your hourly value, or a quote. The roofing contractor was quoted $1,200 by a consultant for build, testing and a handover session (an illustrative figure; get your own).
- Testing on real examples: often as long as the build. Budget it even if you're building yourself.
- Clean-up of the data it depends on: if half the quotes have no customer email, fixing that is part of the build. The data readiness checklist helps size it.
- Your time in the handover: an hour or two to learn how to adjust it.
If you're weighing doing it yourself against paying someone, the trade-offs are in hiring an automation consultant or building it yourself.
Step 6: cost a year of running it
Running costs are small per run and easy to underestimate over a year. Work them out from the design:
- Platform fees. Zapier's Professional plan starts at $19.99 a month billed annually ($29.99 monthly) for 750 tasks. Each successful action step is a task; triggers, filters and formatter steps aren't. An "AI by Zapier" step uses 1, 3 or 5 tasks per run depending on the model tier (1 task if you connect your own API key).
- AI usage, if you call a model directly. Drafting a 120-word follow-up from job details might use 800 tokens in and 200 out. On Claude Haiku 4.5, at $1 and $5 per million tokens, 90 drafts a month cost under 20 cents.
- Review time already counted in step 4.
- Upkeep: an hour a month is a fair allowance for fixing a changed login, adjusting wording or handling a failed run.
For the roofing contractor, each follow-up uses an AI step (say 3 tasks at a mid tier), creating the draft (1 task) and updating the CRM (1 task): 5 tasks × 90 = 450 tasks a month, inside Professional's 750. Year one running: $240 platform + 12 hours upkeep at $45 ($540) = $780. The broader list of post-launch costs is in AI maintenance costs after go-live.
Step 7: add errors avoided and revenue, with evidence
These lines often dwarf the time saving, which is exactly why they need evidence. Only include what you can support from your own records.
The roofing contractor checked last year's quotes: of quotes that were followed up, 30% were won; of those never followed up, 20% were won. About 18 quotes a month currently get no follow-up. If automation followed up all of them and they converted like the followed-up group, that's 1.8 extra jobs a month. At an average job of $3,000 and a 30% margin, each job adds $900 of profit.
That's up to $19,440 a year, which is precisely the kind of number that makes people build things without thinking. The records show a correlation, not proof: the quotes that were followed up may have been the keener customers. So the cautious scenario below assumes only a sixth of that uplift.
Error savings work the same way: count what the errors cost you last year, not a hypothetical. Picture an electrician's firm considering an automation that builds invoices from completed job sheets instead of retyping them. Its records show 14 credit notes last year for wrong amounts or missed items. Each took about 20 minutes to sort out, and five customers paid late while disputing, delaying about $6,000 of cash by a month on average. The error line is then 14 × 20 minutes (about 5 hours of time) plus the cost of that delayed cash, plus any items that were never billed at all, which the firm estimated from a sample of 30 job sheets: two missed call-out charges. Every one of those figures came from records, so the line can be defended.
An avoided hire is the largest line of all when it's real. A cleaning company whose office manager spends 12 hours a week on booking admin might plan to hire a part-time assistant next year. If automation removes 8 of those hours and the hire is genuinely postponed, the saving is the assistant's cost, not 8 hours at the manager's rate. But only count it if you would really have hired, and write the decision down: "we will not recruit the assistant while booking admin stays under 5 hours a week".
Step 8: run three scenarios and the payback
Put the numbers into a small table with cautious, expected and best cases. The cautious case applies a realism discount to the time saving (70% of it) and takes the low end of each assumption.
| Roofing contractor, year one | Cautious | Expected | Best |
|---|---|---|---|
| Time saved (value) | $1,435 | $2,050 | $2,050 |
| Extra jobs a month from follow-ups | 0.3 | 0.9 | 1.8 |
| Extra profit a year | $3,240 | $9,720 | $19,440 |
| Total benefit | $4,675 | $11,770 | $21,490 |
| Build (quote) | $1,200 | $1,200 | $1,200 |
| Running for a year | $780 | $780 | $780 |
| Total cost | $1,980 | $1,980 | $1,980 |
| ROI: (benefit − cost) ÷ cost | 136% | 494% | 985% |
| Payback | About 5 months | About 2 months | About 1 month |
Now the sensitivity check, which is the most useful line in the whole exercise. Remove the revenue line and the cautious case is $1,435 of benefit against $1,980 of cost: it doesn't pay back in year one on time alone. So the decision rests on one assumption, that follow-ups win extra jobs. That can be tested for almost nothing before building: the owner follows up every quote by hand, using a template, for six weeks, and compares the win rate with last year's. If it rises, build. If it doesn't, the automation is a $1,980 way to save $1,435 of time.
A copyable version of the worksheet:
AUTOMATION ROI WORKSHEET Cautious Expected Best
Volume per month ______ ______ ______
Minutes per item today (logged) ______ ______ ______
Minutes per item after ______ ______ ______
Exceptions per month x minutes ______ ______ ______
Hours saved per year ______ ______ ______
Value of an hour ______ ______ ______
Time value (x0.7 in cautious) ______ ______ ______
Errors avoided (from records) ______ ______ ______
Extra profit (from records) ______ ______ ______
TOTAL BENEFIT ______ ______ ______
Build cost ______ ______ ______
Platform + AI usage per year ______ ______ ______
Upkeep hours x value ______ ______ ______
TOTAL COST ______ ______ ______
ROI = (benefit - cost) / cost ______ ______ ______
Payback months = cost / (benefit / 12)
The one assumption this depends on: ______________________
How I'll test it before building: _________________________Asking an assistant to attack your estimate
Once the worksheet is filled in, ask ChatGPT or Claude to find its weak points. It won't know your business, but it's good at spotting optimistic assumptions. Paste the worksheet with a brief like this:
Here is my ROI estimate for an automation, with every assumption.
Act as a sceptical accountant. List the three assumptions most
likely to be wrong, what would happen to payback if each were 50%
worse, and a cheap way to test each before I spend anything.
Don't invent figures; use only mine.
[paste the filled worksheet and a one-paragraph description]For the roofing contractor, the reply might include (illustrative):
1. Extra jobs from follow-ups (0.3-1.8 a month). The biggest
lever. At 50% of the cautious figure, payback moves from about
5 to about 8 months. Test: six weeks of manual follow-ups.
2. One-minute approvals. If approvals take 2 minutes, the time
saving falls by about 40%. Test: time yourself approving
ten sample drafts.
3. Upkeep of one hour a month. Plausible, but the first two
months usually need more. Budget 3 hours a month initially.Check its arithmetic before acting on it: in this case, halving the cautious revenue gives about $3,055 of benefit and payback nearer eight months, which matches. The list of tests is the useful part. Each one costs an afternoon and replaces a guess with a measurement.
Step 9: set kill criteria before you build
Write down, before building, what would make you switch it off, and when you'll check. For the roofing contractor: at 30 days, at least 90% of drafts are approved without heavy editing; at 90 days, the follow-up rate is above 95% and the win rate on followed-up quotes hasn't fallen. If drafts need rewriting more than one time in ten, most of the time saving vanishes and the design needs changing. Deciding these numbers in advance stops sunk-cost thinking later.
A removals firm's automation that failed the test
Not every idea passes, and finding that out on paper is the cheapest outcome. An illustrative removals firm wanted to automate booking home surveys: a customer requests a quote, gets an AI-written email with a booking link, picks a slot, and the surveyor's calendar and the CRM update.
- Volume: 30 surveys a month. Logged time to book each by phone and email: 6 minutes. Total 3 hours a month.
- Automated: about 70% of customers would self-book; the rest still phone. Saving: 2.1 hours a month, 25 hours a year, at the office manager's loaded $30 an hour: $756.
- Costs: a scheduling tool plan (assume $15 a month, $180 a year), a build quote of $600, and six hours of upkeep a year ($180). Year one: $960.
- Revenue: none evidenced. Customers weren't being lost because booking was slow; the firm already replied within two hours.
Cautious benefit: $756 × 0.7 = $529 against $960 of cost. Payback in the cautious case is nearly two years. The firm instead added a free booking link to its existing quote email, without the AI step or the CRM integration, and got most of the time saving for nothing. The worksheet didn't kill the idea; it shrank it to the part worth doing.
When the estimate turns out wrong
Compare the forecast with reality at the 30 and 90-day checks, line by line. A realistic surprise from the roofing example: approvals took two minutes, not one, because the owner kept tweaking wording. That halved the time saving, but the win rate rose as hoped, so the automation still paid. The owner fixed the wording problem by adding two of his own past emails to the AI step's instructions as examples, after which edits dropped to about one in ten.
Keep the worksheet with the automation's documentation. When you next consider an automation, your own past estimates versus actuals are the best calibration you'll have. For the wider method across the business, including tools that aren't automations, see how to calculate AI ROI with a worked example, and for checking that a process is stable enough to automate before any of this, how to tell if a process is ready.
Questions about estimating automation returns
What payback period should a small business aim for?
For a small automation, I'd want the cautious scenario to pay back within 12 months, and ideally within six. Tools, prices and processes change quickly, so a payback that relies on year two or three is exposed to all of that. A longer payback can still make sense for something strategic, but decide that deliberately rather than by optimistic arithmetic.
Should I count the owner's time at the same rate as staff time?
Count it at what that hour is really worth to the business. If the freed hour goes into quoting, selling or supervising, value it at what those activities earn, which is usually more than a wage. If it goes into an evening off, that has value too, but don't present it as cash in the calculation. Keep time savings and cash savings on separate lines.
How do I estimate costs if I haven't chosen a tool yet?
Price the two or three likeliest tools at their published plans for your volume, and use the dearest in the cautious scenario. For per-task platforms, count the action steps in your design and multiply by monthly runs. Add an allowance for AI usage and a monthly hour of upkeep. If the automation only pays back with the cheapest tool, the margin is too thin.
Further reads
- Should You Pay an AI Consultant on Results? Success Fees Explained — Why paying on results needs an agreed baseline like this one.
- How to Judge Whether Your AI Consultant Delivered Value — Checking whether a paid build delivered what it promised.
- How to Stop Retyping Data Between Apps With AI Automation — A common first automation whose savings are easy to measure.
- How to Set Up Missed Call Text Back for a Service Business — A cheap automation where the revenue line is easy to test.
- How to Measure Whether Copilot Is Paying for Itself — The same discipline applied to a Copilot licence.
- Is It Worth Automating a Task You Only Do Once a Week? — The payback sum for weekly tasks, the factors beyond time that change the answer, and three worked cases: one yes, one partly, one no.
- Is AI Ticket Automation Worth It for a Small IT Support Shop? — When AI ticket automation pays for a small IT support shop, when it quietly cuts your revenue, and a 30-day test to find out.
- Is AI Marketing Automation Worth It for a Small Business? — When AI marketing automation pays off for a small business and when it doesn't yet, with a payback sum, a scorecard and three worked cases.
- How Long Does AI Marketing Automation Take to Set Up and Pay Off? — Setup times and payback windows by automation type, a four-number payback formula and a delicatessen's first seven months, month by month.
- How Much Does It Cost to Automate Lead Follow-Up With AI? — What each part of an automated follow-up costs, why the AI drafting is the smallest line, a costed furniture-maker example and the one-extra-sale test.
- Can a One-Person Business Automate Its Marketing With AI? — What a solo owner can hand to AI and automation, what should stay personal, a worked week with costs, and the checks that stop silent failures.
- How Much Does AI Visual Inspection Cost a Small Manufacturer? — Real price bands for one AI inspection station, the line items quotes leave out, and a worked payback sum for a small moulding shop.
- Predictive Maintenance With AI: Realistic for a Small Workshop? — When AI predictive maintenance pays in a small workshop, the five cheaper rungs below it, and a CNC shop costed end to end.
- n8n Self-Hosted vs Cloud: Real Costs for a Small Business — Both price lists side by side, the upkeep hours self-hosting really takes, and a roofing contractor's year costed both ways.
- Break-Even Analysis With AI: A Worked Example for a New Product — A specialty coffee roaster costs a canned cold brew launch step by step: missed costs, channel mix, scenarios and the shelf-life check that changed it.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: facts sheet for Zapier plans, task counting and AI by Zapier task usage, and API token prices. All business figures are illustrative. Checked September 2026.