You signed up for an AI tool a couple of months ago. Maybe it drafts quote follow-ups, answers after-hours messages, or turns job notes into invoices. It feels like it helps. But the card gets charged every month, your office manager still spends time checking its work, and you honestly cannot say whether it is earning its keep or just feels modern.
Here is the short answer. Pick one task. Before you change anything, write down how often it happens, how long it takes, who does it, and how often it gets redone. Then add up everything the tool costs, including setup hours and the time someone spends reviewing its output. Compare the time and mistakes saved against that total each month. If the number stays negative after 90 days, stop.
Key Takeaways
- Measure one task at a time. "AI saves us time" is not measurable; "quote follow-ups take 2 minutes instead of 6" is.
- Take your baseline before you switch anything on. Afterward, nobody remembers how long things used to take.
- The cost people forget is review time. If someone checks every draft, that time comes off your savings.
- Time saved only counts if it goes somewhere: fewer paid hours, more jobs booked, or work that used to slip now getting done.
- Set a 90-day decision date in advance, and be willing to cancel.
What should you measure before you start?
You cannot say a tool saved time if you never wrote down how long the task took.
Choose one task that is repetitive, frequent, and easy to check (the first five processes worth automating is a good shortlist). Then spend one or two normal weeks writing down four things:
- How many. How many times does this task happen per week? Count them, do not guess.
- How long. How many minutes does each one take, start to finish? Time a handful with your phone and average them.
- Who. Who does it, and what does an hour of their time cost you? Use their hourly wage plus a bit for payroll taxes. If it is you, use what your hour is worth to the business, not zero.
- How often it goes wrong. How many get redone, corrected, or forgotten? A salon that sometimes forgets to send the cancellation-policy reminder, or a plumbing shop that quotes the wrong after-hours fee, should count those.
Write it somewhere you will find in three months.
What does an AI tool really cost?
The subscription is only part of the cost. Add up all of these:
- The subscription. Monthly fee, plus any per-message or per-use charges. If the price depends on usage, look at your actual bill, not the plan page. Our guide to AI automation pricing for small businesses walks through the common pricing models.
- Setup hours. The time someone spent signing up, connecting accounts, writing instructions, and testing. One-time, but real.
- Review time. This is the one people forget. If your office manager reads every AI-drafted email before it goes out, those minutes belong on the cost side.
- Upkeep. Updating instructions when prices change or you add a service. Estimate hours per month.
- Cleanup. When the tool gets something wrong and a customer notices, the time spent fixing it counts too.
A worked example (replace these numbers with yours)
Every number here is an example to show the math, not a prediction. Swap in your own.
Say a small HVAC shop sends a follow-up email on every quote that has not been answered after three days.
Baseline (before AI):
- 30 follow-ups a week
- 6 minutes each to look up the quote and write a personal note
- Done by the office manager, who costs the shop $30 an hour
- 30 x 6 = 180 minutes a week, or 3 hours
- About 13 hours a month (3 hours x 4.33 weeks), worth about $390
After (AI drafts, office manager reviews and sends):
- Still 30 a week
- 2 minutes each to read, adjust, and send
- 30 x 2 = 60 minutes a week, or about 4.3 hours a month, worth about $130
Time value saved: $390 minus $130 = $260 a month. Notice the review time is already subtracted here.
Costs:
- Subscription: $50 a month
- Upkeep: 1 hour a month at $30 = $30
- Setup: 6 hours once, at $30 = $180
Monthly net: $260 saved minus $80 in monthly costs = $180 a month.
Payback on setup: $180 setup divided by $180 a month = about 1 month.
ROI over the first 3 months: benefit of $780 (3 x $260), total cost of $420 (3 x $80 plus $180 setup). Net gain of $360. $360 divided by $420 is about 86%.
In this example, the tool pays off. Change one assumption and it might not. If review took 5 minutes instead of 2, the monthly saving would shrink to about $65, which is less than the $80 the tool costs each month. That is why you measure.
The simple formula
Three lines cover it:
- Monthly net = (hours saved per month x hourly cost) + value of other gains, minus (subscription + upkeep)
- Payback in months = setup cost divided by monthly net
- ROI % = (total benefit minus total cost) divided by total cost, x 100, for whatever period you choose
"Hours saved" means hours after review time is subtracted. "Other gains" is the next section.
One honest caveat: saved hours are only money if they turn into something. If the freed time goes to calling back leads, catching up on invoices, or avoiding a part-time hire, it counts. If it just disappears, you saved nothing. Write down where the time went.
What should you measure besides time?
Sometimes the real win is not minutes but the things that used to slip. Track these alongside time:
- Missed follow-ups. How many quotes, callbacks, or rebooking messages went unsent per month before, and how many now?
- Response time. How long does a new inquiry wait before someone answers? Compare a handful of examples from before and after. An after-hours lead answered in minutes instead of the next morning may book with you instead of the next shop.
- Mistakes caught. Wrong prices quoted, a missing deposit line, an onboarding checklist step skipped for a new agency client. Count them before and after, plus any mistakes the AI introduced.
- Redo rate. How often does a finished task need to be done again?
Some of these will not convert neatly to dollars. That is fine: a count that went from "we miss a few every week" to "we missed one this month" is a real result you can weigh.
The worksheet
Copy this table into a note or spreadsheet. Fill in the Baseline column before you start, then add a column at each check.
| Item | Baseline | Day 30 | Day 60 | Day 90 | |---|---|---|---|---| | Task being measured | | | | | | Times per week | | | | | | Minutes per task (including review) | | | | | | Who does it, and hourly cost | | | | | | Hours per month (times x minutes x 4.33 / 60) | | | | | | Redone or corrected per month | | | | | | Missed or forgotten per month | | | | | | Typical response time | | | | | | Mistakes caused by the AI | | | | | | Subscription and usage fees this month | | | | | | Upkeep hours this month | | | | | | Setup hours (one time) | | | | | | Monthly net (saved minus costs) | | | | | | Where did the saved time go? | | | | |
How do you run a 30/60/90 day check?
Put three dates on your calendar the day you turn the tool on.
Day 30: Is it working at all? Expect the first month to be messy. The question is not "did it pay off" but "is it heading the right way?" Is review time dropping? Is the redo rate reasonable? Fix the instructions where it keeps getting things wrong.
Day 60: Has it settled? Fill in the worksheet and run the formula. If the monthly net is close to zero, look at review time first.
Day 90: Decide. Keep it, change it, or cancel it. Decide on the numbers in the worksheet, not on how the tool feels.
When is the honest answer "stop"?
Some tools do not pay off for some businesses. Cancel, or go back to doing it by hand, when you see any of these at day 90:
- Monthly net is still negative, and review time has stopped dropping. It is not going to fix itself.
- Someone is quietly redoing the work. If your team rewrites most drafts from scratch, the tool is adding a step, not removing one.
- Mistakes it causes cost more than mistakes it catches. One wrong price sent to a customer can wipe out a month of saved minutes.
- The task was too rare to matter. Saving 4 minutes on something that happens twice a month will never cover a subscription.
- Nobody can say where the saved time went. If the hours were not used for anything, there was no return.
Stopping is a result, not a waste. You now know that task is better done by a person. For more on that call, see AI vs. human: which business tasks to automate.
Where Mycel fits
Mycel does not run your automations and does not calculate ROI for you. It is the memory your business runs on: a shared vault where your team keeps policies, prices, how-we-do-it notes, and spreadsheets in plain folders. That makes it a sensible home for your baseline worksheet and the notes from each 30/60/90 check, so the numbers are still there when you need them in three months.
If you also use the Martello apps (Relay, TradesBackbone, ReputeLift, SynthPrism), what happens in them gets filed into that memory, and each morning the Today view shows what needs a person, with a link straight to it. Anything that would change something inside another app waits for your approval. See how Mycel works or check pricing; there is a free tier.
Frequently Asked Questions
How do I calculate ROI on an AI tool for my small business?
Take a baseline of one task (how often, how long, who does it, and their hourly cost), then measure the same things after the tool is running. Monthly net is hours saved times hourly cost, plus any other gains, minus the subscription and upkeep. ROI is total benefit minus total cost, divided by total cost.
What costs do people forget when measuring AI automation?
Review time is the big one. If someone checks every AI draft before it goes out, those minutes come off your savings. Setup hours, ongoing upkeep, usage-based fees, and time spent fixing the tool's mistakes are also easy to leave out.
How long should I wait before deciding if an AI tool is worth it?
Give it about 90 days, with checks at 30 and 60. The first month is usually slower because instructions need adjusting and people are still learning to trust the output. By day 90 you should have stable numbers to decide on.
Is AI automation worth it for a small business?
It depends on the task, not on AI in general. Frequent, repetitive, easy-to-check tasks are the most likely to pay off. Rare tasks, or tasks where the output needs heavy rewriting, often do not. The only way to know for your business is to measure one task before and after.
How do I measure benefits that are not time savings?
Count them. Track missed follow-ups, response time to new inquiries, mistakes caught, and redo rate before and after. If you know roughly what a missed follow-up or a wrong quote costs you, you can turn the count into dollars; if not, the before-and-after count is still a useful result.
Your next step this week
Pick the one task you most want to hand to AI. Before you sign up for anything, count it for one week: how many times it happens, how many minutes each takes, and who does it. Fill in the Baseline column of the worksheet above. That one column keeps every later decision honest.