Watch Yourself Work: A Three-Hour Audit That Decides What AI Should Actually Handle
Most small teams do not have an AI problem. They have a “we haven't actually looked at our own admin work since 2023” problem. Which is fine — you have been busy running the business. But if you are going to hand pieces of your week to software, you should first know what those pieces cost you.
That is all this audit is. Three hours, once. No consultants, no free trial funnel, no vendor demo with a countdown timer.
Why the audit has to come first
Buying the tool first is the expensive way to learn. You sign up for something with a nice landing page, you poke at it for two weeks, you realize it solves a problem you do not have, and then you forget to cancel until the card statement shows up in March.
Starting with the audit inverts that. You end up with a ranked list of your own repetitive work, priced in hours per month, before a single dollar leaves your account. And you will probably discover that the biggest time sink in your week is not the thing the ads are selling.
The setup: ten minutes, one spreadsheet
Grab a timer and open a blank sheet. Set these columns across the top:
Task | How often | Minutes each | Who does it | Rule you could write in plain English? | Cost of a mistake | Hours per month
One ground rule for the next three hours: you are not fixing anything today. The moment you start fixing, you stop measuring. Fixing comes in hour three.
Hour one: narrate your own work
Shadow yourself. Literally say out loud what you are doing as you do it: “Now I am pulling the numbers for the Alvarez quote. Now I am retyping them into the proposal template. Now I am emailing to ask whether they want net 15 or net 30.”
It feels ridiculous for about six minutes, then it gets useful, because you start noticing the retyping, the tab-switching, and the copy-paste that happens four times a day.
Four things show up on almost every two-person team's list:
- Quoting. Same template, different numbers, a fresh PDF every time.
- Scheduling. Three emails to land one 30-minute call.
- Invoice chasing. The polite nudge, then the less polite nudge, then the phone call.
- Review requests. You know you should ask every happy customer. You ask maybe one in five.
The sneaky ones are the tasks nobody thinks to name: copying a customer's details between your CRM and your invoicing tool, and answering the same six questions over and over in your inbox.
Hour two: score it like a skeptic
Now do the arithmetic. Frequency times minutes gives you hours per month. An invoice nudge that takes four minutes and happens thirty times a month is two hours. That is real, and it is invisible until you write it down.
Then apply two filters to each task:
- Could you write the rule in plain English? “If the invoice is 7 days past due, send the first reminder with this wording.” That is automatable. “Decide whether this client is being shady” is not.
- What does a mistake actually cost? A wrong appointment time is annoying. A wrong price on a quote is a margin hit and an awkward phone call.
| Task | Hours/mo | Rule-based? | Mistake cost | Verdict |
|---|---|---|---|---|
| Invoice reminders | 2.0 | High | Low | Test first |
| Quote assembly | 3.5 | Medium | High | Template it, don't hand it over |
| Booking back-and-forth | 1.5 | High | Low | Test |
| Review requests | 0.75 | High | Low | Test |
Anything with high hours, a clear rule, and a low cost of error is your candidate. Anything with a high cost of error stays human for now — you can revisit it in six months once you trust the workflow.
Hour three: pick one, write the acceptance test, attach a kill switch
One test. Not three. You are a two-person team, and three simultaneous rollouts means you will learn nothing about any of them.
Write down three things before you sign up for anything:
The baseline. “Invoice chasing takes me 2 hours a month.” No baseline, no proof.
The acceptance test. “If I am not getting at least 60 minutes a month back by day 30, I cancel.” Pick the number before the tool's dashboard can flatter you.
The kill switch. A calendar reminder on day 40 that says, in your own words, cancel this if it hasn't earned its keep. Most subscriptions do not die from bad products. They die from nobody remembering.
Then do the boring math: if a tool runs $45 a month and buys you 90 minutes, and your time is worth $60 an hour, you are ahead by $45. If it buys you 20 minutes, you are paying $45 for a chore. Both numbers are knowable in a month — most people just never check.
What usually wins, and what usually loses
For a shop your size, the reliable winners are invoice reminders, review request sequences, a scheduling link that ends the calendar tennis, and quote assembly from a locked template library.
The reliable losers are anything that talks to customers before you have written down your voice rules, anything that requires migrating your whole pipeline to a new system in week one, and anything pitched as a platform when you needed a button.
The guardrail against subscription creep
Three house rules keep this from turning into eleven monthly charges:
- One in, one out. A new tool earns its spot by replacing one, or by replacing a task you can name.
- Every tool gets a review date the day it is approved — not a vague someday.
- Keep a shared note listing each tool, who owns it, and what it replaced. When the owner leaves, the note survives.
The honest catch
Automation does not delete work. It moves it. You will still spend time checking the output, tweaking the wording, and noticing when the reminder email starts going to a customer who already paid.
Budget a 30-minute check once a month and the whole thing holds. Skip it and you get the familiar mess: five tools, no owners, and a nagging sense that the robots are winning.
Three hours now, thirty minutes a month after. That is a trade almost any two-person team should take.
