AI Fundamentals
How to Audit Your Business for Automation Opportunities
A one-week method for finding what is genuinely worth automating, using a notebook rather than a consultant, and ranking it honestly.
· 8 min read · Flowmatix
The short answer
Log every recurring task for one week with who does it, how long it takes and how often. Then score each on volume, rules-clarity and consequence of error. Automate high-volume, rule-clear, low-consequence work first. Most businesses find their biggest time sink was never on the original list.
Most automation projects begin with a solution someone read about. The better ones begin with a week of boring observation. This is the method, and it needs a spreadsheet rather than a consultant.
Step one: log, do not estimate
For five working days, everyone logs recurring tasks. Four columns: what it was, how long it took, how many times, and what triggered it. No judgements yet, no proposed fixes.
Estimation is the enemy here. Ask an owner where their team's hours go and they will name the visible annoyance. The log almost always surfaces something duller and larger — usually re-entering the same information into a second system.
Step two: score each task on three axes
| Axis | Question | Good candidate |
|---|---|---|
| Volume | How many times a week does this happen? | Many, and predictable |
| Rules clarity | Could you write the decision as if-then rules? | Yes, in under ten lines |
| Consequence of error | What happens if it is done wrong and nobody notices? | Recoverable, and visible |
High volume, clear rules, low consequence is the sweet spot and it is where every first project should live. High consequence work can absolutely be automated later — it just should not be the thing you learn on.
Step three: check it should exist
Before automating anything, ask what happens if this stops. A meaningful share of recurring work is a report nobody reads, a check that duplicates another check, or a step that survived a process change three years ago. Deleting is free and permanently cheaper than automating.
Step four: separate the three kinds of work
- Moving data between systems. Almost always automatable, usually the biggest single sink, and rarely the thing anyone complains about.
- Making a judgement call. Automatable only where the judgement is genuinely rule-based. Where it is experience, keep the human and automate the preparation.
- Talking to people. Automate the routine acknowledgement and the scheduling, never the parts where being human is the point.
Step five: rank by hours, not by irritation
Multiply time per task by frequency to get hours per month, and sort. The most annoying task is often not the most expensive one. Fixing the irritating five-minute job feels better and returns less than fixing the tedious forty-minute one nobody mentions.
Step six: pick exactly one
Take the top item, confirm it is rule-clear and low-consequence, and do that alone. Measure the baseline before you change anything, because after go-live nobody remembers what it used to take. Run it for a month before starting the next one — adoption fails from too many simultaneous changes far more often than from bad tooling.
What a good audit output looks like
One page. A ranked list of recurring tasks with hours per month attached, three marked as candidates, one marked as first, and a measured baseline for that one. If your audit produced a forty-slide deck, it was a sales document.
Frequently asked questions
How long does an automation audit take?
One week of logging and roughly half a day of analysis. The logging cannot be shortened without losing the point — the value comes from measuring real time rather than estimating it from memory.
What should a small business automate first?
The highest-volume task with clear rules and recoverable errors. In practice that is usually moving data between two systems that do not talk, or acknowledging and routing inbound enquiries.
Do we need a consultant to run an audit?
No. The method is a spreadsheet and a week of honest logging. Outside help is worth paying for at the build stage, or when nobody internally has the authority to delete a step that should not exist.