How to Find What to Automate First: A 5-Day Audit You Can Actually Run

Nearly every automation conversation I have starts the same way: someone has seen a demo, they’re excited, and they want to know which tool to buy. That’s the wrong end of the problem. The tool is the easy part. Picking the right task is what decides whether the project saves ten hours a week or becomes another thing nobody uses.

Here’s the audit I run before I build anything. You can do it yourself in a week, with a spreadsheet and no budget. It works whether you’re three people or three hundred, and it will tell you more about your business than most consultants will.

The five-day log

For five working days, every time someone on your team does something they’ve done before, they log one row:

What they did	What started it	Who did it	Minutes	Tools touched
Copied order details into the accounting sheet	New Shopify order email	Rima	4	Gmail, Sheets
Answered “where’s my order?”	Customer WhatsApp	Rima	3	WhatsApp, Shopify
Chased an unpaid invoice	Weekly manual check	Karim	12	Xero, Gmail

Five columns, nothing more. Ask for it to be honest rather than complete — a log that captures 70% of reality is far more useful than a perfect one nobody fills in.

Two rules make this work. First, log the trigger, not just the task. “What started it” is the single most valuable column, because that’s what a workflow will eventually listen for. Second, log small things. The four-minute jobs are the ones people forget to mention in meetings, and they’re usually where the hours actually go.

Getting your team to actually do it

This is where the audit usually dies, so it’s worth being direct about it. If people think you’re measuring them, you’ll get a log that makes everyone look busy and productive, which is useless. Say plainly that you’re looking for work the computer should be doing, not for people who are slow. It helps to log your own work too, visibly.

Keep the friction low. A shared spreadsheet open in a browser tab beats any dedicated tool. Better still, log at two fixed points in the day — after lunch and before leaving — rather than expecting people to record things as they happen. Recall over five hours is imperfect but good enough, and it’s the difference between a log that gets filled in and one that doesn’t.

Turn the log into a number

At the end of the week, group identical rows and calculate monthly minutes:

times per month × minutes each = minutes lost per month

That four-minute order copy, done 60 times a month, is four hours. The 45-minute monthly report is 45 minutes. People almost always assume the report is the bigger problem. It isn’t, and the number is what proves it.

There’s a second cost the log won’t show you directly, and it’s often larger than the first. A task that interrupts someone forty times a month doesn’t just cost four minutes each time — it costs the concentration on either side of it. Two tasks can consume identical minutes while one of them quietly wrecks somebody’s afternoon. When two candidates score similarly, automate the interrupting one first.

Then filter with three questions

Volume alone doesn’t make something automatable. Ask three things about each task at the top of your list:

Is it rule-based? Could you write the decision down as if-this-then-that, without using the word “usually”? If the answer depends on someone’s judgement of the situation, it’s not a clean automation candidate — but it might be an assist candidate (more on that below).

Is the input predictable? A form submission with five fixed fields is predictable. A customer email that might contain an order number, or might not, is not. You can still automate the second one, but it costs more and needs a fallback.

What does a mistake cost? Sending a duplicate internal notification costs nothing. Sending a wrong invoice to a client costs trust. High-cost tasks need review steps, which changes the design — and sometimes changes the answer to “should we automate this at all.”

Sort into four boxes

• High volume + rule-based → automate now. This is where your first project lives. Order syncing, lead routing, invoice reminders, data moving between two tools.
• High volume + judgement → assist, don’t replace. Draft the reply and let a human send it. Summarise the ticket, don’t close it. This is where AI actually earns its keep.
• Low volume + rule-based → template it. A saved reply or a checklist solves this for free. Don’t build a workflow for something that runs twice a month.
• Low volume + judgement → leave it alone. Seriously. This is where automation budgets go to die.

What this looks like with real numbers

Here’s a composite of a fairly typical first audit — a small eCommerce business, four people, selling physical products.

Task	Per month	Minutes each	Hours/month	Verdict
Copy order into accounting sheet	240	4	16.0	Automate now
“Where’s my order?” replies	150	3	7.5	Automate now
Chase unpaid invoices	20	12	4.0	Automate now
Write product descriptions	12	25	5.0	Assist with AI
Handle a damaged-item complaint	6	30	3.0	Leave alone
Monthly sales report	1	45	0.75	Template it

Notice what the numbers do to the intuition. Before this exercise, the owner was certain the monthly report and the complaint handling were the problem — those are the tasks that feel heavy. They’re the bottom two lines. The thing quietly eating two full working days a month was a four-minute copy-paste nobody had ever mentioned.

That top line is also the easiest thing on the list to build. That combination — biggest cost, lowest difficulty — is what you’re hunting for, and you will almost never find it by discussion alone.

The things I tell people not to automate

I say this on every intro call, so I’ll say it here too. Don’t automate a process nobody agrees on — if two people do the same job differently, automating it just picks a winner without a discussion. Don’t automate something that changes every quarter; you’ll spend more time maintaining it than doing it. And don’t automate a broken process. Automation doesn’t fix a bad process, it just makes it fail faster and more consistently.

Fix the process on paper first. Half the time, that alone gets you most of the hours back and the automation becomes much simpler to build.

One more: don’t automate something only one person understands, without writing it down first. If the workflow encodes knowledge that exists nowhere else, you’ve moved a bus-factor problem rather than solved it.

“We’re too small for this” and “we’re too messy for this”

These are the two objections I hear most, and they’re both usually wrong, for opposite reasons.

Too small: automation economics work on repetition, not headcount. A one-person business doing the same 4-minute task 200 times a month has exactly the same 13 hours to reclaim as a ten-person one. If anything the case is stronger, because that one person has no colleague to hand it to.

Too messy: mapping the process is step one, so “we’re not ready” collapses on inspection. You don’t need a documented, agreed process to start the log — the log is how you discover what the process actually is. Businesses often find the audit itself is the valuable part, because it’s the first time anyone has seen the whole operation written down in one place.

Start with one, and measure it

Pick the single highest-volume, rule-based task from your list. Before you build anything, write down three numbers: how long it takes today, how often it runs, and how often it currently goes wrong. That’s your baseline, and you will want it later — memory is generous about how bad things used to be.

Automate just that one thing. Then measure the same three numbers after two weeks of real use, and add a fourth: how much time you spent fixing the automation. That last one is the honest test. A workflow that saves 16 hours a month but needs two hours of babysitting is still an excellent trade. One that needs ten is a liability, and you want to know that early, on a small project, rather than after you’ve automated six things.

One working automation with a real number attached will do more to convince your team than any demo. And you’ll have a log full of the next ten candidates, already sorted.

Where to go next

Once you’ve picked the task, the next decision is how the workflow should start — which sounds trivial and isn’t. That choice affects speed, running cost and reliability more than anything else you’ll do, and it’s covered in webhook, poll or schedule. After that, the thing that decides whether your automation survives contact with reality is how it handles failure.

If you’d rather not run the audit alone, that’s essentially what the first session of any project with me looks like — see how an engagement runs, or book a free intro call and we’ll do the first pass together. If you already know the work is there, n8n workflow automation is usually where it ends up.