What to Automate First: The 5-Question Test We Run Before Touching a Workflow

A spreadsheet saved my agency eight billable hours a month. Not a clever tool, not a new model, a scoring sheet that told me in black and white which task to automate first.

That is the real question behind most searches for what to automate first. You have seen the hype on LinkedIn, Instagram and X, you believe you can bolt AI onto almost anything, and the frustrating part is that you are half right. You can automate almost anything. The trouble starts when you pick a task at random and automate that, because automating a broken process does not fix it. It just accelerates the mess.

This is the exact test we run before we build anything for a client. Five questions, a score out of 85, and a traffic light that tells you whether to build it now, plan it, or leave it alone. By the end you will be able to score your own workflows and stop guessing.

Why most businesses automate the wrong thing first

The instinct is to automate whatever irritated you most this week. The inbox that will not empty, the report that ate your Friday, the follow-up you keep forgetting. Irritation is a real signal, but it is a bad ranking system, because the most annoying task is rarely the most valuable one to fix.

Here is the trap in one line: you can automate a task and still be worse off. If the process underneath is a mess, you have not removed the mess, you have made it run faster and cost less to reproduce. If the task needed a person's judgement and you handed it to a model, you may have traded eight out of ten quality for a five and never noticed until a client did.

So the job is not to find something to automate. Almost everything can be automated. The job is to decide what to automate first, and that is a ranking problem, not a technical one. Rank it wrong and you spend weeks building the thing that changes nothing. Rank it right and the first build pays for the next three.

The gate before the test: does it even qualify?

Before you score anything, run one coarse, ten-second filter over the task. I borrowed this framing from Michael John at Sidekick, and it has earned its keep. Ask which of three buckets the task lands in:

• Revenue. Does automating it make you more money, or protect money you already have?
• Efficiency. Does it save you real time or cost?
• Risk. Does it reduce a genuine risk in the business, an error, a compliance gap, a dropped ball?

If a task lands in none of the three, you do not build it. That is the whole rule. This gate is deliberately blunt: it stops you building the wrong thing before you have spent a minute on it. Whatever clears the gate then goes into the scored test below, which tells you the order to build the things worth building.

The 5-question test we actually run

At Montaj we settled on five criteria that decide whether a task is worth automating. Each gets a score from 1 to 10, and each is weighted by how much it actually matters, so the final number is biased towards money rather than just hours. Multiply, add it up, and you get a weighted priority score out of 85.

I walk through the whole thing on camera here, spreadsheet on screen:

The five questions, and the weight each one carries:

• Revenue impact (×2.5). Score it high when automating it grows or protects revenue directly.
• Time saved (×2.0). Score it high when the task eats 7 or more hours a week right now.
• Error and quality risk (×1.5). Score it high when the manual version often makes mistakes or needs reworking.
• Feasibility (×1.5). Score it high when you can build it with off-the-shelf tools, not a custom model.
• Repetitiveness (×1.0). Score it high when it runs the same way every time, rule-based, low judgement.

Notice that revenue impact carries the heaviest weight and repetitiveness the lightest. That is on purpose. The tool is meant to bias you towards the automations that change what your business is worth, not just the ones that clear your afternoon.

1. Revenue impact

Does automating this grow revenue or protect it? Protect is the word people underrate. In a service business your utilisation rate, the share of your team's time spent on billable work, is close to everything. Move ten hours a week off admin and onto strategy or selling, and you have not only saved time, you have changed what that person is worth to the business. That is why this one is weighted heaviest. If you want the full argument on why admin quietly eats a salary, I wrote it up in the true cost of an employee.

2. Time saved

Simple: how many hours a week does this task cost you today? One hour is a 1, ten or more is a 10. Be honest about the real number, including the switching, the chasing, and the tidying up afterwards, not just the bit that feels like the task.

3. Error and quality risk

How often does the task go wrong when a human does it? Score it high when the manual version is error-prone. Moving hundreds of numbers between spreadsheets by eye, for instance, is far more fragile than people admit, and a machine is genuinely the safer option there, not just the faster one. But read this one carefully, because it cuts both ways. If your current output is a 9 out of 10 and handing it to AI would drag it down to a 5, that is a high risk to your business, and it can veto an otherwise attractive automation. Nobody wants to ship faster rubbish.

4. Feasibility

How easily can you build it with the tools you already have? If it needs a custom model wired into your own cloud stack, that is complicated, closer to a 1. If it can be done with Zapier, Make, n8n, GoHighLevel or a well-prompted Claude, that is very feasible, closer to a 10. Rough tool ladder: Zapier to start, Make when you outgrow it, n8n when you need real control.

5. Repetitiveness

Is it the same every time, rule-based, X then Y? Score it high. Does it change with every case, full of nuance and judgement? Score it low. This is worth remembering: most of the highest-value systems we build contain no AI at all. AI is not some intelligent being, it is very good at sorting information and predicting the next step, and the more identical a task is, the more cleanly plain automation handles it. Putting AI into everything is a good way to break things that already worked.

Add the weighted scores and you get a number out of 85. Then a traffic light:

• 🟢 70 or more: automate now. Top priority.
• 🟡 40 to 69: important, plan and build it, but do not drop everything.
• 🔴 Under 40: revisit later, or forget it.

Worked example: weekly reporting scored 77 out of 85

Take the task that started this. Across all our clients we used to spend around two days a week on reporting: pulling numbers from each ad account, dropping them into a sheet, writing up the analysis against the previous period, and sending each one out by email or Loom.

Here is how it scored:

• Time saved: 10. Two days a week across the business. Ten is the ceiling anyway.
• Revenue impact: high. Those are ten to sixteen hours of salary I can move onto strategy and client-facing work. In a service business that is about as valuable as it gets.
• Error and quality risk: 7. Moving hundreds of rows from A to B by eye invites small, hard-to-spot mistakes. I am not ashamed to admit we made them.
• Repetitiveness: 8 to 9. Move the numbers, compare to last week, run them through a model with a strict prompt. Nearly identical every time.
• Feasibility: 10. Reporting automation is well-trodden ground. We did it with a simple n8n workflow and our structured data.

Weighted, that came out at 77 out of 85. Verdict: automate immediately. So we did, and 95% of that task now runs itself. That is the eight billable hours a month, handed back to the work that actually grows the business.

Worked example: chasing invoices scored 66, and why we paused

Now a task that looks like an obvious win and is not. Chasing unpaid invoices.

• Time saved: 8. For a firm without a payment processor doing the collecting, this can eat serious hours.
• Revenue impact: high. A settled invoice is money in the bank. Hard to argue with.
• Error and quality risk: 4. Depends entirely on how clean your accounting data is.
• Repetitiveness: 4. And here is the catch the score surfaces. If you are a high-volume, low-value business, automated chasing is fine. If you are low-volume and high-value, with a handful of clients you have spent years building trust with, do you really want a bot sending them automated payment reminders? Probably not.
• Feasibility: 10. Tools like QuickBooks and Xero already do most of it.

Weighted, that landed at 66 out of 85. Genuinely useful, but not urgent, and the score made the trade-off visible instead of letting us blunder into it. That is the whole point of scoring: these are arbitrary numbers with real nuance behind them, and the act of assigning them forces you to see the financial and relationship impact before you spend a penny building.

How to run this on your own business

You do not need our spreadsheet to do this, though the version I walk through in the video is free and linked in the description.

1. Brain-dump 5 to 10 repetitive tasks. Follow-up emails, social posting, onboarding, invoicing, reporting, quote-writing. The more you list, the more useful the exercise. Treat it as a living document you add to over time.
2. Run each through the three-bucket gate. Revenue, efficiency, or risk. Anything that hits none, cross it off now.
3. Score the survivors on the five criteria. Be honest, especially on error and quality risk, where the veto lives.
4. Sort by weighted score and read the traffic lights. Build the greens first, plan the ambers, park the reds.
5. Sit with your team for an hour and do this properly. They know where the time actually goes better than you do.

Do this once and your automation roadmap stops being a wish list and becomes an ordered plan with money attached to each line.

The catch: the test tells you what, not whether you are ready

The matrix tells you what to automate first. It does not tell you whether your business is ready to automate at all. Automate on top of messy data or a broken process and you just speed up the chaos, which is the one mistake this whole exercise is meant to prevent.

"You can't just stick AI into a business without really understanding the challenges, the processes, and everything else." Paz B, who runs the web development firm Kickass Online, after one of our AI Week sessions.

That understanding is the foundation. Get it wrong and the best-scored automation in your list will still disappoint you.

If you want to know where your business actually sits before you build anything, book a free AI audit: we look at your real workflows and hand you a prioritised starting point, whether that is tidying your data, starting to automate, or stepping back to rethink the workflow entirely.

Final words

Almost anything can be automated. That is exactly why "can I automate this?" is the wrong question. The right question is "what should I automate first?", and that is a ranking decision you can make with a pen, an hour, and five honest scores.

Score the task, respect the veto, fix the foundation, then build. Do it in that order and automation stops being a pile of half-finished experiments and starts compounding, because the first build funds the next three. Guessing gets you busy. Scoring gets you paid.