Which process should you automate first? A decision framework

The most important automation decision is the first one, and almost everyone gets it wrong.

They pick the biggest, most painful process — the tangled, cross-department mess that causes the most grief — because it hurts the most and fixing it feels like the point. That is precisely the wrong choice, because the biggest mess is the hardest to automate, the slowest to show results, and the most likely to fail. It burns your credibility and your momentum on the exact process most likely to defeat you.

Your first automation is not about solving your biggest problem. It is about proving the approach works, fast, so you get to do a second one. Choose accordingly.

Score candidates on three axes

For each process you might automate, rate it on three things.

1. Pain — how much does this hurt today?

Real, felt pain that people would thank you for removing. Not theoretical inefficiency — actual weekly grief. High pain means motivated users and obvious value.

But pain alone is a trap, which is why it is only one axis. The most painful process is often the most painful because it is the most complex, and complexity is what kills first attempts.

2. Clarity — how well-defined is this process?

Can you describe exactly how it works, start to finish, without hitting "well, it depends" every third step? A process with clear rules automates cleanly. A process full of exceptions and judgement calls does not — you will spend your whole first project discovering that nobody actually agrees how it works.

High clarity is the most underrated axis and the best predictor of a smooth first win. If two people describe the process differently, clarity is low, and you have found an organisational disagreement to resolve before any software helps — see what AI still cannot do in ERP.

3. Blast radius — what happens if it goes wrong?

If this automation misbehaves, what is the damage? A process that sends internal reminders has a tiny blast radius — a wrong reminder is mildly annoying. A process that pays suppliers or emails customers has a large one. For your first automation, you want a small blast radius, so that the inevitable early wrinkles are cheap to fix and do not scare everyone off.

The rule

Your first automation should be high pain, high clarity, low blast radius.

• High pain so people care and the win is visible.
• High clarity so it actually works without a discovery saga.
• Low blast radius so early mistakes are cheap and safe.

This combination is your sweet spot. It is almost never the process you first thought of, because the process you first thought of is high-pain-low-clarity-high-blast-radius — the biggest mess. That one comes later, once you have proven the approach and built the muscle.

What tends to fit

Processes that usually score well on all three:

• Chasing — overdue invoices, missing documents, unsubmitted timesheets. High pain (nobody does it), high clarity (the rules are simple), low blast radius (a reminder cannot break much). Often the ideal first automation, and the reasoning is in getting paid faster with AI.
• Approval routing for a well-understood approval — a purchase, a leave request. Clear rules, real pain, contained risk.
• A single tracker replacing a chaotic spreadsheet — enquiries, jobs, requests. Clear, painful, low-risk.
• Document capture for a high-volume document type — invoices, receipts. Tedious, well-defined, and errors are caught at a review step.

What to avoid first

• The cross-department epic. The end-to-end process spanning four teams. Maximum pain, minimum clarity, large blast radius. This is your eventual prize, not your opener.
• Anything moving money unsupervised. Large blast radius by definition. Automate around it first, build trust, then approach it carefully — never as your first move.
• A process nobody agrees on. If clarity is low, you are not automating; you are refereeing a disagreement, and no software settles it.

Why the first one matters so much

Because momentum is the actual scarce resource, not budget.

Get a clean, fast first win and three things happen: people believe, they bring you the next candidate, and you have learned how to do this. The second automation is easier and better-chosen than the first. You are on a curve.

Pick a hard first project that drags for months and stumbles, and the opposite happens: people conclude "this doesn't work here", the sponsor stops showing up, and there is no second automation. The approach was fine; the sequencing killed it.

This is the same logic as the one-workflow start and the reason ERP projects that lead with data migration stall. Start where you can win, win visibly, and let the wins compound.

A five-minute version

List your candidate processes. For each, score pain, clarity and blast radius from one to five. Rank by pain plus clarity, minus blast radius. Start at the top.

You will notice the top of the list is rarely the process you would have picked on gut feel — the gut picks the biggest fire, and the biggest fire is exactly the wrong place to learn. Trust the framework over the instinct. The instinct is optimising for heroism; the framework is optimising for a second automation, which is the thing that actually gets you to a transformed business.

Common questions

What should I automate first?

Whatever scores high on pain, high on clarity and low on blast radius. High pain so people care and the win is visible; high clarity so it works without a discovery saga; low blast radius so the inevitable early wrinkles are cheap to fix and do not scare everyone off. Score each candidate one to five on the three axes, rank by pain plus clarity minus blast radius, and start at the top.

Why not start with our biggest, most painful process?

Because the biggest mess is the hardest to automate, the slowest to show results and the most likely to fail, so it burns your credibility and momentum on the process most likely to defeat you. Your first automation is not about solving your biggest problem. It is about proving the approach works, fast, so that you get to do a second one. The cross-department epic is your eventual prize, not your opener.

What kinds of process tend to make a good first automation?

Chasing is often the ideal one — overdue invoices, missing documents, unsubmitted timesheets. High pain because nobody does it, high clarity because the rules are simple, low blast radius because a reminder cannot break much. Approval routing for a well-understood approval, a single tracker replacing a chaotic spreadsheet, and document capture for one high-volume document type all score well on the same three axes.

What should I keep away from at the start?

Anything that moves money unsupervised, which has a large blast radius by definition. Automate around it first, build trust, then approach it carefully — never as your opening move. Avoid the end-to-end process spanning four teams too, and any process nobody agrees on: if two people describe it differently, clarity is low, and you are not automating, you are refereeing a disagreement that no software settles.

The next step is usually smaller than people expect. Talk to us about one process worth starting with.

Related: the one-workflow start and how to describe a process so AI can build it.

Also worth reading: what a workflow really is.