Ask a team what they'd like to automate and you'll get a list of whatever annoyed them most recently. That list is almost never the right one.
Annoyance and cost aren't the same thing. The task everyone complains about is often a five-minute job that happens twice a month. The task nobody mentions is the twenty-second one that happens ninety times a day, because it's small enough that it never feels worth raising.
The arithmetic nobody does
Before automating anything, work out the actual annual cost:
Minutes per run × runs per year = hours you're buying back.
A 5-minute task done twice a month costs 2 hours a year. Automating it might take a day. That's a bad trade, and no amount of irritation changes the maths.
A 20-second task done 90 times a day costs about 125 hours a year. That's three working weeks, and it never appears in a single conversation because no individual instance is worth mentioning.
The three properties that matter
The tasks worth automating tend to share the same shape:
- Frequent. Volume is what makes the arithmetic work.
- Deterministic. Same input, same output, every time. If a human makes a judgement call halfway through, you're not automating - you're building a decision engine, and that's a bigger project than you think.
- Consequential when wrong. Manual data entry between two systems isn't expensive because it's slow. It's expensive because it's occasionally wrong, and the wrongness surfaces three weeks later in a report someone acts on.
That third one is where the real money hides. People price automation as time saved and forget it's mostly error eliminated.
What not to automate
Anything that runs once a quarter. Anything where the rules change more often than the code would. Anything where the current process is broken - automating a bad process just produces bad outcomes faster, and now they're harder to see.
If the process only works because a specific person quietly fixes it, automate the fixing, not the process.
That last case is the most common and the most misdiagnosed. Someone has been patching a broken workflow by hand for so long that the patching is the workflow, and nobody has written it down.
Start where the systems meet
In practice, the highest-return automation is almost always at the seams - wherever data gets copied from one system into another by a human. CRM into accounting. Orders into fulfilment. Timesheets into payroll.
Those handoffs are frequent, deterministic, and expensive when wrong. They're also the ones people have stopped noticing, because copying data between systems feels like work rather than waste.
Find yours
If you want to know where your seams are, the fastest route is to ask your team what they do every morning before the real work starts. The answers are usually the roadmap.
Let's map them - the assessment is free, and the arithmetic tends to make the decision for you.