Free resource · Guide
What to automate first
Your first automation decides whether you ever build a second one. Here is a two-question test that finds the task that pays back, a way to score your real week against it, and the two traps that bite beginners.
Free, and free of a catch. See the whole library.
Why the obvious pick is usually wrong
Ask someone what they want to automate and they name the task they hate most. But hate is a bad compass: the most annoying tasks are often annoying precisely because they are full of exceptions and judgement calls — the things automation handles worst. Pick one of those first and the automation fails, and the lesson you take is 'this does not work for my job'. The right first pick is not the task you resent; it is the task that repeats.
The Leverage Test — two questions
Put any task on two axes. One: how often does it repeat, in the same shape — not just frequently, but the same steps with different inputs? Two: how much judgement does it need — how often does doing it well depend on context a set of instructions could not carry? Where a task lands tells you what to do with it.
Score a real week
Do not score hypothetical tasks — score last week. List everything you did more than once, then run each through the two questions honestly. 'Compile the Monday status update from the same three sources' scores high-repeat, low-judgement: automate it. 'Reply to a difficult client' scores high-judgement no matter how often it happens: at most, let AI draft while you decide. If listing and scoring by hand feels like a chore, make the AI run the interview.
Here are tasks I did more than once this week: compiling the Monday status update from three spreadsheets, chasing overdue invoices by email, replying to new-enquiry emails, and formatting the weekly figures for the team meeting. For each one, ask me the two questions that matter for automation — does it repeat in the same shape each time, and how much judgement does it actually need — one task at a time. Then rank them into three lists: automate now, let AI draft while I decide, and leave alone. Challenge me where my answers sound optimistic.Trap one — automating a broken process
Automation is an amplifier: it multiplies whatever is there, including the flaws. If the weekly report is compiled from three sources because nobody ever merged them, automating the compilation entrenches the mess and makes it faster. The test: before automating a task, ask whether you would teach it to a new colleague exactly as it stands. If the honest answer is 'well, first I would fix…' — fix that first. It is cheaper to repair a process than to repair an automation built on one.
Trap two — automation that fails silently
The first automation must fail loudly. A report that arrives wrong is worse than a report that does not arrive — because the second gets noticed. So keep yourself in the loop until trust is earned: the automation drafts, you send. Give it output you can check at a glance — a count, a total, a list of what it touched — rather than a black box that says 'done'. Only when checking has become boring, because it is never wrong, do you step out of the loop. Boredom is the promotion criterion.
The simplest rule of all
Automate the task you could hand to a capable temp with a page of written instructions and expect done right on their first morning. If you cannot write that page — because the instructions keep sprouting 'except when' — the task is not ready, whatever the tooling promises. Write the page, automate the page.
More like this, as we write it.
The library is still filling. Start free and we will tell you when the next resource ships — one email, no spam.
Start free