What to automate first when everything is manual
When all of it is manual, the instinct is to start with the most annoying task. That is usually the wrong one.
Founder
The most annoying task is rarely the most expensive one. Annoyance tracks how unpleasant something feels, and cost tracks how often it happens. Those two come apart constantly, which is why businesses so often automate the wrong thing first and conclude that automation was oversold.
Here is the scoring method we use, which you can run on a piece of paper.
Score each task on three things
List every recurring task that eats time. For each one, note:
- Frequency. How many times a week does it happen? A two-minute job done forty times beats a two-hour job done monthly.
- Decision content. Does it require judgement, or is it a rule somebody follows? Rules automate cleanly. Judgement does not, and pretending otherwise is where automation projects go wrong.
- Failure cost. What happens when it is skipped? A missed reminder that loses a customer scores far higher than a report nobody reads.
The best first candidate is high frequency, low decision content, high failure cost. That combination is almost always some form of chasing: the payment reminder, the document request, the appointment confirmation, the enquiry follow-up.
Why chasing wins so often
Chasing scores well on all three axes at once. It happens constantly. The rule is trivial, so much so that the only reason a human does it is that somebody has to remember. And the cost of skipping it is direct and measurable: an invoice paid late, a slot left empty, a member who lapsed, an enquiry that went cold.
It also has a property that matters more than it should: nobody enjoys it. Automating chasing removes work that a person was doing badly because they disliked it, which means the automated version is often better than the manual one rather than merely cheaper. That is the rare case where the annoying task and the expensive task coincide.
What to leave alone at the start
Two categories are worth deliberately postponing.
The first is anything requiring judgement, especially where a customer sees the output. AI can draft these, and a person should approve them. That is a genuine time saving and it is a different project from the mechanical work. Start with the mechanical work, because it is where the certainty is.
The second is any process that is currently broken. If your quoting process produces inconsistent prices, automating it produces inconsistent prices faster and with more authority. Fix the process, then automate it. Skipping this step is the single most common reason automation makes things worse.
Start narrower than feels worthwhile
The instinct is to automate a whole category at once: all reminders, everywhere. Resist it. Pick the single highest-scoring sequence, put it live, and watch it for a fortnight on real cases.
You will learn something you could not have specified in advance, usually about an exception nobody mentioned because everybody in the business knows about it. Discovering that on one sequence is a small correction. Discovering it across eleven sequences is a rebuild.
The narrow start also matters commercially. One working sequence is evidence, and evidence makes the next decision easy. A large first project is a bet, and bets make the next decision political.