It is twenty to six on a Thursday. The last customer has gone, and someone at the desk is typing a stack of delivery notes into the accounting package, line by line. Twenty-two of them. Tomorrow there will be another stack.

That is the moment most owners first think about automation. It is also the worst moment to choose a tool, because at twenty to six everything looks worth automating. The question is not can this be automated — nearly anything can. The question is which single task pays you back, and which ones you should leave exactly as they are.

Start with a tally, not with software

Before you look at any product, spend one ordinary week counting. A sheet of paper next to the keyboard is enough. Every time someone does a repetitive office job — retyping something, copying between two systems, sending the same email with three words changed, building the same weekly overview — write down three things: what it was, how long it took, who did it.

One week, no judgement, no improving anything yet. At the end you will have a list of perhaps fifteen jobs. In our experience two or three of them account for most of the lost time, and they are almost never the ones people complain about loudest. The noisy tasks are the annoying ones. The expensive tasks are the quiet ones that come round forty times a week and take four minutes each.

The sum that decides

Take your top three and multiply: minutes per time × times per week × 45 working weeks. Four minutes, forty times a week, is 120 hours a year. That is three working weeks of one person's time for one task — worth a serious look. A job that takes twenty minutes but happens once a month is four hours a year. Leave it alone, however irritating it is.

Then set that against what it costs to build and keep. A small, tightly scoped automation — pull the details out of incoming orders, put them in one place, send one confirmation — is usually a few days of work, not a project. If the yearly time saved is not at least three or four times the first-year build cost, this is probably not your first candidate. And note the word keep: every automation needs somebody who notices when it stops. Budget a little for that.

Three kinds of work to leave alone

  • Work that needs a judgement call. Deciding whether a complaint gets a refund, which patient can be squeezed in, whether a quote is worth doing at all. Automate the preparation — all the facts on one screen — but leave the decision with a person.
  • Work that is about to change. If you are switching accounting packages in January or rewriting your menu, do not automate the current version of it. You will pay twice.
  • Work that happens rarely. Year-end, the annual price list, the quarterly report. Rare tasks are exactly where automation breaks quietly, because nobody uses it often enough to notice it has stopped.

Automate the boring half first

Most office work is a sandwich: a mechanical bit, a human bit, another mechanical bit. A quote means pulling up the customer's details and last order (mechanical), deciding the price (human), and sending a tidy PDF with a reminder in five days (mechanical). You do not need a system that writes quotes. You need one that hands you a half-finished quote and sends it properly the moment you press the button.

That half-and-half approach is also far easier to trust. A person stays in the loop, mistakes stay visible, and if something hiccups you fall back on the old way for a day instead of grinding to a halt. Pick the task whose output is already standard, too. If the same email currently goes out in five personal styles, automating it starts with agreeing on one version — and that conversation, not the technology, is what takes the time. Sometimes you finish the conversation and find you are already halfway there.

What to do this week

  • Put a sheet of paper by the desk and tally every repetitive job for five working days: what, how long, who.
  • On Friday, multiply your top three by times-per-week × 45 and see which one crosses 50 hours a year.
  • Write down in two sentences what the ideal version of that one task would look like — before you open a single tool.

If you end up with a task on paper and no feel for whether it is half a day of fiddling or a real build, we are happy to look at it with you at ITSysPro. One conversation is usually enough to tell.