Finding the hours worth buying back

Automation projects usually fail at the first step, which is choosing what to automate. People automate the thing that is annoying rather than the thing that is expensive, and those are rarely the same job.

Write down a week

Before buying anything, record what you actually did for five working days. Not a timesheet to the minute: a line per task with a rough duration, written as you go rather than reconstructed on Friday.

Almost everybody who does this is surprised, and the surprise is usually about frequency rather than duration. The job that eats the week is the four-minute one done eleven times a day, not the two-hour one done on Thursday.

Sort what you find

• Frequency times duration. This is the real cost, and the ranking it produces is rarely the ranking your irritation produces.
• How much the task varies. Identical every time is automatable. Mostly identical with judgement in the middle is not, or not completely.
• What a failure costs. Automating something that produces an invoice is different from automating something that produces a folder.
• Whether the task should exist. The best automation is often deletion, and it costs nothing to run.

The jobs that are usually worth it

Across small businesses of almost any kind, the same handful come up, because they are high frequency, low judgement and cheap to connect.

1. 01Copying the same information between two places: a form into a spreadsheet, an order into accounting, a lead into wherever you track them.
2. 02Sending a message that is the same every time, with two fields changed.
3. 03Scheduling, including the back and forth about when.
4. 04Reminding somebody, including yourself, about a thing on a date.
5. 05Filing and naming: documents, receipts, images, in a consistent place with a consistent name.
6. 06Producing a recurring document from data that already exists: an invoice, a report, a statement.

What not to start with

Anything that touches money leaving the business, anything that sends a message somebody will read as personal, and anything where the input varies enough that you cannot write down the rule. Those come later, with a person checking, and some of them never come at all.

The tasks that look automatable and are not

A specific trap, because these are the ones people reach for first: they feel mechanical from the outside and contain a decision in the middle.

• Replying to enquiries. The reply is a template; deciding what this particular person needs is not.
• Quoting. Standard jobs can be priced by a rule. The ones that make money usually cannot.
• Scheduling around constraints you have not written down, such as not that customer on a Friday, or not two of those in one day.
• Anything whose input is a photograph, a conversation or somebody's handwriting.
• Categorising things where the categories are still being argued about internally.

Each of these can be partly automated, which is the useful move: the surrounding work goes away and the decision stays. What does not work is automating the decision because the surrounding work was tedious.

Start with the one you dread

A qualification to the arithmetic. Among tasks of roughly equal cost, pick the one you avoid, because avoidance has its own price: the invoicing that happens three weeks late, the follow-up that never goes out, the filing that becomes a day's work in March.

Those tasks are also where automation produces something better than time saved, which is the thing happening at all. An invoice that goes out on the day the work finished is worth more than the twenty minutes it took to send.

Ask the person doing it

If somebody else does the task, they already know where the time goes and they have usually never been asked. They also know which steps exist because of a problem that happened once, which is the information that makes or breaks a workflow. Asking takes ten minutes and prevents the specific failure of automating a process as described rather than as performed, which is how machinery ends up producing something nobody can use.