Which Business Processes Should You Automate First?

Start by listing every bottleneck. These are the processes where work visibly piles up behind one person, or where the same manual step repeats dozens of times a week. A sawmill selling plywood and lacquer through a trade counter, distributors and a small online shop would list sales-order entry, payment reminders, stock reconciliation and expense approvals. Each one argues to go first.

Ask the people who do the work. The person keying orders knows which step eats the morning, and that detail rarely survives being summarised upward to a manager who sees only the finished report. A list built from real complaints is more useful than one built to look impressive on a slide, and it earns the team’s cooperation later.

Ask four questions of each. Volume asks how often the task happens and repetition asks whether it follows the same steps almost every time in a normal working week. Stability asks whether the process has held steady for months or keeps being redesigned, and risk asks what breaks when an edge case is handled wrong.

Volume and repetition alone mislead. A process that scores high on both but is reshaped every month is not ready, because you would be automating a moving target and rebuilding the rule. A stable, repetitive task that touches nothing when it fails, such as logging a photocopier meter reading, is a safe place to build confidence even if savings are modest. Write the four scores in a shared sheet beside the owner’s name and the date, so the next review starts from evidence instead of recollection.

Good scores can mislead. Customer records that live in three spreadsheets, with the same company spelled three different ways, turn automated follow-up emails into confused messages sent faster and more often.

Run two checks first. First, can the systems exchange data automatically through an API, an export file or a shared database, instead of only a screen that a person reads? Second, can whoever keeps the customer and stock records vouch that they are accurate enough to act on without a manual check? That cleanup is the real first project.

Skip the biggest process. A pilot covering one branch, one product line or one team proves the approach and surfaces exceptions nobody predicted, and at that scale a mistake is cheap. A lacquer reorder alert for the sawmill stockroom is a fine pilot.

Give the pilot one named owner. This person watches output for the first few weeks, compares it with the manual result, and decides when the automation is trustworthy enough to widen. Without an owner, a pilot runs unwatched until a customer or an auditor notices a wrong figure that has already been copied into the daybook and the ledger.

Measure the manual process first. Before the pilot starts, record how long a batch of creditor invoices takes to key, how many errors surface weekly and how many staff hours it consumes. Without that baseline, nobody can say whether the automation helped or merely felt faster because it was new, so the review turns into a debate of opinions.

Then repeat the measures. If the pilot is faster, more accurate or frees time without creating new problems, move to the next candidate and apply the same criteria, not first-success enthusiasm. Agree the stop rule in advance. If error counts rise above the manual baseline for two weeks running, pause the automation and return the work to people while the cause is found.