Automating business processes: where to start

Most automation projects do not fail in the code. They fail earlier, at the moment someone picks the wrong process to automate. The wrong pick is usually the most visible annoyance instead of the most expensive one.

Step one: map a real week

Ask each person on the team to note, for one normal week, the tasks they repeat and roughly how long each takes. No tooling needed; a note on a phone works. The result is always surprising. The task everyone complains about often costs two hours a week, while some silent routine like re-typing order data eats ten.

Step two: score, do not debate

For each recurring task, answer four questions on a 1 to 5 scale: How often does it happen? How much time does it take in total? How clear are the rules? How bad is a mistake? Multiply the first three, divide by the fourth. This crude formula settles arguments better than any meeting, because it forces frequency and rule-clarity into the decision.

Step three: automate one task completely

Take the highest score and automate it end to end, including the boring parts: error handling, the escalation path to a human, the log that shows what happened. A process that is 90 percent automated still requires someone to babysit it, which means the saved time quietly evaporates.

• One process, fully done, beats five processes half done.
• Measure the before state (hours per week) or you will never prove the after.
• Keep a human approval step in the first month, then remove it where the data says you can.

What this looks like in practice

A typical first project at a services company: inbound requests arrive by email, someone reads each one, checks two or three criteria, and either replies with a standard answer or forwards it. That is four to eight hours a week of pattern matching a system can do in seconds, with the unusual cases still landing on a human desk. Start there, bank the hours, then go after the next one.