Every Monday, someone downloads a spreadsheet, removes a few columns, copies the remaining rows into another tool and emails three people. The task is familiar enough that nobody talks about it anymore.

That is a reasonable place to investigate automation. Familiar work often contains a mixture of predictable rules and decisions that still need a person. The first step is to separate those parts.

Watch the work before redesigning it

Ask the person who performs the task to show the entire process, including the checks they do without thinking. They may recognize duplicate records by sight, correct a date format or exclude a customer with a special agreement.

These details matter. Automating only the visible copying can move mistakes faster. Record the input, each decision, the destination and the conditions that make the operator pause.

Choose a bounded first workflow

A useful first candidate has a clear trigger, a limited set of rules and an outcome someone can verify. For example: when an approved enquiry becomes a project, create the project folder and notify the assigned owner.

Avoid beginning with a process that changes every week or depends on undocumented judgment. It may need clarification before automation. Sometimes the first improvement is simply agreeing on a shared definition of “approved.”

Keep ownership visible

Write down who receives exceptions and who maintains the rules. If a folder cannot be created because the project name contains an unsupported character, someone needs to know. A failed run should become a visible task with enough context to repair it.

Give the operator a way to pause the automation when inputs are wrong. Include a record of what happened so they can check whether a step has already been completed before attempting it again.

Decide which actions need human review

Creating an internal task carries different consequences from sending an invoice or deleting a record. Start with preparation steps where that suits the workflow, and keep approval for decisions that need judgment or carry a meaningful cost.

For example, an automation could prepare a draft follow-up email with the relevant project details, while a person reviews the message before it is sent. That boundary should be a deliberate product decision.

Measure the whole job

Track more than the time spent clicking. Include the time needed to check results, handle exceptions and maintain the automation. A task that runs quickly but creates a pile of repair work may need a different design.

  • Document the current workflow with its operator.
  • Choose one trigger and one verifiable outcome.
  • Build repeat handling and a visible exception queue.
  • Run a small pilot alongside the existing process.
  • Review whether the total workload became easier.

Reliable automation gives a team a process it can understand and supervise. Start with a piece of work people recognize, make its rules explicit and expand only when the first version behaves predictably.