Which business processes can be automated?

Five criteria

A process is a suitable candidate if as many as possible of these points apply:

• Repetition: It runs regularly and in a comparable form. A one-off case once a year is not worth the effort.
• Clear inputs: The information needed can be named — not necessarily all available, but nameable.
• Describable rules: You can say in plain sentences what should happen in which case.
• Reachable systems: The programs involved have an interface, an export option or at least a file they can read.
• Tangible benefit: The work saved or the errors avoided are in proportion to the effort.

If one of these criteria is missing, that is not an exclusion but a task: rules can be sharpened, access can be created. If three are missing, the process is probably the wrong place to start.

Three examples

The following three processes are examples, not finished product modules. They show how varied “automation” can look.

• Taking in enquiries in a structured way: What arrives today as free-text emails is captured through a form, checked for completeness and assigned to the responsible person. The gain is fewer follow-up questions, not fewer conversations.
• Reconciling data between two systems: Stock levels, master data or orders are transferred, and discrepancies are made visible rather than silently overwritten.
• Preparing recurring reports: The figures are brought together and made available. The assessment stays with the people who know the business.

Where the line runs

There are processes that could technically be automated but should not be. The line lies where a decision requires discretion or directly affects people.

• Decisions about people — such as job applications or employment-law matters — do not belong in a rule that nobody checks any more.
• Cases with a large impact but low frequency: here a prepared decision with human approval is better than an automatic one.
• Processes whose rules keep changing: every change then becomes an adjustment. Here it pays to clarify the rule first.

A good automation does not have to decide every case itself. It has to recognise which case it cannot decide, and put it to a person — with everything they need to decide.

Self-check in four questions

Take a process that costs you time today and answer in order:

• How often does it run — per day, per week, per month?
• How many minutes does one run take, from first touch to completion?
• What information does it need, and is it in one place or in four?
• At what point does a person have to decide — and do they need information for it that they currently have to search for?

Frequency times duration gives today’s manual time spent. That is not a promise of savings: approvals, monitoring and exceptions remain work. But it is the figure on which you can decide whether a closer look is worthwhile.

Where to start

Not with the biggest process. A sensible choice is one where you can answer all four questions above and which runs at least weekly. A manageable first step gives you experience with your own data — and with the second automation, that is worth more than any amount of advance planning.