Automate the ordinary: how to choose your first automation

Start with resentment, not ambition

Ask a team what they would happily never do again and you will get a precise list within minutes: copying invoice details into the accounting system, chasing approvals, compiling the Monday report, answering the same five customer questions. That list is more valuable than any strategy deck, because it points directly at work that is frequent, rule-bound and low on judgement.

A simple scoring frame

For each candidate, consider four things. Frequency: how often does it happen? Variability: does it follow the same pattern each time, or does every case need thought? Data access: can a system actually reach the inputs and outputs, or does it depend on information in someone’s head? Cost of error: if the automation gets it wrong, is the result a minor correction or a serious problem?

The sweet spot is high frequency, low variability, accessible data and a recoverable cost of error. Invoice intake with a human approval step usually scores well. Fully automated payment release usually does not — at least not yet.

Automating a broken process only makes it break faster.

Fix the process before you automate it

Many manual processes contain steps that exist only because of an old workaround. Before automating, map the process as it really happens, remove steps that no longer serve a purpose, and agree on what “done” means. The automation you build afterwards will be simpler, cheaper and far easier to maintain.

Earn autonomy gradually

Our default pattern is suggest, then approve, then act. The system first proposes an action for a person to confirm. Once its accuracy is measured and trusted, routine cases move to automatic handling while exceptions still reach a human. This keeps risk low and builds the confidence teams need to expand automation to the next process.