Starting an Automation Project: The First Four Weeks

Week one: measure

Before deciding anything, measure what you have. Pick the two or three candidate processes and record, for a week: how many times each happens, how long each takes, and how many go wrong.

This week costs you almost nothing and it settles every argument that follows — which process to choose, whether the project paid for itself, and whether to fund the next one.

Measure by observation and tally, not by asking. Self-reported figures are wrong in both directions.

Week two: choose and specify

1. Score candidates on frequency, tedium, error cost and how little judgement each needs
2. Pick one — the one with the most repetition and the least thinking
3. Write the happy path as numbered steps
4. Write the exceptions, and keep going until you have at least five
5. Name the internal owner and confirm they have half a day a week

The exceptions list is the part that determines whether the project succeeds. A specification without one is not finished.

Week three: get comparable quotes

Send the same written scope to three suppliers. Ask each for the price broken into phases, the assumptions they are making, what is explicitly out of scope, and who would do the work.

State your budget range. Withholding it produces proposals scoped for the wrong scale and wastes everyone's time.

Week four: decide

• Compare on what is being built, not on the headline number
• Read the assumptions section first — it reveals who has thought about your problem
• Take one reference from a project of similar size, not the flagship
• Check ownership of code and accounts is yours, in writing
• Confirm the maintenance arrangement before signing, not after

What happens next

A well-prepared first phase runs six to twelve weeks: specification, build, parallel running against the manual process, then cutover. Expect to spend half a day a week throughout.

Then re-measure against your week-one baseline. That number is what funds phase two, and it is why week one mattered.

What to avoid in the first project

Anything spanning several departments, anything where the process is still being argued about, and anything that requires a third party to build something for you first.

Save those for when you have one success behind you and an organisation that believes automation works.

Frequently asked questions