Observation before prescription

The fastest way to waste an automation budget is to automate a process nobody has looked at. The second fastest is to look at it by asking people to describe it.

What people remember and what systems record

Ask three people how a renewal is processed and you will get three honest, different answers. None of them is wrong. Each describes the part they touch, in the order they remember it, minus the exceptions that happen every week but never make it into a description. The systems, meanwhile, recorded all of it: who opened which file, which spreadsheet was updated, when the email went out, how often the same document was attached twice.

So I start there. Before any design, the current state is derived from what the systems actually recorded, then written up as process records that say what happens, in what order, how often, and by whom. The people who do the work then validate each record. They correct what the evidence misread, and they add what the evidence could not see. That combination, evidence first and memory as review, is far more accurate than either alone.

The records describe; they do not recommend

The discipline that makes this work is a refusal. The process records contain no recommendations. Nothing in them says what should change. That is deliberate, and it is harder than it sounds, because the moment you see a process from evidence you can see what is wrong with it.

Keeping observation and prescription separate does two things. It keeps the records honest, because nobody is shading a description to support a conclusion they have already reached. And it keeps the people who do the work engaged, because validating a description of your own job is a very different experience from being handed someone else’s plan for it.

Then one piece at a time

Only after the records are validated does the question of automation come up, and it comes up one piece at a time. The first piece is the one that hurts most and can be measured. It gets built, put in front of the people who use it, and measured before and after, in units the business already uses: hours, days late, items rekeyed. If the number did not move, that is the finding, and it is a useful one.

The next piece is chosen from the records, informed by the measure. There is no twelve-month program with phases on a slide. There is a map of the current state, a list of pieces, and a scorecard that says what each one did. Sponsors like that more than they expect to, because for once they can see what they paid for.

Why this is slower, and why it is not

Observation takes weeks that a proposal could skip. But the proposal that skips it automates the parts that were already wrong, and finds out six months later. Measured against that, looking first is the fast option.