What we will not automate

An automation vendor with nothing on the list of things they will not do is not measuring anything. They are quoting.

Three categories are scored low deliberately, and the hours behind them are largely kept out of the figure.

Work that needs human judgment

Some processes look highly repetitive from the outside and are not. Code review, sprint planning, deciding which customer complaint is a pattern and which is an outlier: these run to a rhythm, they have inputs and outputs, and they are made almost entirely of judgment. The rhythm is what makes them look automatable. The judgment is what makes them not.

What can be taken off them is the administration wrapped around them: the reminder, the collation, the writing up, the chasing of the person who has not responded. That is worth having and it is worth counting. The core is not on offer.

Physical work

This one cost us a real correction. Early in testing, a delivery route and a night baking shift were scored medium feasibility with their full hours attached, which put well over a hundred hours a month each into the automatable column. Nobody is automating a van or an oven. The hours were real and the classification was nonsense, and the headline it produced was worse than useless, because it was the kind of number that falls apart the moment somebody who does the job reads it.

Physical core work now scores low, and only the administrative overhead around it counts: the route sheet, the stock count typed in twice, the delivery confirmations copied from one place to another. In that one case the correction took a figure of a hundred and sixty-nine hours a month down to seventeen.

Anything running on a system with no way in

A process can be perfectly repetitive and still be out of reach, because the software it lives in has no interface, no export worth the name, and a vendor with no interest in either. Scraping is sometimes an answer and it is always a liability: it breaks when the page changes, and it breaks quietly.

Where scraping is the only route, the process scores medium at best, and the proposal says out loud that its maintenance cost is higher than the rest.

Why the refusals make the number better

Every hour removed from the automatable column makes the remaining figure harder to argue with. That is the whole trade. A report claiming a company can automate two thirds of its work is dismissed in the first meeting by anyone who does the work. A report that says a hundred and forty hours a month, here is which ones, here is what stays exactly as it is, survives the meeting and gets used to make a decision.

The refusals are also the fastest way to tell whether an audit happened at all. If nothing in a report is marked as out of scope, nothing was examined closely enough to rule out.