How to put a number on work before you automate it

Every automation proposal contains a number meant to justify it, and most of those numbers cannot survive being asked where they came from. This is how ours is built, including the parts that make it smaller.

Hours times a blended rate

Each mapped process carries an estimate of hours per month built from two things the employee said: how often it happens, and how long it takes. Frequency times duration. That is the whole derivation, and it is deliberately boring, because anything more sophisticated would be a model rather than a measurement.

Those hours are priced at one blended fully-loaded hourly cost for the company, set by the owner during setup. Blended rather than per-person, for two reasons. The first is practical: fully-loaded cost includes employer taxes, benefits, equipment and the space someone sits in, and few owners have that to hand per head. The second is that per-person rates turn a process map into a salary comparison, and the moment a report can be read that way it stops being about processes.

Ad-hoc time is counted at half

The last question in every interview asks how many hours a week disappear into interruptions. That is real time and real money, and leaving it out would understate the problem. But it is self-reported, with no trigger, no tools and no steps behind it, which makes it the one figure in the report that cannot be checked against anything else.

So it is counted at fifty percent, it sits on its own line, and the line says what it is: self-reported, no mapped process. The mapped processes carry their own subtotal above it, so anyone who wants to argue about the soft half can do that without touching the half that is evidenced.

That is not modesty, it is self-defence. Fold a soft number silently into a headline and challenging it puts the entire figure in doubt, including the part that would have survived on its own.

The basis is printed under the number

Underneath the headline the report prints its basis in full: how many processes were mapped, across how many people, and that the hours are employee-estimated per month. Not in an appendix. Under the number, in the same eyeline.

Any figure that cannot show its own basis should be read as a guess with confidence added.

Four ways this goes wrong

These are worth knowing because they are the ones we have had to fix:

• Two denominators in one view. A share-of-value percentage measured against mapped-process value, sitting beside a headline that also includes ad-hoc time, lets a reader see "one hundred percent of the value" next to a figure twice the size. Every percentage has to name its own denominator on the page.
• Rounding twice. Display hours get rounded for readability. Multiply the rounded hours back up and the column of workings no longer sums to the total printed beside it. Money is derived once, from unrounded hours, and the column has to add up.
• Physical work priced as if it were software. A delivery route or a night shift is a hundred and thirty hours a month of real work, and none of it can be automated. Scoring it as automatable inflates the headline enormously.
• The same person counted twice. Somebody who starts an interview, stops, and comes back later can appear as two people with two sets of findings, both feeding the total. Sessions are grouped by identity and their findings merged rather than added.

Why it is worth being this careful

The number is opened in front of the owner and used to price the work. It is not a marketing figure. If it is inflated, the first thing that happens is not a larger contract, it is an argument, and the argument is unwinnable, because the person on the other side knows their own company better than the report does.

A number the owner attacks and fails to dent is worth several times a bigger number they can take apart in a minute.