Count one job
Pick one job that keeps coming back. Put rough numbers on it and see what it actually takes out of a year. Then answer four questions and get a straight answer on what I’d do with it: automate it, assist it, fix the process, keep it human, or leave it alone for now.
Worth knowing
The total is everyone’s time added together. Three people losing twenty minutes a day shows up here as one big number, but nobody gets 184 hours handed back in a lump — what comes back is pieces of people’s weeks.
It won’t put money on any of it. I stop the sums at hours on purpose: multiply by a salary and the conversation stops being about the work and starts being about the people doing it, and you lose the honesty you need from them.
A big total settles nothing on its own, either. Plenty of heavy, repetitive jobs should stay exactly where they are, and the four questions are there to catch those: whether the job follows a settled method, how much judgement it takes, whether a mistake would get noticed in time, and what one would cost if it didn’t.
The five answers it can give are the same five I use on a Census day: automate it, assist it, fix the process, keep it human, or leave it alone for now.
The full method is written up in Before You Automate Anything: How to Find the Work AI Should Actually Do.
This counts one job. The Census counts the lot, in a day, with your own people round the table writing them down in their own handwriting. Then we weigh them and sort them the same five ways.