First published February 8, 2021.
I approach problems as a digital marketer who studied economics and likes technology. I also get bored easily. So I always want things to work better and faster, and to run on their own when they can.
Handing decisions to a machine takes more than good software. You have to build up to it in stages.
Stage 1: See what’s happening
- Are you measuring the right things?
- Can you build reports and dashboards without a lot of effort?
If you can’t produce reliable reports, you aren’t ready to let a machine make decisions.
Stage 2: Turn what you see into action
- Can you find insights you can act on?
- Can you repeat the cycle of insight, action, and learning?
- Are patterns showing up, and are you writing them down?
Once insights repeat reliably, people can start making those decisions by hand using clear rules. As you build a record of the decisions and their limits, you can slowly hand more of them over.
Stage 3: Let technology decide
- Do you see the same insights again and again?
- Do you have limits and fail-safes in place?
- Will you know when something unusual happens?
- Is the culture ready?
The culture question is as hard as the technical one. Maybe harder. People have to trust the system, and they have to know when to step in.
Most companies haven’t gotten past the first two stages.
Looking at this again in 2026
I wrote this in 2021. AI has changed what machines can do since then. It hasn’t changed the order. Teams that skip the first two stages and go straight to automation get fast decisions built on data they can’t trust.
Get new essays and notes by email.
One email when I publish something new.
