How big a correction

When teaching a model to learn, there is one tiny dial that decides almost the entire result. The field calls it the learning rate. It sounds technical, and you can picture it simply: after each round of learning from experience, the model has to decide whether to change itself a lot or a little.

Like someone practising darts. If after each miss they adjust their hand too hard, the dart flies past on the other side, then overcorrects back, and never settles on the centre. Adjust too timidly and they improve, but so slowly it is maddening. Machines are the same: correct too hard and it overshoots the best point and oscillates; correct too gently and it learns properly but sluggishly and expensively. What is interesting is that there is no single right number for every case, only a rate that suits each problem.

Change in a business seems to me to move to exactly that rhythm. Change everything at once and the organisation wobbles; hesitate too long and the moment passes. A skilled operator, I think, is like a well-tuned learning rate: knowing when to accelerate and when to go slowly and surely.

At this moment, does your organisation feel like it is changing a little too fast, or being a little too cautious?

To your desk

Change too fast and you wobble, too slow and you are left behind. This is exactly the balance every model has to strike.

A sensible approach for this week: instead of introducing a batch of new tools at once, pick the single process eating the most time and give the team about two weeks to master it before widening. Fast enough to see a result soon, slow enough that nobody is left behind.