The critical key to reliable forecasting: Be less wrong.

The key to good forecasting, that magic elixir most of us take, is not to be right, but to be increasingly less wrong.

We know  the future will be different, being less wrong about that difference is better than consuming resources trying to be right, because you never will be.

For a decade, several decades ago, as marketing manager of a very significant business, I did a weekly sales record for about 50 SKU’s, by hand. It was in the late eighties, early nineties, the days before this was made easy.

Every Monday morning, I took about 15 minutes to record the sales on a sheet, with a 5 week rolling average, and a 5 week rolling forecast. Every month I did the same, but it took a little longer, as there were comparisons to the relevant quarter the year before, and budget, which took about 45 minutes.

In 10 years, I only ever got one forecast right, but was usually very close. Nobody took any notice at all of the forecasts of the sales force, despite them being part of the sales KPI’s. When manufacturing had choices to make about factory utilisation and what not to make, they came to me, and ignored the rest.

This was simply the building of a qualitative knowledge over time.

We routinely defer to ‘Bayesian’ statistics, a theorem proposed by English statistician Thomas Bayes in 1763, that dealt with the probability of a future event, and how that probability becomes more certain with the addition of information relevant to the outcome. We see Bayesian thinking all around us all the time. Every time we see an outcome to an action, and adjust before we repeat the action, we are using Bayesian thinking. Artillery is the obvious example. Use one cannon to get as close as you can, observe the degree to which you are long or short of the target, and adjust accordingly. When you land one on the target is when all  the other cannons in the group adopt the same settings and blast away.

In business, we can spent inordinate amounts of time and energy trying to get the last 5% accuracy, when it would be far better to take a decision, and move ahead knowing that the chances are you will be wrong, but able to adjust and accommodate the degree of ‘wrongness’ with far less effort. This is the basis of continuous improvement, Plan, Do, Check, Act. 

Bayesian theory at work, every day.