Benchmark secret #3: Your business is unique so simple comparisons will not work

Imagine that you are part of a study on human weight. It turns out that you are heavier than 98% of all the other people in the study. Is this good ?

I hope we can all agree that bigger is not always better so perhaps being at the 98th percentile is not a good thing? If you are a 12 year old and 5'1", there is potential for some serious health complications. If you are a 25 year old 6'9" wrestler, you may be underweight.

We all, automatically make some allowance for height, gender and age. We expect men to weigh a little more than women, the tall to weigh more than the petite, grown adults (especially muscle-bound wrestlers) to weigh more than children. These are all relationships that we have an intuitive sense for.

With a very large sample, such as is possible when they weigh and measure you every single time you go to the Doctor's office, you can simply disregard a large part of your sample for comparison purposes to find results for people like you. "So let's compare your weight to other men, 6'1" and over aged 51 and ... oh dear, you have some work to do".

With smaller samples, such as you get in any business-benchmarking exercise this is just not going to work. Businesses are complex organisations and each one is unique, with their own DNA, a result of their environment (things you cannot change) and managerial choices (things you can). So, how can you get a fair, useful, actionable benchmark?

What you need is a quantitative model. (Whether you call this statistics, machine learning, data science, predictive analytics or AI is unimportant, your choice.) A model quantifies and explains the relationships between the performance measure and all the inputs that influence the measure. The model's prediction of your performance (your personalized benchmark) is not based on companies that are exactly like you but on all companies, in the same general arena, with the model allowing for the differences in their inputs. If you can perform better than this benchmark, you're really doing something worthwhile. If not, the model can give you strong guidance on what you might want to change to improve performance.