Claudia worked on this piece of analytics for the last 6 months. It was nerve-wracking. She was working for a large global provider of syndicated market research that had assembled a mind-blowing dataset: Data about all new product launches in CPG in the US. Details product perception and presales purchase intent, sales data, distribution data, everything.
Claudia’s task was to build a mechanism that predicts -based on presales shopper assessments- whether or not a product will sell and survive.
Nothing worked. Purchase intent correlated with success ZERO. Regression delivered R2 close to ZERO. Then, some destiny let the company reach out to us at Success Drivers.
We ran our holistic causal machine learning approach and achieved great explanation power but something was still odd. Typically, the method gives great transparency about causal relationships, nonlinearities, and two-way interactions of any kind.
We paused. The look at the skewed distribution of success (very few are very successful) brought the Eureka moment. Such a distribution evolved only if you multiply four or more independent success drivers. Its rare that all 4 hit the mark at a time, so it becomes rare that a product survives the first year after lunch.
Suddenly all this made sense. You will not survive with bad packaging or messaging, hardly with a missing brand. You will not survive if the pricing is not appropriate. You will not survive if the product is not that great, so people don’t want to rebuy it. You will not survive if retailers do not put the product on their shelves.
There are so many MUST Dos, it is likely that you miss just one of them. If you do, you fail.
Methodically speaking, this is a 4-way interaction – the success can be computed by multiplying success drivers variables. Causal machine learning can learn this out of data.
Later, I realized this learning applies to some degree to all business success drivers. Drivers seemed to work indeed independently on an incremental/micro level, but not on a macro level. This is why we can get easily fooled.