To productize or not

‘To productize or not’ is the question that many boutique analytics firms are facing. Faced with an urge to grow beyond the small company / start up tag, and being unable to match these aspirations with the current business model of providing high end, customized solutions analytics firms are now trying to evolve products based on the experience that they had with their experiences.

At a certain level this indeed seems the way ahead and a very logical extension of the domain and technical skill sets that these companies have build in the past year.

However, the potential of ‘productization’ of analytics solutions need to be viewed from the prism of two critical things:

1. Re-applicability
2. Statistical robustness

1. Re-applicability:
Some of the questions that concern with re-applicability of analytics solutions are:
- Is the product, with minor customizations, ready to be installed?
- Are there any category / industry / domain specific nuances critical and can be they be programmed in the model?
- Is the list of variables / scenarios considered for building the model exhaustive?
- Do the variables change so much over a short period of time that using an outdated model would actually back-fire? E.g. A model that is used to predict likeability of clothes (Since fashion changes so much over time)

Is the effort to address the above issues significantly less than creating a new model?
If answer to all above is yes, then the product can be said to have high re-applicability.

2. Statistical robustness:

Any product needs to stand the test of accuracy. In analytics, it means it has to be statistically robust. A weak model would result in unpredictable results. A weak model is unable to stand the test of integrity, of the model as well as the modeler.

If Re-applicability as well Statistical robustness is high, then go forth comrade and productize - and may your tribe muliply!