Recently, I came across a very interesting article in New York Times, How Company Learns Your Secrets. In this article, the author Charles Duhigg describes the application of analytics and behavioral science to large retailers’ marketing efforts. The real fascinating part of the article was the story of Target’s efforts to use purchase history and demographic data to identify which of its female customers were in the second trimester of a pregnancy, so that they can better target those customers, and try shifting their buying habits. Very thought provoking story. Somehow, the story and discussion followed by the story made me rethink the idea of using the static data like purchase history and demographic data to construct complex mathematical, predictive models. It also made me think about the questions like when to rely on these models, to what extent, and how to use customer feedback and insights to make these models more reliable to support complex decisions.
Analytics seems to be a very hot topic right now, and the software vendors make it look even hotter by spreading bright and shiny success stories about it. We have heard a lot about the benefits of analytics used successfully. However, we don’t hear much about what happens when analytics done poorly and used inappropriately. Analytics provides highly powerful insights if given in the right hands, but, what happens when those powerful tools are given in the wrong hands. I have heard some stories of how analytics goes wrong in the absence of right feedback mechanisms; like, inaccurate plagiarism detection failed the brightest student, the most influential customers got fired because of the assumption of low value, the smartest employee quit because of the wrongly done performance tracking, and I still get the diapers and baby-formula samples when my kid is seven! These all reminds me of The Monkey Story, which is about a bird who owned a gold coin but could not hold it in his beak, so could not really have it.
In the obsession to throw coupons and samples to get some more business, you may find it tempting to spy on your target customers’ purchase history. However, if you want to make effective decisions to gain a competitive edge, watch out. And, be aware of the inherent biases of predictive analytics, have dialogs with your customers, get useful feedback from them, engage them, and effectively use that data to improve your model and augment your predictions and decisions. I love this video that emphasizes importance of having dialog:
For owning the analytics gold coin and for really having it, it is crucial to constantly use your judgement and wear a skeptic hat to explore and understand the data patterns and existence of its inherent biases. When you are building a predictive model, what matters are the associations between the variables, and the associations may or may not have causal relationships. Cotton-ball purchases and pregnancy may have high correlations, however, that does not mean that cotton ball purchase causes pregnancy. That’s why the predictive models may or may not have good explanatory powers and vise a versa. And, that is why, to come up with better predictions and decisions, you need multiple approaches and techniques to verify unintuitive results of your model. To enhance predictive and explanatory power of your models, it is important to augment the operational data with customer experience data generated through chats, emails, questions, dialog or surveys, and, the qualitative customer insights and feedback gained from on-going social media conversations. You don't need to assume much when you have the opportunity to ask.
Are you using analytics to detect patterns in your data?
Is your analytics helping you gain competitive edge?
Are you augmenting and integrating the customer surveys, feedbacks, insights and experiences with your predictive models? How?
Please share your ideas and experiences.