2X Founder & Chief Decision Scientist | TEDx Speaker | Contributor to MIT SMR & Forbes | B-School Adjunct Prof. | C-Suite Advisor
Early in our journey, we helped a telecom client identify consumers who were likely to churn. Simple #algorithms like decision trees could improve customer retention by 39%. It is intuitive to predict churn with factors such as ‘last call date’ or ‘bill amount’. While the client's marketing team was thrilled to try it out, our ambitions were higher! We experimented with more advanced, black-box algorithms such as neural networks. We were excited when customer retention shot up to 66%. Starry-eyed, we presented the results. To our utter horror, the users dumped the solution?? Ultimately, the project failed. The graveyard of #datascience projects is filled with solutions like these. Well-intended and high performing, but complex. As a result, users give them the cold-shoulder. How can a leader funding a high-profile #analytics project avoid this fate? That's the focus of my #newsletter this week. Read on by clicking the link. To get the next post in your inbox, please subscribe. https://lnkd.in/eJFj8Cg #data #business
deepak rout
4 年
This is so true.Model explainability at many times is a deal decider.
Kumar Jayant
4 年
Well written indeed. If it ain't easy to comprehend and hence execute, businesses will have hard time wrapping around stargazed recommendations. Wonder if so many competitions are being won using recent algos like xgboost, when will if ever, the industry will start using these in LIVE business decision. What's your opinion Ganes Kesari
Software Dev Engineer at Amazon
4 年Yes true,sir. Simple models which are easily understandable will be adopted faster. Algorithms too complex to be used ,though delivering better results might not be acknowledged. Happens when testing people too . Like when a Sudha Murthy book is easier and more satisfying to read than maybe a book written by Shashi Tharoor where someone needs to sit and search every word. Audience is the key. Wonderful article ,sir