Two years ago, I was invited to a $300m start-up to advise them on choosing the proper "big data technology". We managed to find the proper technology for their "expressed" needs. However, all the cells in my body wanted to tell the VP of Analytics that their targeted questions reminded me of the story of the drunk man under the lamppost [1]:
A policeman sees a drunk man searching for something under a streetlight and asks what the drunk has lost. He says he lost his keys and they both look under the streetlight together. After a few minutes the policeman asks if he is sure he lost them here, and the drunk replies, no, and that he lost them in the park. The policeman asks why he is searching here, and the drunk replies, "this is where the light is".
The streetlight effect is a common disease of business analytics. Managers are drunk with all the news on big data and analytics. They have the data (the light!) and they need someone help them find the keys to value creation. So, they employ data scientists to turn those data into insights. But, that does not mean data scientists will be able to find the "keys" to value creation. The keys can be somewhere else...
I described the issue and recommended the VP an "Action Research" to "develop a strategy for analytics". That would help them prioritize the business questions and focus on the ones that could create the highest value (ROI). The VP of Analytics politely dismissed my point and expressed that they already know such questions. Today, however, I heard that the company's income has dropped significantly, the CEO is dismissed, and many employees have been laid off. It seemed foreseeable three years ago as analytics have been at the core of their business model.
Asking the right question is the basis of a successful business analytics. According to Jeremy Howard, asking the right question means you need to a) identify the right objective, and b) select the right lever. Jeremy, one of the most well-known data scientists of our time, said [2]:
"The most important thing is to look at the question you are trying to answer. And I split that into two parts. The first is the objective that you are trying to drive. For example, insurance is focused around the objective of maximizing the profit from each customer. The second thing, we look at is what I call the levers. What are the things that I can actually change that’s going to impact the business? And in insurance, one of the most important one by far is the price that you set. With this in mind, maximizing profit by changing the price, I can now go and say what kind of model do I need to build which would hook up those two pieces [price and profit]."
Right objectives source in business strategies. Does the business want to increase the revenue and market share? or it targets increasing the profit? Does it aim at developing a brand or a new product? or it intends to decrease the cost per customer or cost per product? Some may want to decrease the risk rather than increasing the value. In a nutshell, the objective must be aligned with the business strategy.
Finding the right lever can be more challenging. For instance, a data science team that is asked to find a way to increase the sale of an e-business can target either of the following questions:
- What is the competitive price for each product?
- Which products are frequently bought together?
- Why did not the customer, who read the details of several products, make the purchase?
- Why do our customers rarely come back for the second purchase?
- Why do current customers buy from us, but not from our rivals?
- How can a purchase become "fun"?
It is evident that each business faces a list of questions to be answered. These questions should be prioritized based on their ROI, i.e, the value they can create as opposed to the cost of required technologies, skills, datasets, etc. This is where it becomes complex as there are few people who have the capability to carry out this prioritization. Nevertheless, its complexity should not lead companies neglecting prioritization, or they end up in a condition similar to the following case. According to Gartner [3, 4, 5],
A global car manufacturer decided to run a sentiment analysis project. Six months and $10M later the findings from big data were distributed to all dealerships. All thousands of them were laughing out loud as every one of them already knew about the provided insight.
Consequently, I recommend my data scientist fellows focus on the keys, not the light. I hope I find time to write about how this prioritization can be conducted. I invite you to comment your opinion and related experiences.
Amin K. Amiri
Assistant Professor of Information Systems
Tilburg University
Netherlands
References:
[1] David H. Freedman (2010). Wrong: Why Experts Keep Failing Us. Little, Brown and Company. ISBN 0-316-02378-7.
[2] https://www.youtube.com/ watch?v=yPGzOw_KcBk
[3] http://searchcio.techtarget.com/news/4500251611/Seven-big-data-failures-to-watch-out-for
[4] https://datafloq.com/read/top-reasons-of-hadoop-big-data-project-failures/2185
[5] http://blogs.gartner.com/svetlana-sicular/big-botched-data/