From time to time, I’m offered a job or consulting assignment in which the primary goal is to find out how to extract value from available data.
This typically happens when an executive team realizes that their company has been accumulating large volumes of data (sensor readings, transaction logs from an e-commerce website, event logs from a mobile app or process automation system) without the ability to translate that data into any kind of benefit or competitive advantage.
At that point the company doesn’t know what it’s looking for, but it doesn’t want to risk losing space to a competitor using “cutting-edge analytics” either. This can be an interesting challenge for a data scientist, allowing the combination of science and art to identify and exploit a variety of actionable data insights. However, the results are unlikely to be as good as what can be achieved when advanced analytics efforts are driven by relevant business questions rather than a desire to “embrace big data” or “make the company machine learning ready”.
Some questions for analytics are obvious; others aren’t
Consider an online retail company that sells a wide variety of products in different categories. It wouldn’t be surprising to see that company use advanced analytics that relies on historical data and competitive analyses to make informed decisions about pricing, promotions, assortment management. Questions like What is the optimal retail price for this product? and Which specific items can be used to maximize the value of our promotions? will come naturally to decision-makers.
Likewise, we would expect to see an airline using its vast amounts of customer and operational data to dynamically calculate ticket and baggage fee prices rather than sticking to hard-coded decisions with pre-determined, non-optimized outcomes.
Yet, the same companies may be blind to other big opportunities to use analytics to drive results, leaving money on the table when they fail to ask:
Online retail company: How can we move the consumers who currently buy from us in a single niche to buying in several?
Airline: Which change to one of our preflight digital touchpoints can produce the highest return on investment?
To understand the benefits of answering those questions, the companies would first need to apply expert judgment and insight from data in order to understand its capabilities and performance gaps:
Online retail company: recognize that the business is composed of traditional “hit” products plus a “long tail” of less popular ones that in aggregate make for a significant market. Machine learning could then be used to perform microsegmentation to connect individual customers with products they want but wouldn’t easily find on their own.
Airline: realize that while a large portion of the prebooking digital touchpoints is owned by flight search aggregators, a lucrative percentage of those touchpoints happen in the company’s mobile app and website. Traditional data analysis could then be used to determine, for example, that smartphone users have a significantly higher conversion rate than website users, and an investment in improving the website experience (in particular, increasing response times to reduce bounce rates) has the potential to produce the highest ROI.
Starting from business questions rather than high-tech solutions is the best way to prevent the “man with a hammer” syndrome
On purpose, one of my examples above would benefit from a solution involving machine learning, while the other required traditional analytics.
When an executive asks, How can we start using machine learning to drive results? (a question I’ve seen asked by more business leaders than I can count), they are inadvertently creating a “man with a hammer” trap. Machine learning is a great tool, but it’s not a panacea. Many business problems will require a different approach to turn data into useful insights.
The questions you ask determine which solutions you’ll see and which will remain hidden.
When a decision-maker starts sprinkling terms like AI, machine learning, big data, or deep learning in conversations, the best thing to do is to quickly move their focus from solution-based innovation to question-based innovation. Starting all your analytics efforts from a business question is one of the best protections against poorly conceived and implemented advanced analytics solutions that are as likely to decrease as to increase performance.