Meri Gruber had an interesting post on her Competing on Execution blog this week – Zappos: Make Me Happier with Business Analytics – that prompted me to want to add my own thoughts. She says:
Why is Zappos only offering me help from other customers, when they are sitting on a wealth of order (and return) information? Zappos know what styles and sizes are being returned and exchanged for other sizes. It’s all in their order database. They must see which manufacturers size 8.5B is consistent year to year, style to style, and which ones jump all over the place. Why not use this information to create a “Zappos size” by manufacturer, by style? One that improves over time?
An analytic model or score, in other words, that could be used to drive suggestions or recommendations on their site. A model could score how likely it is that each shoe size is likely to be right for you (given what you say your shoe size is and given what they know about your returns and the returns/comments of others) and then the decision could be made as to what size to recommend you order. Helpful, effective decisions that are specific to a customer and driven by analytics.
Zappos is, of course, not the only retailer guilty of thinking that the only way it can help people make decisions about which products to buy is through social media. Despite how widespread this opinion is, it is missing a big part of the picture. Past behavior is the best predictor of future behavior. When you have data about large numbers of people it will be possible to find patterns, segments and sub-segments, that act similarly. Any company that tracks sales, returns, complaints has a tremendous amount of data about customer behavior.
As Meri suggests, retailers like Zappos could put this data to work using analytics. They could find the products or brands that get returned for a different size (larger or smaller). They could identify products that get returned or complained about by people who are not typically returners or complainers. They could see what kinds of customers like or dislike a product (based on reviews) and see if data they have about customers that haven’t reviewed the product that would predict a like/dislike review. They could analyze their operational, transactional, structured data to help their customers and they could combine this with their social media data to extend its reach and relevance (“people with a similar order/return history to you really don’t like these shoes”). Add-in the power of text analytics to combine the reviews/comments with the rest of the structured data and you can get more accurate, more targeted predictions.
The move to social media and the embracing of reviews/comments by mainstream retailers is a good thing – customers like to know what other customers think. But customers tell you more through their actions than they do through their reviews. And reviews have biases too. The context, the writer’s emotional state, the time of day and much more can influence a review. Plus, of course, only a small percentage of customers even write reviews? You have all this additional data, so why not use it?
Disclosure: Meri works down the corridor from me (and remember, I work from home), so it is no surprise that I agree with her.
