In anticipation of his upcoming keynote co-presentation, Picking the Right Modeling Technique for the Problem, at Predictive Analytics World London, October 12-13, 2016, we asked Michael Berry, Analytics Director at TripAdvisor Hotel Solutions, a few michael_berry-imagequestions about his work in predictive analytics.

Q: In your work with predictive analytics, what behavior do your models predict?

A: At TripAdvisor for Business, one of our most important products is subscription-based. We price our subscriptions based on the value our product will deliver to hoteliers in the form of increased direct bookings on their web sites. This means predicting their future traffic, click-through rates, conversion rates, room rates, average length of stay, and so on. Beyond that, I worry about all the usual things subscription-based businesses worry about: What is the probability that a subscriber will renew? What actions of ours can increase that probability? Which non-subscribers are the best prospects? What actions on our part will lead to increased owner engagement?

Q: How does predictive analytics deliver value at your organization? What is one specific way in which it actively drives decisions?

A: I’ve already mentioned pricing.  Another area is sales efficiency.  There are over 900,000 hotels listed on TripAdvisor and our salespeople can’t reach all of them. We use predictive models to pick which properties to call.

Q: Can you describe a successful result, such as the predictive lift of your model or the ROI of an analytics initiative?

A: No. In a public forum like this, I generally show graphs with no numbers on the axes. Of course internally we measure things like the increase in expected value of sales leads so we know how valuable our work is.

Q: What surprising discovery have you unearthed in your data?

A: Here’s one that surprised me a bit when I first started looking at hotel ratings data: The average bubble rating of all reviews is higher than the average bubble rating of all hotels.  Both are pretty high since people tend to like the places they picked, but the difference is noticeable.  How can that be?  Well, some properties have enormous numbers of reviews. Think The Bellagio in Las Vegas.  These properties tend to be traveler favorites so their thousands of reviews bring up the average review score. But the Bellagio is still just one hotel, so it doesn’t affect the average hotel score any more than a Motel 6 on a truck route somewhere.

Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Predictive Analytics World.

A: There is no one best type of predictive model; you need to pick your tools to match the problem you are trying to solve.


Don't miss Michael’s keynote co-presentation, Picking the Right Modeling Technique for the Problem on Wednesday, October 12, 2016 at 11:45 am at Predictive Analytics World London. Click here to register to attend.