Sifting Through the Signal vs. the Noise
One of the most challenging (and rewarding) parts of building a demand-driven pricing engine is the ability to sort out the signals from the noise in order to build the world’s best dynamic pricing recommendations.
Wheelhouse analyzes millions and millions of data points every day. We process and analyze booking data across all major marketplaces in thousands of cities all over the world every day. It takes some serious brain power and computational efforts to be able to synthesize this information into industry-leading pricing recommendations.
For example, a couple just booked an apartment in Cape Town, South Africa on Booking.com for an avg. of $84/night in February. A party of 8 people just booked a house on HomeAway in South Lake Tahoe, CA, USA for the week of 4th of July. A solo traveler just booked a flat in London on Airbnb for this weekend. And that’s just a few. A thousand other bookings have happened since you started reading this blog post. That’s a lot of data to ingest every single day. But how do we turn those booking patterns into daily price recommendations for any property anywhere in the world?
Sifting through signals vs. noise can be especially difficult in smaller markets where there isn’t an abundance of booking data every single day. For example, it’s fairly reasonable for someone to book a 4-bedroom Airbnb in Delavan, WI (a small town in Wisconsin of about 8,000 people) for June 20, 2019. But what happens if 3 other bookings are made a few days later for those exact same stay dates that are 7+ months away from today? Is that a random coincidence (i.e. noise), or is there a big event taking place in Delavan, WI on June 20, 2019, and are there about to be many more bookings to come? What about 15 more bookings on June 20th, all houses with at least 3 bedrooms? Does that signal there’s a big event happening? Should your prices automatically go up because your neighbor got a booking? That’s what we mean when we say we are sorting out the signal from the noise.
So how were we able to do this? We’ll spare you from all the details, but because of a resampling method in statistics called the jackknife technique — which you can read more about here — our Data Science team has improved our ability to identify the signals of high demand, instead of overreacting to noisy outliers. This jackknife technique is essentially a way to systematically remove bias from smaller subsets of data.
Let’s look at a real-life example of this in Greensboro, GA. The charts below show how price recommendations in this smaller market change before and after the improvements. The first three recommendations (the “before”) show a lot of “events” appearing and disappearing from one model date to the next (left label). The bottom chart (the “after”) consistently only has the ones that were present in all previous model runs, i.e. the dates that are certain to be events, or periods of higher demand.
Conclusion
We recently launched these updates to the Pricing Engine, and customers in smaller markets should immediately notice pricing recommendations that are less noisy. With these improvements and our Data Science team, Wheelhouse is better positioned than ever to separate the signals from the noise and, more importantly, generate pricing recommendations that maximize total revenue for markets of all sizes, anywhere in the world.
Ready to start taking advantage of our world-class Data Science for your own listings? Sign up with the promo code SIGNALS for $20 in Wheelhouse credit after a free 30-day trial and start earning more and saving time.
If you’re a current customer and have already noticed these improvements to the pricing model, we’d love to hear your thoughts too! Drop us a line at hello@usewheelhouse.com or leave a comment below.
