I Invented Surge Pricing For Uber. Here’s Why it Wouldn’t Have Worked for Wendy’s.
Lessons from the dynamic pricing front lines...
KN Note: I stopped working at Uber in 2017. Nothing in this blog post should be all construed as representing Uber in any way, shape, or form. These thoughts and perspectives are entirely my own.
Quick Takeaways:
Dynamic Pricing worked for Uber because ridesharing is a stochastic system; if a car is available to be dispatched to you, it results from a series of decisions drivers and riders make that can be influenced economically.
Because it’s stochastic, demand outpacing supply can lead to longer transactions, lowering the number of trips the fleet can do in an hour. Uber sets prices to maximize the number of trips for the fleet as a whole - this surplus of trips compounds multiplicatively with the surge surcharge, creating a big economic win for Uber.
To be best of my knowledge, fast food is not a stochastic system, meaning that the upside for Wendy’s is pretty limited.
So, I'm the guy who invented and implemented surge pricing at Uber. As you can guess, I've been getting a ton of messages about Wendy's recent announcement to use dynamic pricing (and nearly immediate retraction) for their food. I wanted to share some thoughts on why surge pricing works for Uber and why Wendy's might not find the same success.
Surge Pricing (aka “Dynamic Pricing”) was first implemented in late 2011 to respond to the demand spikes Uber was experiencing every Friday and Saturday night, as well as on major holidays like New Year’s Eve. I joined Uber in mid 2011 as employee #21 and their second data hire, and was immediately put to work on a long list of data science problems we were wrestling with, even in those early days. I personally implemented the entire surge pricing product (ML models, backend system, internal ops tools, even the surge pricing screen you saw in the app) and ran the team that supervised surge until mid 2015. Other Uber Data team members took my humble beginnings and built far more elaborate and advanced systems, but I was there right at the messy beginning of Uber’s journey with dynamic pricing.
Why Surge Pricing Worked At Uber
To understand why surge pricing worked so well for Uber, you need to understand a few basic insights. First, you need to know that Uber drivers usually need to move after dropping off a rider to be in the best position for their next pickup. It's pretty rare for the drop-off spot to be the ideal place for the next ride (airports are an exception). So, a typical night for a driver involves a cycle of driving to pick up a customer, taking them where they need to go, repositioning themselves, and waiting for the next ride request. Repositioning the car also burns gas (or kwH) and generally costs money, so as a rule, drivers are looking to reposition as little as possible without a good reason.
Second, it's important to understand that all phases of an Uber driver's cycle are connected. The whole system is what data scientists call "stochastic," meaning the current state of the market (where drivers are located and who's available) depends on decisions made by drivers and riders in the past. Why does this matter for dynamic pricing? Well, we figured out early on at Uber that longer wait times for pickups actually hurt our ability to maximize the number of trips our fleet can handle.
For example, if an Uber is five minutes away from picking you up and your destination is 15 minutes away, that driver spends about 20 minutes on the trip and can do up to three trips per hour. But - if an Uber is 15 minutes away, they spend 15 minutes getting to you and 15 more minutes driving you – that's a 30-minute transaction, so they can only do two trips per hour. (Real life is more complex than this simple example, but it illustrates how increasing wait times can reduce the overall efficiency of our fleet.) Put another way, the future efficiency of Uber's fleet (expressed as possible trips per hour) largely depends on the average ETA for a typical ride request right now. If you apply this insight to a typical Friday night at Uber before dynamic pricing, whenever we had a localized surge in demand, each sequential ride was being dispatched to cars further and further away, eventually lowering the carrying capacity of the Uber fleet to the point that nobody could get a ride.
This understanding of the problem also motivates Uber’s dynamic pricing objectives - by setting prices to both discourage customer demand and, much more importantly, motivate drivers to relocate to optimal spots in town, we can keep arrival times low enough that the carrying capacity stays high and we maximize the number of trips for everyone involved. This was beneficial for drivers (more trips), riders (more fulfilled requests), and of course, Uber. Maximizing the number of trips per hour has a more significant impact on revenue than the actual surcharge applied to Uber trips. In fact, early on, the Uber fleet could complete two or three times as many trips per hour with dynamic pricing. This effect compounds multiplicatively with the per-trip surge surcharge we would apply to each trip, resulting in a supercharged revenue effect for drivers and Uber.
Should we be surging Baconators?
So, what does this have to do with Wendy's? Unlike Uber, Wendy's and fast food industries don't operate in a stochastic system. The availability of a cheeseburger isn't determined by everyone's decisions before you, except when it comes to inventory management (which Wendy's likely has under control). Therefore, the effect size of Wendy's dynamic pricing experiment will probably be quite modest.
Dynamic pricing could still be relevant and potentially beneficial for Wendy's in certain situations. For instance, they may charge slightly more during high-demand times or possibly offer discounts during low-demand periods (though, based on my experience with dynamic pricing in food and beverage industries, the latter seems less likely).
The primary goal of implementing dynamic pricing at Wendy's is probably to incentivize customers to eat earlier or later than the typical mealtime rush. Wendy's might have found that customers tend to eat all at once during lunch hour, making employee scheduling and staffing more challenging. By encouraging customers to adjust their meal times slightly through dynamic pricing, they could improve operational efficiency and ultimately enhance their bottom line.
Put another way, most employees work shifts of four to eight hours. If all the demand happens within 30 minutes, you'll likely have at least three and a half hours where you're overstaffed, just to make sure you have enough staff to cover that peak period. So, the ultimate goal is probably to shape consumer behavior so that customer demand and employee staffing line up more intuitively.
This isn't exactly a new concept; early bird discounts and happy hours have been trying to achieve this for businesses for decades. Wendy's is simply taking a more modern approach. While it may not be a bad strategy, I suspect the results will underperform compared to Uber, mainly because there's only so much time flexibility the average Wendy's customer has.
It's unlikely someone would order a triple baconator at eight in the morning just because it's economically incentivized (though it would be interesting to see if this happens). Instead, we might see the lunch rush extend from 30 minutes or an hour, to two or two and a half hours. This could be beneficial for Wendy's when implemented on a larger scale, but it probably won't revolutionize the food and beverage industry. The revenue generated from this experiment might not justify the investment, but like everyone else, I'm eager to see how their quarterly financials reflect the results.
Wear your fireproof pants - Dynamic Pricing & PR
Lastly, Wendy's is learning what Uber discovered: dynamic pricing can be challenging from a PR standpoint. Once money and changing prices are involved, concepts of fairness and ethics come into play. Even years after helping develop dynamic pricing at Uber, I still face criticism about the company's decision to implement this product, even though it has proven successful in many ways. I bet Wendy's press team is having a tough week, and I hope they find a path to product success like we did. But honestly, I'm doubtful that'll happen.