Ever wonder why some companies can charge different prices at different time–like United Airlines, or Uber–while others, like Walgreens, don't? Academcis have long had 5 tests to determine if dynamic pricing could work for an industry. With machine learning, we're reducing that to just 2. Read on to learn what separates dynamic industries from those stuck in the 20th century.
First, what is dynamic pricing? Dynamic pricing is the rapid reaction of price to changes in market demand. Think of it like a stock trading higher or lower depending on what buyers are willing to pay or sellers are willing to take to part with their shares. Just like any market, there may be lots of people watching from the sidelines who have no intention of buying or selling–and these people don't matter. Or they might matter if prices swoop low enough, or spike high enough.
This applies in real life too. You get that job interview with Google. Google is paying for the ticket. There are 5 flights, and you have a meeting with your boss at 4pm on Thursday. You will literally pay anything–because it's not your money anyway–that Google HR will approve to make it back on time for that meeting. Unlike, say, a college student taking a trip home for no particular reason. United had better be sure it still has a seat for you (and Google) to buy that Thursday night when the interview is confirmed.
There's no way to allocate a certain number of seats on each flight for job seekers, prospective employers, etc., But there is a way to make sure people who really, really need to fly can get a seat when they need it, and that is dynamic pricing.
Dynamic pricing makes its way into lots of places–from airlines and Uber to restaurants, apparel retailers, Amazon.com and many more places. Even your local grocery store or Walgreens, possibly. But not all dynamic pricing is the same.
Two types of dynamic pricing
Not all dynamic pricing is created equal. The two types of dynamic pricing are:
- Revenue management. Driven by changes in demand, or to incentivize certain consumer behaviors, revenue management is typical in hotel, airline, rental car, event ticketing, and similar industries.
- Supply-driven. Typical in ecommerce and retail, driven by price-matching or competitor out-of-stocks.
The most profitable type, based on revenue management algorithms, adapts prices to changes in demand or availability of supply. You have experienced this if you've ever taken an Uber at a surge price, or flown on a commercial airline. Even rental cars and hotels are beginning to use this type of dynamic pricing.
The other type, supply-driven, can have significant impact as well. Companies either raise or lower prices based on what competitors are doing, or they mark down inventory to clear room for new inventory soon to arrive and recover cash tied up in the inventory that hasn't yet sold. Think about all those bikes in your favorite sporting goods store in October. Where will the skis and snowboards go? Space is as much a constraint for an online giant like Backcountry.com as it is for a Main Street retailer.
How to know if dynamic pricing is right for your business–old school
Dynamic pricing based on demand is extremely powerful. When American Airlines implemented it in the 1980s, it bankrupted People's Express, costing the startup airline $160 million in losses in one year. National Rental Car swung from a loss to significant profitability, after increasing revenue $56 million in just one year due to dynamic pricing–going from a writedown of $744 million to sale of $1.2 billion in under 2 years.
Traditionally, to understand if revenue management type dynamic pricing is right for your business, academics would ask you to answer these questions:
- Is your capacity relatively fixed?
- Is your demand predictable at all?
- Is your inventory perishable? (For example, a seat on an airline or at a live concert)
- Are your fixed or sunk costs relatively significant compared to your variable costs?
- Does demand vary by time? (For example, is there more demand on weekends?)
If you answered "yes" to at least four of these questions, dynamic pricing may work for you. If you answered "no" to two or more, it may still work for you if based on supply, depending on which two.
Use machine learning, reduce to 2
The above 5 are the traditional definition; we think that with the advent of machine learning in pricing, you can reduce this list to two:
1. Does demand vary by time?
2. Will consumers accept dynamic pricing?
You'll note that in our view, supply and predictability are irrelevant.
The supply of Ubers is actually far from fixed: however, dynamic pricing matches to demand in that moment. Early on, during surges Uber drivers had no idea there was a surge. They were just driving (now, they can see where is surging and by how much–but Uber makes no guarantee that what they see is what they get paid). Yet surge pricing worked–by keeping people with lower willingess to pay out of the market so those with real need (or, to be fair, means) could get a ride when they needed it.
Restaurants vary pricing by time of day, day of week, and more. They do it subtly. Similar strategies are pervasive at grocery stores and pharmacies. Here supply is not at issue: the only thing that matters is customer acceptance.
With machine learning, predictability is no longer important. By using novelty detection, surges can be spotted without any forecast of future demand. In fact, these models are provably more accurate and resilient than traditional revenue management approaches which–for example–might depend on last Christmas's air travel demand to predict this Christmas. How did 2009 stack up against 2006-2008, I wonder? Ouch!
By rapidly looking at demand in a given moment, Uber or Lyft can adjust prices without having to factor in rain, the baseball team's schedule, a city's convention schedule, or any of a host of other very complicated externalities. Airlines need all of these inputs, or at least they think they do.
Surge pricing of course highlights perhaps the biggest challenge of dynamic pricing, namely consumer acceptance. In some categories, for example staples like paper towels and toilet paper, consumers know exactly what price they usually pay. Deviating will lead to massive swings in demand–however, it also creates "strategic customers" who wait til the next sale and stock up. Instead, if demand does vary, and customers are willing to accept dynamic pricing, you may be able to move forward.
The classic example of demand varying by time and dynamic pricing making sense–but not being acceptable–is groceries before a hurricane. Because it would be perceived as "gouging", grocery stores (and gas stations) cannot raise the price. So the first person in can buy as much water, canned foods, etc as they possibly can carry. This is a horrible allocation of scarce resources–if each gallon of water cost $30, you would only buy what you need, leaving some on the shelf for the next person. But at $1, you fill your minivan. And yet, consumers are unwiling to accept dynamic pricing in this highly dynamic circumstance.
What if I'm a retailer or manufacturer?
To understand if dynamic pricing makes sense in an ecommerce or retail environment, there are two steps. First, run an experiment on your current systems, adjusting price to see if you can impact your sales and profitability. This can apply to your prices, your prices vis-a-vis your competition, or a markdown or promotion compared to an earlier one. You may have already done this (for example, raising prices after a supplier changed its own pricing).
Data in hand, your second step is to analyze the impact. First, did consumers accept price changes, or was there outrage? Second, what was the financial impact? If there was no impact, or even after a month, results were not statistically significant, dynamic pricing might not be right for you. But if you observe significant changes to revenue, conversion rate, customer acquisition cost, lifetime value, or profitability, dynamic pricing may be a fit.
Dynamic pricing systems
Some companies build dynamic pricing in house; for example, Starwood spent $50 million building its own system. Others use vendors such as Perfect Price, PROS, or Sabre. Any system has three major components.
- Data collection and cleanup: collect, in as close to real-time as possible, indicators of demand such as purchases, page views, etc.
- Demand model generation: generate estimates of demand based on observed data.
- Burst detection: determine if demand in the current moment is exceeding predictions enough to warrant a change in price.
- Pricing service: The API or system which updates prices the consumer sees based on the demand model and any burst in demand. Typically these are highly customizable.
Each component is critical; however not all use cases require burst detection. To our knowledge only Perfect Price offers burst detection, or the ability to detect if a spike in demand in near eral time, separating real spikes from simply noise. Burst detection simplifies long term demand estimates by capturing and adjusting for difficult-to-forecast features such as weather in multiple markets.
Dynamic pricing enables companies to capture more value–and deliver more value. Frequently upon implementing dynamic pricing, overall prices only rise 1-2% but revenue and profitability increase as much as 10%. This is because dynamic pricing both is a form of price discrimination, charging some people more (early adopters, business travelers, etc.) and also improving the utilization of underutilized assets (the hotel room that would have sat empty at $100 but is rented at $50, or the shipping department at 10% capacity where if prices were lowered on key items, it might be at 30%).
When you consider dynamic pricing (or simply price optimization), engage with all stakeholders in your company and outside. Frequently many customers ask for and embrace dynamic pricing; today's generation expect "right pricing" whereas prior generations may have grown accustomed to simplicity and stasis. After all, only two generations ago department stores were a tremendous innovation–with standardized pricing across the country.
Contact us if you'd like to learn whether dynamic pricing is right for your business.