Dynamic Pricing: A Vendo Explainer

http://blog.vendoservices.com/vendo-blog/2015/11/18/dynamic-pricing-a-vendo-explainer

Let’s say you have 100,000 people visit your online store every day. How do you give each visitor the right price?

Simple question. F***ing hard to answer. But we’re going to try in this Vendo pricing explainer.

Let’s start with some definitions. “100,000” is clear, “visitors” ok, “online shop” got it, “each one”…yep. Really the only thing we need define is “the right price.”

First, what’s the right price? The right price gets the sale and gets the visitor to spend as much as possible. That’s the definition. And it’s what you need to do. Anything less than the right price for each visitor kills your return on investment in two things:

  1. Product
  2. Advertising to get that visitor in front of your product.

No one can afford that. Especially not you.

How much do you lose from having the wrong price for each visitor? In our work with leading companies we’ve seen bad pricing drag down revenues by 10% to 20%. Imagine your top line revenue is $1m a month. Pricing right for each visitor would send revenues upwards by $100,000 to $200,000. Your profits grow even more because there is no change to your fixed costs like staff, office space, etc. Most variable costs like bandwidth don’t really change either.

Big difference. Why don’t people do it? Well, they do. It’s just really hard and you can lose a lot of money by doing it badly. Let’s say you just hired some nerd fresh out of math school (or business school). And one morning he comes into your office with a proposal. He says, “hey, charge this guy that price…and charge that one this price.” Would you do it? Probably not. It’s risky. You don’t know if you would be losing money by changing your prices. And the last thing you want to do is lose money by doing something you don’t understand. That’s a recipe for disaster. You tell him your prices are fine. They’re the same as everyone else’s prices.

The only problem is that your prices aren’t fine. Not even close.

Back in the 1978 the US deregulated the airline industry. All of a sudden, from one day to the next, airlines could set their own prices. This is the story of what happened.  It’s also the story of how a shift in regulation in a totally different industry is going to change the way price your products.

Robert Crandall, the CEO of American Airlines, recognized the problem of putting planes in the air with not enough passengers. Each seat was a “perishable good.” Think of a seat like a banana at the grocery store. If the grocery store didn’t sell it in time, it “perished” and had no value at all. An empty seat in an airplane had no value at all to an airline that had to fly that plane on schedule.

Banana-Chair-2 airline_seat-01_Thumbnail1.jpg0e7c339e-4390-454d-86bf-0184564d8bfeLarger

There are other transport models that solve this problem. Take a bus in a developing country. They don’t leave until it’s full. Problem solved. But people pay airlines to take off on time. So they had to approach the problem differently.

What did they do? They put a bunch of Ph.D.s in a room right next to an even bigger room that contained an IBM mainframe. They set out to crack the code. To fill each seat at the highest price. There’s an old joke in aviation circles that goes like this:

“Do you know what makes an airplane fly?”

  • “Uhh…maybe higher pressure under the wings, lower pressure on top of wings and sufficient thrust to…”

“Nope, money makes airplanes fly.”

The guys at American Airlines wanted to make more money so they could fly more planes. And they did. Along the way they created the foundation for all contemporary pricing strategies. They recognized differences in people and used powerful algorithms to deliver different prices to them.

Here’s one of the first things they figured out. Business travelers are different from vacation travelers. The airline can separate them by charging different prices for flights that had a round trip including Saturdays. No businessperson wanted to spend the weekend on a business trip so American Airlines could charge them more.

Today there are literally millions of ways that flight prices are adjusted for each visitor. Here’s one you can try right now. Search for a flight, write down the price, then clear your cookies and do the same search. Bet you’ll get a different price. By the mid-nineties pricing was generating $500 million a year for the airlines. Btw, if you want to get really technical on airline pricing check out Preston McAffee, of the California Institute of Technology. He wrote a history* of the development of dynamic pricing in the airline industry. But be careful. There’s a little bit of this…

Screen Shot 2015-11-18 at 4.14.53 PM

and a lot of this…

Screen Shot 2015-11-18 at 4.15.52 PM

In the 1980’s the hotel industry followed the airlines. It turns out that the great Wayne Gretzky’s motto, “You don’t make any of the shots you don’t take,” also applies to hotel rooms. Each night “you don’t make any money on the rooms you don’t sell.”

In the late 1990’s Google started building on what the airlines and hotels had learned to develop tools for understand the differences between users. Google used the tools to give each user the results they were looking for. Today, most major online retailers, including Amazon, are standing on the shoulders of these early giants to deliver dynamic pricing.

So, how do you do it? How do you find the right price for each visitor? Well the price has to get the sale (Conversion) then it has to make sure the person spends as much as possible (Lifetime Value). That’s tricky. Everyone knows that lowering your price from say, $30 to like $10 would sell a lot more. Conversions would go up. Maybe it would look like this:

100,000 visitors see price of $30, 100 visitors buy. Conversion rate is 1 in 1,000 (or, 0.1%)

v.

100,000 visitors see price of $20, 150 visitors buy. Conversion rate is 1 in 667 (or, 0.15%)

If you are only looking at conversion rates then $20 is way, way better than $30 (50% better!). But you are looking at Lifetime Value, too. Let’s add that to the calculation and see what happens.

100,000 visitors see price of $30, 100 visitors buy, Lifetime value is $60. Revenue per Click is $0.06.

v.

100,000 visitors see price of $20, 150 visitors buy, Lifetime value is $40. Revenue per Click is $0.06.

They’re the same. We’re back where we started. Why? Because we are treating all 100,000 visitors the same. We’re showing the same price (either $30 or $20) to all 100,000 visitors. But they aren’t the same.

They’re all different. Here are some of the ways they are different:

  • Rich v. Poor
  • Educated about your product v. Don’t know your product at all
  • Afternoon shoppers v. Late night shoppers
  • On a mobile device v. On a desktop
  • Weekday shoppers v. Weekend shoppers
  • (and there are many more variables…)

And that’s where the opportunity is. Just like the airlines and hotels and Amazon, the goal is to  figure out which price to charge each visitor – not which price is best for all visitors.

So we start looking closer and closer at each visitor. Here’s one that’s visiting your store:

Rich Poor
Educated about your product Don’t know your product at all
Afternoon Late night
Mobile device Desktop
Weekday Weekend
Variable X Variable Y
Variable A Variable B
Variable 1 Variable 2
Variable etc. Variable et al.
and so on… and so forth…

This visitor is rich, does know your product, shopping in the afternoon on a mobile device on the weekend and is variable X, A, 2, etc. and so on…

What price should we give this visitor maximize our revenue?

Compare that visitor with another visitor to your store who is poor, not educated about your product, shopping late at night on a PC on a weekday and is Variable Y, B, 1, et al. and furthermore…

Which price is right for this visitor? Is it the same as the one above? Why would it be?

Rich Poor
Educated about your product Don’t know your product at all
Afternoon Late night
Mobile device Desktop
Weekday Weekend
Variable X Variable Y
Variable A Variable B
Variable 1 Variable 2
Variable etc. Variable et al.
and so on… and so forth…

Let’s say we give our pricing a floor (ex. we won’t price below, say, $20) but, even though we could choose a ceiling (ex. we won’t price above $70) we don’t. We let the ceiling set itself.

We have to test prices to know which ones give us the highest revenue per click for each visitor. That means measuring the conversion and lifetime value. Conversion is easy and immediate. Not so Lifetime Value. Lifetime Value is a slow train that can take months to reach the station. To understand what Lifetime Value will become we need sophisticated forecasting tools that look at upgrade and rebill ratios and tell us how Lifetime Value is shaping up for each price for each visitor.

Next we could divide everyone up into two groups randomly then give one group one price and the other group another price (A/B split testing). That would be the traditional way to do it. The challenge is that we are dealing with 100,000 very different visitors each day. A/B split testing would need very large samples that would ignore the differences between visitors, or, take years to give us answers.

Put simply, the tools we normally use to run our business are not capable of solving the problem of pricing right for each user. They can’t do it in the relevant time frame (which is, in our example, of 100,000 visitors, delivering the right price more than once per second (100,000 visitors / 84,600 seconds = 1.15 times per second).

For this reason we are now stepping out of the world of our parents, grandparents (unless your grandfather was Isaac Asimov) and into the world of artificial intelligence.   

The same technology behind driverless cars is now driving pricing. And it does the same thing. It reads thousands of signals (ex. your product and a visitor that is rich, pc, etc.) and decides on which price will maximize its chances of success. Just like a car driving down the road reads signals like the lane markers, stop lights, scans for pedestrians, measures speed, etc..

Here’s the definition of artificial intelligence (AI): A system that perceives its environment and takes actions that maximize its chances of success

Success for a AI in a car means safely going from point A to point B. Success for AI pricing is maximizing revenue. Or, stopping revenue losses caused by humans that give everyone the same price.

What does the future look like? Google just announced that their AI for search outperforms everything they have ever done by 15%. Whoa! And they expect that number to grow. They have a serious team of researchers who are guiding, teaching and optimizing their AI. It’s a very human-centric effort to build great AI. It takes a lot of human qualities like persistence and creativity. We’ve been at it for five years with a strong team of Ph.D’s.

We don’t know what the upper limit is but every day it makes the human approach to pricing look worse.

 

———

 

*http://mcafee.cc/Papers/PDF/DynamicPriceDiscrimination.pdf

Leave a Reply

Your email address will not be published. Required fields are marked *