00:00:07 Pricing optimization and challenges.
00:01:34 Classical approaches to pricing and their limitations.
00:04:12 Importance of integrating pricing and supply chain optimization.
00:05:48 Successful examples of dynamic pricing in e-commerce.
00:07:26 Balancing market messaging and optimization for dynamic pricing.
00:09:07 Using probabilistic approach for better pricing optimization.
00:11:58 Importance of historical data and pricing practices.
00:13:56 Technical requirements for pricing optimization.
00:15:03 Process and data requirements for price optimization.
00:17:35 Importance of competitor pricing in different industries.
00:19:22 Predicting future pricing and profits, and industries already doing it.
00:20:33 The key lesson on pricing optimization and its relation to demand.
00:21:18 Closing remarks.
Summary
In an interview, Lokad founder Joannes Vermorel discusses the challenges of modeling pricing and demand, emphasizing that they are often disconnected. He argues that traditional forecasting models are limited as they don’t factor in pricing effects. Vermorel believes that pricing optimization and supply chain management are fundamentally entangled and should be considered together. Dynamic pricing strategies, such as those used by Amazon, allow companies to manage stock-outs and maximize margins. Vermorel stresses the importance of a probabilistic approach to forecasting, which accounts for a range of possibilities. By bringing together teams responsible for pricing, planning, purchasing, and production, businesses can execute better optimization strategies using minimal data requirements.
Extended Summary
In the interview, host Kieran Chandler and Joannes Vermorel, the founder of Lokad, discuss the challenges of modeling the link between pricing and demand, and how advances in machine learning and big data have made pricing optimization possible for modern companies.
Vermorel starts by explaining that the most striking aspect of pricing is how disconnected it is from demand planning. Marketing teams typically determine the price points for products, while planning teams decide on the demand. The sales team may coordinate as part of an S&OP (Sales and Operations Planning) process, but pricing is often ignored once a decision has been made.
He goes on to discuss the basic principle of economics that as prices increase, demand decreases. For most goods, this is true, although there are exceptions known as Veblen goods. Vermorel shares that these goods are rare, and the expertise at Lokad primarily focuses on more common products. The challenge arises when considering that most supply chain planning tools do not take pricing into account, leaving planners blind to the strong pricing effects that may impact forecasts.
The reason many supply chain tools do not account for pricing is due to the numerous variables, such as marketing influences, that can affect demand. However, Vermorel argues that the sheer number of variables is not necessarily a problem, as modern computers can process hundreds of thousands of variables. Instead, the issue lies in the simplistic models used, which often only consider demand as a moving average with a seasonal coefficient, without factoring in stockouts or pricing effects.
Vermorel emphasizes the importance of monitoring competitor prices, which has become easier with web crawlers and specialized data retrieval companies. He believes that the starting point for pricing optimization is acknowledging that supply chain optimization must factor in pricing optimization. These two elements are fundamentally entangled and cannot be separated. By recognizing this, companies can begin to work towards more effective pricing strategies and supply chain management.
Vermorel emphasizes the importance of incorporating pricing into the forecasting process. Traditional forecasting models, which treat pricing and demand separately, are limited in their ability to accurately predict the future. He argues that pricing should not be treated as a separate entity to be forecasted, but rather as a factor that can be engineered to optimize supply chain management.
E-commerce companies are cited as leading the way in smart, quantitative optimization through dynamic pricing. Vermorel explains that by adjusting prices based on inventory levels, companies can better manage stock-outs and maximize margins. For instance, if a company knows it is heading for a stock-out, there is no need to liquidate the remaining stock quickly. Instead, they can increase the price to make higher margins on the last few units. Conversely, if competitors are experiencing stock-outs, there is no need to rush and create a stock-out situation for themselves.
Another example Vermorel provides is the common practice among fashion retailers to hold sales at the end of a collection to liquidate excess stock. Dynamic pricing strategies like these are already being used by companies like Amazon, whose prices change frequently based on demand and supply factors.
Dynamic pricing is based on several factors, including historical product performance and demand levels. Vermorel points out that pricing serves multiple purposes, one of which is to send a message to the market about a brand’s positioning. For example, a cheap brand should not raise its prices too much during stock-outs, as it may harm its image.
Another important aspect of pricing is its ability to help manage supply chain challenges and uncertainties by accelerating or slowing down demand. Dynamic pricing can lead to better resource allocation and improve the overall customer experience. If two nearly identical products are available but one is close to running out, raising the price of the scarce item can help direct customers who are indifferent between the two to the more abundant product, while those who specifically want the scarce product are still able to access it at a premium.
When it comes to integrating pricing with forecasting, Vermorel suggests abandoning the idea of static forecasts. He critiques the traditional approach, which involves predicting a specific number of future sales for a product. This approach fails to account for the fact that pricing is a lever that can be used to influence demand. Instead, he argues that forecasts should take into account not only future uncertainties but also the decisions that companies can make themselves to shape demand through pricing adjustments.
Vermorel explains that pricing is not uncertain but rather undecided, and businesses should have the ability to adjust their pricing based on market feedback. This can involve increasing prices to reduce demand when running out of stock or decreasing prices to increase demand.
A probabilistic approach, which accounts for a range of possibilities, can help businesses better understand the effects of pricing changes. Vermorel uses the example of McDonald’s Big Mac, explaining that small changes in price have a measurable impact on demand due to the product’s scale and price sensitivity. However, in typical supply chain situations where fewer units are sold, pricing analysis becomes fuzzier. Probabilistic forecasting helps address this uncertainty by allowing businesses to shift their demand forecasts up or down while remaining relatively uncertain.
Vermorel highlights the importance of bringing together the teams responsible for engineering prices, planning, purchasing, and production in order to optimize pricing. By doing so, businesses can execute better optimization strategies. The data requirements for pricing optimization are not extensive; historical sales data, promotion history, and competitor pricing information (if available) are generally sufficient.
The interview also touches on the importance of competitor pricing. Vermorel explains that the impact of competitor pricing depends on the industry, with luxury brands being less affected by competitor pricing, while industries with easily substitutable products being more sensitive to competitors’ prices.
Vermorel points out that some companies, such as Amazon, are already using pricing optimization effectively. He predicts that pricing optimization will become more prevalent across various verticals in the future. In summary, pricing optimization is an integral part of demand forecasting, and businesses should not ignore pricing effects in their supply chain optimization efforts.
Full Transcript
Kieran Chandler: Today on Lokad TV, we’re going to learn how the advances in machine learning and Big Data have changed all this, and it is now possible for pricing optimization for modern companies. So, Joannes, perhaps you could tell us a little bit about how the classical approaches that many companies are taking to pricing today.
Joannes Vermorel: The most spectacular thing about pricing is typically how disconnected it is from demand planning in general. You have, let’s say, the marketing team that is going to assess what is a good price point for the products, and then, in largely isolation, frequently completely isolated, you will have the planning teams that will decide what will be the demand. Maybe the sales team will coordinate as part of an S&OP process, but numerically, the pricing is mostly ignored once somebody has made up their mind on the price point of a given product.
Kieran Chandler: We already covered S&OP processes in a previous episode, and the pricing itself, how does that work as a mechanism?
Joannes Vermorel: Economics 101 is that, typically, as you increase the price, you decrease the demand. In theory, there is a class of goods known as Veblen Goods, which can be the opposite because a higher price tag makes them more appealing. But, the experience at Lokad is that those goods are exceedingly rare. So, for pretty much anything that is like normal stuff, if you increase the price, you decrease the demand, and this is relatively obvious. Yet, it gets very tricky because, when you think in terms of classic supply chain optimization, you focus a lot on the demand, but pricing is nonexistent. Most supply chain planning tools do not even take into account pricing. So, it should; you’re completely blind. And obviously, it means that whenever there is a strong pricing effect that comes into play, well, all the planning and all your forecasts are just dramatically off.
Kieran Chandler: So, it’s one of the reasons a lot of these supply chain tools don’t take into account pricing because there are so many variables. There’s things like marketing influences and things that can get in the way of demand. Is that the reason why pricing isn’t really taken into account?
Joannes Vermorel: Yes, but also, just because the amount of variables itself is not necessarily such a problem. With the processing power of modern computers, you can deal with hundreds of thousands of variables. This is not, in itself, a showstopper. Yet, those classical models tend to be completely simplistic. You think of the demand as a pure moving average with just a seasonal coefficient, and this is it. You’re not even taking frequently into account things as basic as stockouts. So, obviously, if you have stockouts, you observe zero sales, but that doesn’t mean that you have zero demand. So, let alone pricing effect. Indeed, pricing is more subtle, and ideally, you would take into account the price of your competitors, which nowadays, with web crawlers, can be monitored online as well. It has never been easier to actually get access to the data. There are even companies that are specialized in retrieving those prices for you.
Kieran Chandler: So, if it’s not such a problem and it is possible to keep track of all these variables, how can we begin to work towards pricing optimization? Where do you start off?
Joannes Vermorel: Your supply chain optimization needs to factor in the pricing optimization. You cannot dissociate the two, you see, because the traditional perspective is that you will have people that, in isolation, decide a price and then other people in isolation that decide what will be the future demand. But the reality is that those two elements are completely coupled with one another, and if you cannot disentangle them, they are fundamentally entangled. So that means that the starting point is to acknowledge, process-wise, that those things will go hand in hand. And if you, by design, separate them, then it doesn’t matter how smart your machine learning stuff can be or all your numerical recipes. Once they are completely separated by design, then you cannot acknowledge anymore, from a demand forecasting perspective, the fact that the pricing exists because you’re just blind, numerically speaking. So, the idea is that pricing is, instead of being something that you actually forecast, it’s something that you actually engineer yourself. An example would be pretty good, and there are companies here that are actually doing this well today. E-commerce companies, as usual, are ahead of the pack when it comes to very smart, quantitative optimization. They are aggressively quantitatively optimizing the price. So basically, if they know they are heading for stock-outs, there is no point in rushing to liquidate the stock that remains, because anyway, you will end up with a stock-out. You can inflate your price a bit, and you will still face your stock-out, but at least you will have made bigger margins on those last few units. Conversely, if you see that your competitors are all suffering from stock-outs, there is no point in rushing to the point where you will face a stock-out yourself just because of the extra influx of demand. Conversely, that’s what all fashion retailers do if you end up with too much stock for a given product at the end of the collection: you do a sale to basically liquidate what remains. So, dynamic pricing is already used in production. Amazon, for example, if you look at their prices, they are very much in flux. The prices change from day to day and even from hour to hour, especially during very busy periods like pre-Christmas.
Kieran Chandler: And so, the basis for that dynamic pricing is it basically how products have performed historically and sort of the demand levels at different points?
Joannes Vermorel: The idea is that pricing is many things. Pricing is one message that you send to the market, so that part is relatively rigid. You don’t want to steer too much away from the message. If you position yourself as being a cheap brand, you do not want to vastly increase your price because of stock-outs. So there is this messaging part that’s important, that’s the long-term component. But also, there is the fact that pricing is a fantastic mechanism to accelerate or slow down demand to help you cope with your supply chain challenges and difficulties. The future is uncertain, and it’s very hard to have an accurate forecast when the markets are extremely erratic. So having dynamic pricing is a way to make better resource allocation and even to deliver better service for your clients.
There is no point, for example, if you have two products that are almost perfect substitutes but one of them is going to run out. It’s not naturally a perfect substitute in the eyes of all your clients. If you raise the price of one of those two products, the clients who do not mind using one product over the other will just switch to the other one. The clients who care will still have access to the product they really seek, and they are even willing to pay a little premium.
Kieran Chandler: Now let’s sort of look at things from the perspective of forecasting. I mean, how can you possibly build a forecast that works well in tandem with pricing?
Joannes Vermorel: The idea is that you have to give up on the idea of having a static forecast. That’s also one of my big concerns with a classical perspective on forecasting. You’re supposed to say, “this is it, this is the future, we will sell 1,000 units of this thing,” and the answer is, well, it depends. Your pricing is fundamentally a lever that can influence the demand. So, what you’re trying to forecast is fundamentally driven by something that you can do. It’s not about capturing future uncertainty; it’s also about capturing future decisions that you will make yourself. And those decisions, they are not uncertain; they are undecided. There is no uncertainty about your pricing; your pricing is exactly what you want it to be. If you want to raise your price, you can raise your price, except if you’re in a highly regulated domain. Mostly, you have complete freedom over your pricing, so there is no uncertainty, but you might change your mind.
If you think of a static forecast, it’s as if you were saying, “well, my price will not change no matter what,” which is bad. It’s much smarter, and in business, it means more profitable to be able to adjust what you’re doing depending on the feedback from the market.
Kieran Chandler: So, basically, if you want to reduce demand because you’re running out of stock, you’ll increase the prices, and then if you want to increase demand, you can reduce the price?
Joannes Vermorel: Exactly. And this is where a probabilistic approach fits in, having this range of possibilities might work well. Probabilities really help because, unless you’re selling a product like the Big Mac in a large country like Germany where you have fantastic quantities, if you increase the price of the Big Mac, every single cent will probably have a measurable impact on the demand. So here, you have a product that is sold at scale and is highly price-sensitive. People should be able to measure the exact elasticity. I’m pretty sure there are smart people at McDonald’s who know the price elasticity of the Big Mac country-by-country.
But that works well because they have a huge amount of data. In typical supply chain situations, you’re not selling millions of units a week of a given product. Frequently, you end up with just a few hundred units a week or even less than that, so it means that your pricing analysis is going to be very fuzzy. Where it does help is that having a probability forecast means that you can actually have a forecast where you just shift your demand forecast up or down, but it’s still relatively uncertain. The further away you are from your usual price point, the more extra uncertainty you will add.
Because if you just move the price a little bit, it turns out that, most probably, your historical data is relevant as a baseline to guess what will happen. But if you were to think of what would happen if your price were, for example, 10 times cheaper, chances are that you have zero historical data to support that, just because you never did it. So, because it was not profitable, none of your competitors were about to do it either, so you just don’t know. Nobody knows. Maybe there will be 100 times more demand for your product if it was 10 times cheaper, but it’s just not a point that you have ever explored. So, you just don’t know, and that’s an extreme situation, but the same idea applies.
Kieran Chandler: The further away you are from your usual pricing practices, the less information you have coming from your store, so the more uncertainty you have.
Joannes Vermorel: Okay, so in the example of the Big Mac, what you’re saying is if you’re changing by a couple of cents here or there, you’ll have more of an understanding, whereas if you’re changing by a couple of euros, a larger change, then you’ll have much less of an understanding of the possible future.
Kieran Chandler: Let’s talk about some of the technical requirements for pricing optimization. I mean, what does a company need? You said this about the Big Mac and McDonald’s having lots of historical data. What kind of technical requirements does a company need in order to optimize their prices?
Joannes Vermorel: Your requirement to get started is first to bring together the teams that are engineering the price and teams that are typically doing the planning, and the ones that are doing all the purchasing and production plans. You see, because the problem is that you need to, as a requirement, bring all those functions together; otherwise, you would not be able to perform the optimization. It’s like, by design, you prevented yourself from even being able to execute that. Then, I would say the process requirements, in terms of data, you don’t need that much. Actually, your traditional historical sales data is good. You need to have the usual suspects: your history of promotions, because promotions are like temporary price movements that are interesting to analyze what happens when you move the price. You need to have historical stockouts as well to avoid having pollutions and outliers that you cannot explain price-wise. Ideally, if you can have the price of your competitors, at least on the web, that really helps. That’s not strictly required, but it really helps to understand, I would say, the non-linearity effects when you get lower than the competitor or the competitor gets lower than you. That can have little nonlinear effects on the demand that you can see, spikes and drops that are just explained because people shift from whoever is the least expensive.
Kieran Chandler: By bringing all these departments together, how does this kind of differ from a more traditional S&OP approach?
Joannes Vermorel: It differs from the fact that, fundamentally, in your pricing, you give up the idea of having a fixed price for a product. Instead, you have a pricing strategy. Let’s say you’re sourcing products in Asia, and it’s going to take 13 weeks between the time you’re passing an order to your Asian supplier to the point in time where you can actually put the product on display and start selling in North America and Europe. Why would you, at day zero, decide what your price will be when you start to sell your product 13 weeks later? You know, this decision can be delayed. If, 13 weeks later, you see that demand has surged for this type of product and your initial order that was of 1,000 units, you already realize that it’s going to be way too few. There is no point in sticking to what you had in mind at that time, 13 weeks ago. You can reassess based on the latest data that you have and decide something that is smarter.
Kieran Chandler: How important is keeping track of competitors’ pricing, especially now that we have easy access to their prices online?
Joannes Vermorel: It really depends on the verticals. For example, if you are a luxury brand, it doesn’t really matter. A luxury brand like Louis Vuitton does not decide to lower their price just because Cartier did. They are both premier luxury brands that focus on having the best quality for their products, and they are going to price accordingly. A luxury brand is supposed to be something that does not have a substitute. On the other hand, you have products that are nearly perfect substitutes. If you’re buying sugar by the kilogram for a restaurant, it doesn’t really matter who your supplier is, as long as it meets the quality standards.
Kieran Chandler: Looking ahead to the future, can you envision a time when companies are setting pricing points with a solid understanding of their resulting profits?
Joannes Vermorel: Some companies, like Amazon, are already doing it. We have a few clients in very aggressive e-commerce who are also doing it, even if they don’t advertise it as much. In more sophisticated verticals like aerospace, they are already doing it as well. Airlines have been doing yield management for decades to sell tickets, and even with aircraft parts, they are doing it. What I see is that it’s coming to pretty much all other verticals, but it may be more or less applicable.
Kieran Chandler: To conclude, what is the key lesson we should take from today about pricing optimization?
Joannes Vermorel: Pricing optimization is an integral part of demand and demand forecasts. There is no demand without price – price defines the demand and vice versa. If your supply chain organization is trying to optimize while ignoring pricing effects, you are missing an elephant in the room. It doesn’t matter if you micro-optimize your moving average; there is still this elephant that you’re ignoring, and that’s bad.
Kieran Chandler: Hopefully, nobody’s missing the elephant in the room anymore. That’s everything for this week. Thanks very much for tuning in, and we’ll see you again next time. Goodbye for now.
