00:00:08 The problem of separate entities in pricing and planning.
00:01:02 Traditional software companies and their approach to pricing and forecasting.
00:03:24 The accidental convergence of pricing and forecasting technologies.
00:05:46 Data integration as the starting point for convergence.
00:07:33 Addressing the meta problem and applying it to both pricing and planning.
00:09:20 The breakthrough in deep learning and predictive demand models.
00:11:22 The convergence of planning and pricing models in different industries.
00:13:36 The benefits of having one team to handle both planning and pricing.
00:15:00 Amazon’s successful approach to dynamic pricing based on stock availability.
00:16:01 Discussing the shift towards pricing optimization in e-commerce.
00:17:03 The accidental development and convergence of pricing and planning solutions.
00:19:01 The challenges of traditional software companies unifying pricing and planning.
00:21:20 The rise of new companies addressing pricing and planning together from day one.
00:22:37 The market dominance of companies unifying pricing and planning, like Amazon and Alibaba.


Kieran Chandler interviews Joannes Vermorel, founder of Lokad, about the significance of integrating pricing and demand planning in supply chain optimization. Traditionally, these aspects were handled separately, resulting in data silos and poor communication. Lokad discovered that pricing and planning share similar data sources and numerical recipes, leading to the development of a domain-specific programming language, Envision. By leveraging shared data storage and processing, Lokad created more sophisticated models, incorporating pricing and time effects. Vermorel believes that successful companies must unify pricing and planning, as they are interconnected aspects crucial to almost every industry.

Extended Summary

In this interview, Kieran Chandler, the host, discusses with Joannes Vermorel, the founder of Lokad, the importance of integrating pricing and demand planning in supply chain optimization. Historically, pricing and planning have been treated as separate entities, leading to data silos and a lack of communication between departments. The conversation revolves around how software can help address these tasks in tandem, and why the industry should view them as two interconnected aspects.

Vermorel explains that anticipating future demand is crucial to effectively serve clients. Companies need to produce or source products ahead of time to meet the demand, as instant production is not yet possible. However, demand is also strongly influenced by price. If a product is too expensive, demand will be low, while a competitive price can lead to significant demand.

When Lokad was founded, the software industry was divided into two camps: forecasting companies, which focused on predicting demand, and pricing optimization companies, which ignored forecasting and planning. For the first five years of Lokad, the notion of prices and pricing was absent from their focus, as the company primarily dealt with forecasting. Similarly, pricing optimization companies did not consider forecasting and planning as part of their scope.

Vermorel emphasizes that demand and pricing are profoundly interconnected. Companies need to anticipate demand while also pricing their products correctly. Historically, organizations kept pricing and demand planning separate, utilizing tools like price forge and sales cast. However, at some point, Lokad decided to integrate these two aspects.

The decision to join pricing and demand planning came from the realization that certain established practices can become so ingrained in the overall mindset that people fail to see the connections between them. By integrating these two aspects, Lokad aims to improve supply chain optimization by addressing the interdependence of pricing and demand planning.

Initially, Lokad developed two separate products, Salescast for forecasting and Priceforge for pricing, with completely distinct technologies. The company decided to bring these two apps together because they noticed that both used similar data sources, such as sales history and product catalogs, as well as the same transactional data from DRP, WMS, and e-commerce platforms. They decided to create a unified architecture for data storage and processing, but at this stage, they had not yet realized the deep connection between pricing and sales forecasting.

The first degree of convergence between the two solutions came about accidentally. The company realized that the numerical recipes for pricing and sales forecasting were similar, and they started to work on a domain-specific programming language for predictive optimization of supply chains. This led to the development of Envision, which began on the pricing side.

At this time, Lokad was less familiar with pricing than with planning, so they decided to tackle a meta-problem: how to develop numerical recipes faster and more reliably. They hoped that by improving the time to market and the reliability of new numerical recipes, they would be able to succeed in the pricing domain. As they experimented with these techniques, they discovered that the same programmatic approach they were using for pricing made sense for the planning side as well. This realization led to a second degree of convergence, with both pricing and planning sharing the same data crunching layer and programming language.

However, the solutions still ran separately, and it was during the deep learning era that Lokad had its first breakthrough. By incorporating more advanced predictive demand models, they found that integrating pricing and time effects led to more sophisticated models for both domains. For example, the willingness to pay for a product might be seasonal, so it made sense to incorporate these factors into both pricing and sales forecasting models.

Lokad’s journey in combining pricing and sales forecasting began with the observation that both solutions shared similar data sources and evolved through several stages of convergence. By leveraging shared data storage and processing, as well as a unified programming language, Lokad managed to create more sophisticated models that incorporated both pricing and time effects. While the discussion in this portion of the interview does not provide a conclusion, it highlights the iterative process Lokad underwent to develop and refine its approach to supply chain optimization.

Vermorel shares how Lokad began to realize that willingness to pay was seasonal, and how this affected their approach to supply chain optimization. As they worked on their models, they noticed that the pricing dimension was becoming more prominent. This was especially true in industries like fashion, where price is used as a lever to ensure that stocks are liquidated at the end of a collection. As they continued to work on their models, they found that the problems of planning and pricing were converging, eventually leading them to develop a single model that could address both issues.

When asked which industries this approach is most applicable to, Vermorel explains that pricing is crucial in almost every industry because it can make the difference between no margin and significant margin. Planning and pricing are often entangled, with a tight relationship between dealing with scarcity and willingness to pay. Different industries have unique perspectives on the problem, but at the core, the two issues are almost always interconnected.

However, the majority of the market still offers software focused on either pricing or planning, rather than a unified solution. Vermorel attributes this to the fact that until there is a solution, there is no perceived problem. Lokad’s approach to unifying pricing and planning emerged accidentally, as they found that their models for both issues shared many common components. As they continued to develop their models, they realized that they needed to tackle both problems simultaneously.

When asked about the possibility of software companies partnering up to provide a dual solution, Vermorel is skeptical, as the technology needed to unify pricing and planning is radically different from traditional methods. Instead, he predicts that new companies will emerge that leverage insights from the beginning to address both issues together.

Vermorel emphasizes that companies that successfully unify pricing and planning can achieve market dominance. He uses Amazon and Alibaba as examples of companies that continue to gain market share because they have gotten the fundamentals right. His conclusion is that businesses should embrace a solution that addresses both pricing and planning, as these two aspects are inextricably linked.

Full Transcript

Kieran Chandler: Today on Lokad TV, we’re going to discuss how software can deal with these tasks in tandem and why the industry should see them as two sides of the same coin. So, Joannes, previously we discussed a little bit about pricing. What’s the angle today?

Joannes Vermorel: The angle is the fact that when you think about the demand, obviously, if you want to be able to serve clients, you need to anticipate what the future demand will be so that you can produce or source ahead of time what you intend to serve. You can’t instantly 3D print things on demand, not yet at least. But obviously, demand is strongly influenced by the price. If you have something that is exceedingly expensive, demand is going to be nil, and if you have something that is exceedingly competitive, the demand can be absolutely gigantic on the grand scale.

Yet, when I started Lokad, there were clearly two camps in terms of software companies. There were the forecasting companies, Lokad was one of them, who really took the problem of analyzing the demand from the predictive perspective. And traditionally, pricing did not even exist. Prices did not even exist, and that’s kind of weird, but actually, for the first five years or so of Lokad, the very notion of prices and pricing was absent. It wasn’t even part of the roadmap. It wasn’t anywhere. Obviously, I knew like everybody that prices are things that you can see, but there was no specific connection. And then there were other companies who only dealt with pricing optimization, and from their perspective, forecasting and planning were absent. It wasn’t even part of the landscape. They didn’t care. Again, they knew that forecasting exists because they watched the weather forecast on the news, like everybody else. But, nonetheless, just like I was on the forecasting side, I was completely ignoring what was happening as far as pricing was concerned. Software pricing companies were completely ignoring what forecasting was about. And yet, when you think about it, it’s profoundly connected for the demand, and demand is super important; you need to anticipate but you need to correctly price as well.

Kieran Chandler: Yeah, and historically, you kind of kept those two entities very much separate in kind of Salescast and PriceForge, but then at some point, they kind of joined together. So why did you decide to join them?

Joannes Vermorel: That’s very interesting. You see, sometimes there are things that are so well-established in the overall thinking of your time that you can’t even see it. And when Lokad started to tackle the pricing problem, we had actually a pricing mission. I thought of it as something that was just completely separate. So, we had one product called Salescast, and it was basically the forecasting part of Lokad, and we had another product that was completely distinct, which was PriceForge.

Kieran Chandler: And we were selling those two things separately and actually even the technologies, the respective technologies, were completely separate. In terms of data analytics, how easy was it to combine the two?

Joannes Vermorel: That’s the weird thing. We brought those two apps together, not because I thought, “Oh, it’s the same problem that we’re solving, by the way, we’re solving it twice.” No, that would have been very smart, but no, we did the very dumb thing. We said, “This app is using the sales history, this app is also using the sales history. This app needs the catalog with prices, this app also needs a catalog. We don’t use the price, but it’s pretty much the same thing, it’s a catalog.” And literally, we were looking at the data and the data that was needed for, I would say, pricing analytics or for planning analytics. It was essentially the same transactional data that was coming from the same sources: ERP, WMS, e-commerce platforms, etc. So, we decided to mutualize the architecture to have the same base layer for data storage and data processing because it was the same data. But at the time, we hadn’t even started to conceptually realize that it was literally two sides of the same coin.

Kieran Chandler: So, it’s kind of the data integration and having those two sets of data in the same place was where you started. But when did the solutions start converging in terms of pricing and the sales?

Joannes Vermorel: We gradually started to realize that the numerical recipes were kind of the same, which was puzzling. What Lokad is nowadays, Envision, with a domain-specific programming language dedicated to predictive optimization of supply chain, this thing started on the pricing side of things. At the time, pricing was something very new to us, and we didn’t really know how to tackle that. So instead of solving the pricing problem, we said, “We’re not really sure how to approach that, so we are going to solve a meta-problem, which is: how can we come up with numerical recipes faster and in ways that are more reliable?” We didn’t know the recipes yet, but if we could improve our time to delivery in terms of how much time it takes to deliver a new numerical recipe that we haven’t invented yet to the market, and then how much time it takes to get this thing running reliably in production, then we would be kind of okay. So the thinking went into, “We are going to do that for pricing because I knew at the time much less about pricing than I did about planning.” And yet, when we started to tweak, we were dealing literally with the same data sets, and we realized, “Oh, it’s very interesting. What we’re doing on the pricing side kind of makes a lot.”

Kieran Chandler: It’s quite a challenge to deal with the planning part of supply chains, as they are made up of a series of accidents. Some companies have one ERP, some have two, some have one per country. Sometimes the e-commerce platform is integrated, sometimes it’s separate, and sometimes there’s a WMS. It’s very heterogeneous. We realized that we needed a completely bespoke, programmatic approach for both the pricing stage and the planning stage. This allowed us to have a second level of convergence with the same base layer for data and the same data crunching layers. But how well does it actually work? Is it more of a rough approximation?

Joannes Vermorel: It took us some time to improve the system. Initially, we mutualized the data layer, as we were dealing with the same data sources, and we also mutualized the data crunching approach. We realized that both sides needed to be programmatic, and we had distinct sets of numerical recipes. The first breakthrough emerged from the deep learning era, where we started playing with more advanced demand models. On the planning side, if you want a sophisticated model, you need to integrate price, and on the pricing side, you need to integrate time effects, such as seasonality.

For example, the willingness to pay for a beach towel is higher at the end of spring than at the end of summer, because if you buy the towel at the beginning of the season, you can enjoy it for the entire summer. Conversely, if you buy it at the end of summer, your use of the product will be much more limited. We realized that willingness to pay was actually seasonal, and on the planning side, our more advanced models had the pricing dimension becoming more and more prominent. In fashion, for example, you want to use price as a lever to make sure that you liquidate your stock entirely at the end of a collection.

So, we saw these convergences, and in the middle, we had a predictive model of some kind for demand. We realized that these models were really converging, and the loop was closed with the differentiable programming era.

Kieran Chandler: So, is the fashion industry the one this solution is most applicable to? Because if you think about the airline industry, their prices are fluctuating by the second, and then you’ve got other industries where the pricing hasn’t changed in years.

Joannes Vermorel: The solution is literally applicable everywhere. Pricing is always super important when you’re selling something because it makes a difference between no margin or tons of margin. And there is a tight relationship between planning, where you’re dealing with scarcity, and pricing, where you’re dealing with willingness. The two are actually completely entangled. If you anticipate more demand, you can produce more at a lower price and thus have a better margin and potentially outcompete your competitors on price. If you have a super high price, maybe you can play on scarcity and make your products even more desirable, like luxury brands do. They want pricing to increase over time and play on the idea of scarcity. Every industry has a fairly distinctive perspective on the problem, both on the planning side and on the pricing side, but the fact that those two problems at the core are completely entangled is almost always true.

Kieran Chandler: So, to get the best results, the planning teams need to work very much with the pricing teams?

Joannes Vermorel: I would disagree with that. My initial perspective was: why should you have a planning team and a pricing team? Why have two? That’s the exact same mistake I made initially with separate software for each. It’s the same problem; you’re just looking at it from two angles. There should be one team.

Kieran Chandler: Are there any companies that you’ve observed doing that well and working with just one team?

Joannes Vermorel: Yes, there is the usual suspect: Amazon. They are very smart, and they do things that are very obvious when you think about it. For example, if something is running out of stock, they’ll raise the price. There’s no reason to rush towards a stockout. The price is just a way to shape the demand so that you make the most of the stock you have. You can literally see it in action.

Kieran Chandler: Amazon did that before Christmas with the toys they’re selling. My daughter, who is 10 years old, is a big fan of LEGO toys, and when you look on Amazon, literally every couple of hours when there is a bit less stock, they will actually bump up the price of certain LEGO boxes. I’m pretty sure they have some basic heuristics that inflate the price when they’re running out of stock, and it makes sense. It’s literally common sense, but it’s not super obvious. I’m pretty sure that just through public observation, they have an algorithm that entangles stock availability, which is a projection, with pricing optimization.

Joannes Vermorel: But then, from the local customer base, I see that other very smart, tech-minded, aggressive e-commerce companies are also already doing that, so it’s not just Amazon. It’s also the biggest challengers in terms of supply chain of Amazon today that are doing that.

Kieran Chandler: Okay, so if this is such a good way forward, then why is the majority of the market still offering software that is just focused on pricing or just focused on planning?

Joannes Vermorel: Because until you have a solution, there is no problem. That’s a puzzling thing. You would say, “Oh, there is a problem, so people should be looking for a solution,” but that’s not how it works. Nobody is looking for a new way to reinvent cars if we had anti-gravity engines. We don’t have anti-gravity engines, so nobody cares about finding solutions for a problem that is not there yet. If you don’t have a solution to frontally address pricing and planning, is there even a point in thinking about the potential solution for this problem? I would say no.

How did Lokad come to that? Did I think historically, in a stroke of genius, “Those are two sides of the same coin, I need to address that and think of those two problems as one”? Absolutely not. What I did was pretty much the opposite. We were thinking, “Well, we have a problem where there is a set of solutions known in the literature for forecasting, and there is pricing, and there is a set of solutions known in the literature for that as well.” So, I will actually have one product that implements all solutions or maybe what I think to be a slightly better variant of those solutions, and the same thing on the forecasting side—a slightly better variant of what is known in the literature.

As I was describing, the convergence emerged in a way that was completely accidental. It was just that we were using the same datasets, so why not? Let’s do it. Let’s merge like that. Then we see that the kind of the recipes are a bit alike, so let’s mutualize that. The two distinct software products are brought together just because accidentally they share the same components.

But you see, it was not like grand thinking; it was more like accidental development. And then, five years down the road, you have another accident where, with new machine learning techniques, you realize, “Oh, crap, it’s the same model that I end up using on the two sides of the problem.” So, I have my solution that does the two things together, and thus I realize I should be tackling those problems as one because I have a solution.

Kieran Chandler: Now, in inside it seems obvious, but looking back it wasn’t. And thus, I think most companies have kept that team separate because it is a reasonable thing to do until you have a solution to unify the thing. If you don’t have, you know, a solution that can unify pricing and planning, then basically, well, you’re stuck having two distinct teams. Because there is otherwise those people would not even be able to start working on those two problems. The first thing they would do is to internally re-split the problem. And so, if they would have like people work on one side of the program and another set of people who work on the other side of the department, because they don’t have, you know, and that’s that would become the theme of planning and pricing.

Joannes Vermorel: Okay, sounds like a bit of a happy accident then.

Kieran Chandler: And maybe if we look forward to the kind of the future, can you see kind of some of those software companies kind of partnering up and joining together and there’s providing a dual solution?

Joannes Vermorel: I don’t think so because literally the sort of things that work to unify the two problems are completely unlike what was traditionally done for pricing and what was traditionally done for forecasting. That’s a bizarre thing that this the class of solutions we have now to do to address those two problems at once, you know, and frontally optimize both prices and the plan which is basically how much you produce, how much you buy, how much stock you put all across your networks. The sort of technology that we developed are radically different from what we had historically either on the pricing side or on the planning side. So, literally, we had to entirely scrap and discard our respective technologies both on both sides. So, imagine, you know, would you actually bring two companies, two software companies together so that they can both agree that they are both going to discard the entire tech stack to rewrite everything together? Nah, I don’t think so. That sounds like super weird. So, probably what will most likely what we will most likely see…

Kieran Chandler: I don’t know, I mean, it’s very hard to fork out to predict the future, especially the future. But what we will probably most likely see is new classes of companies that, from the start, leverages insight to say, “We are going to address those two problems together,” and from day one decide to go for a solution that would encompass the world spectrum of problems.

Joannes Vermorel: Okay.

Kieran Chandler: If we start kind of wrapping things up today, um, what’s our main conclusion? Is it kind of that a company which unifies pricing and planning has a lot more control?

Joannes Vermorel: No, it’s just that, as usual, the companies who have unified that will just, you know, keep uh, pushing, I would say, to bankruptcy companies who don’t. You know, it’s literally, um, if you’re able to execute that, it will not just be more control; it’s just, it’s literally market dominance. And, again, when we see it’s one more ingredient where you see why is Amazon still gaining market share, although they are, like, one of the largest of the largest companies on earth, and they’re still winning market share, and same for Alibaba in China. Or, and you see, well, because they have those fundamental things right, and thus, over time, it just, you know, it’s just going to crush all the companies that do not. So, why should you do that? Well, I would say just question yourself about, can you really think that you can, you know, in thinking the demand that you can literally isolate the pricing part from the planning part? Is it reasonable business-wise? And usually, when people, you know, stop for a second and think, no, those things are completely entangled. So, yes, if I have a solution to just, you know, embrace that, I should do it. And, uh, my conclusion would be, for people watching us, is to do it.

Kieran Chandler: Always seems to come down to survival of the fittest, absolutely. Okay, that’s everything for this week. Thanks very much for tuning in, and we’ll see you again in the next episode. Thanks for watching.