00:00:07 Leveraging weather data for optimizing purchase decisions.
00:00:41 Weather-sensitive products and their correlation to weather.
00:02:00 Historical context of weather data startups and their niche status.
00:03:23 Electricity providers efficiently leveraging weather data for consumption prediction.
00:06:02 The complexity of weather data compared to supply chain data and technical challenges.
00:08:00 Weather’s impact on IPO markets and demand for products.
00:10:00 The limited benefit of weather forecasts for supply chains.
00:12:45 The usefulness of past weather data in explaining product demand.
00:14:00 Using climate change to improve seasonal forecasts and its limitations.
00:15:49 Comparing daily temperature fluctuations to long-term climate change impact.
00:17:01 The importance of using transactional data for supply chain decisions.
00:18:04 Leveraging web traffic data for better supply chain understanding.
00:18:36 Using competitive intelligence in supply chain decision-making.
00:19:27 Conclusion: Prioritizing other data sources before exploring weather data.


Joannes Vermorel, founder of Lokad, discusses with Kieran Chandler, the host, the potential and limitations of weather data in supply chain optimization. Vermorel highlights that weather data has strong explanatory power, but it may not be very useful for refining forecasts. The multi-dimensional nature of weather makes it challenging to incorporate it effectively into forecasts. Vermorel emphasizes the importance of prioritizing transactional data and non-transactional data, such as web traffic and competitive intelligence, over weather data and social data. Weather data may be more useful in analyzing past sales performance. Despite the challenges, as weather forecasting technology and supply chain management evolve, there may be greater opportunities for businesses to leverage weather data for strategic decision-making.

Extended Summary

In this episode, Kieran Chandler, the host, interviews Joannes Vermorel, founder of Lokad, a software company specializing in supply chain optimization. They discuss the role of weather data in optimizing purchase decisions and the complications that arise when combining different forecasts.

The idea behind using weather data for supply chain optimization is simple: many products are highly weather-sensitive, and their demand is influenced by weather conditions. Examples include barbecue meat and even cars, as people in different climates may choose different vehicles. Many supply chain practices aim to improve forecasts by incorporating weather data.

A decade ago, there was a buzz around weather data and startups focusing on refining forecasts. However, despite the initial interest, using weather data for supply chain optimization has remained a niche area. Lokad conducted several missions for large companies about a decade ago, but these efforts eventually died down because it was too much effort for the benefits gained.

One of the key insights from Lokad’s experience is that weather data is incredibly complicated. It is a multi-dimensional problem with a high degree of geographical diversity. Weather conditions can vary significantly even just 20 kilometers apart due to factors such as altitude. Additionally, weather is not uniform throughout the day, making predictions more challenging.

However, there are industries that have successfully leveraged weather data, such as electricity providers. They use this data to predict electricity consumption, allowing them to provide just enough power to meet the grid’s needs at every minute of the day. Electricity storage is inefficient and impractical, making precise forecasts essential for managing supply.

While the concept of using weather data for supply chain optimization seems logical, it has proven to be complex and has remained a niche area. The weather’s highly variable and multi-dimensional nature makes it challenging to incorporate into forecasts effectively. However, some industries, like electricity providers, have found success in using weather data for their specific needs.

Vermorel explains that weather data is highly complex due to its geographic specificity and the various factors that must be considered, such as rain, wind, humidity, and sunlight. This complexity makes weather data significantly larger and more difficult to work with compared to traditional supply chain data.

While obtaining weather data has become easier thanks to cloud providers, processing and correlating this data with sales patterns remains a challenge. Vermorel highlights that weather can have highly localized effects on sales, and businesses must navigate various heuristics to account for these nuances.

In response to the suggestion of using simpler, more reactive triggers based on temperature, Vermorel acknowledges that temperature can be an important factor but emphasizes that it is not the only one. For instance, a hot, rainy, and windy weekend might not lead to a spike in demand for barbecue products. Furthermore, Vermorel points out that customers monitor weather forecasts just like businesses, leading to potential shifts in buying patterns based on weather expectations.

Looking at the short-term accuracy of weather forecasts, Vermorel explains that forecasts beyond ten days are typically not very useful for optimizing supply chain decisions. During certain events like heatwaves, the forecast only truly matters at the beginning and end of the event, resulting in a narrow window of usefulness.

Discussing the future of supply chains and weather forecasting, Vermorel acknowledges the potential for more accurate forecasts to have a greater impact. However, he highlights that the most interesting use of weather data may actually be in analyzing past sales performance. For example, understanding whether a successful ice cream sales campaign was due to effective marketing or simply a heatwave in Paris can provide valuable insights for businesses.

Incorporating weather data into supply chain optimization is a complex and challenging task, with potential benefits in understanding past sales performance and limited short-term forecasting applications. As weather forecasting technology and supply chain management continue to evolve, there may be greater opportunities for businesses to leverage weather data for strategic decision-making.

Vermorel shares his insights on the importance of weather data, the impact of climate change, and the types of data that companies should focus on for better supply chain optimization.

Vermorel explains that although weather data has strong explanatory power, it may not be as useful for refining forecasts. Many supply chain practitioners find it helpful for understanding the past but not necessarily for making future predictions.

When asked about the impact of climate change on seasonal forecasts, Vermorel points out that, while it is a significant global issue, its effect on supply chain optimization is minimal due to the difference in timeframes. Climate change predictions span centuries, whereas supply chain decisions focus on months or a few years. As a result, the effect of climate change on supply chain optimization is relatively small.

Vermorel highlights the importance of transactional data for supply chain optimization. He notes that many companies are not effectively using this data, as they often do not quantify their stock and stockout costs in financial terms. By making the most of transactional data, companies can optimize their supply chain decisions.

Additionally, Vermorel suggests focusing on non-transactional data that is easy to collect and highly relevant to a company’s supply chain. Web traffic data, for example, can provide valuable insights into customer behavior and product performance. Competitive intelligence, such as competitor pricing, is another valuable data source, although it can be more challenging to collect.

Weather data and social data can be useful, but Vermorel recommends that companies prioritize transactional data, web traffic data, and competitive intelligence data first. Weather data and social data should be considered when a company has already made the most of other data sources and has a large data science team.

Weather data can be useful and interesting for understanding past supply chain performance, but there are other data sources that are more important for optimization. Companies should prioritize transactional data, web traffic data, and competitive intelligence data before exploring weather data or social data for supply chain optimization.

Full Transcript

Kieran Chandler: Today, we’re going to discuss where some of the complications can lie when combining different forecasts and understand whether this data can be leveraged in order to give us any valuable insights.

Joannes Vermorel: The idea is a simple one: there are tons of products that are highly weather sensitive in terms of demand. You can think of maybe barbecue meat that you’re going to buy for the next weekend if there’s sunshine, and you’re likely to do a barbecue. But more generally, there are entire classes of products that are highly weather sensitive, and to some extent, pretty much everything is kind of weather sensitive. I mean, even your car; if you’re in a region that is very cold, you may not pick the same car as if you’re living in a region that is very hot or where it rains a lot, for example. So it is of interest to have this data, and obviously, because it’s an obvious factor, many supply chain practices think of improving the forecast by trying to include this data from the weather, in particular from looking at weather forecasts.

Kieran Chandler: Surely the more you actually know, there’s a better place to make a decision based on the future. So how does it work in practice?

Joannes Vermorel: That’s interesting, especially the part about how it works in practice. When I created Lokad ten years ago, weather data and weather startups were all the rage. At the time, probably in France, there were three startups that were all about refining forecasts with weather data, and there were probably something like 20 in the US. So it’s interesting because that was a decade ago, and it was a small buzzword of its own, having everything that was weather-enabled or weather-powered in terms of analytics. But what is interesting is that, in practice, it has remained incredibly niche. Even at Lokad, we did quite a few missions for large companies about a decade ago, and it died down. The long story short is that it’s way too much effort for what it’s worth.

Kieran Chandler: So what did we actually learn when we were putting it into practice all those years ago?

Joannes Vermorel: We learned quite a lot. We had missions with a large European electricity provider, which I believe is exactly the sort of class of company who are nowadays leveraging weather data very efficiently. Electricity providers are leveraging weather data to predict electricity consumption so that they can really feed just enough to what the grid needs at every single minute of the day. By the way, you cannot stock electricity, or actually there are ways, but it’s very inefficient, very slow, and impractical in practice. So you need very precise forecasts. But back to weather data, the key insights were many. One is that weather is just so incredibly complicated. People don’t realize that it’s such a multidimensional problem. You have the geographical grid, and one thing about the weather is that it’s a super local thing. I was a bit surprised to realize how different the temperature can be just 20 kilometers apart. You can literally have a 15-degree difference 20 kilometers apart because one place is just a kilometer higher in altitude compared to the other place. So you have super diversity in terms of geography, but also the weather is not a thing for the day. When you look at TV. that you will have like five instances of rain, ten minutes each, spread during the day. So it’s a very precise thing that changes literally minute by minute. So okay, you have the geography, then you have the time with a granularity that is super thin. Obviously, it makes a lot of difference whether it rains during the night or during the day, you know, these sort of things. But then the weather itself, it’s not just about the temperature and whether it rains or not. There is like half a dozen of metrics like wind, humidity, light, wind speed and direction, and whether it will blow continuously or not. So it’s a very multidimensional thing and as a result, I would say it’s a small world of its own. I mean, you’re already dealing with all your supply chain data, which is already kind of very complicated, and then you just discover that there is this weather data set that is just next to your supply chain data, and it’s something that is literally at least ten times more complicated than all your supply chain data put together.

Kieran Chandler: Yeah, so there are obviously these complications, but they are complications that are very much dealt with. They’re very much collected, and so what kind of technical challenges does that introduce because they’re things that we kind of start to know now?

Joannes Vermorel: Yes, I mean first, you end up with a volume of weather data that is literally 10 to 100 times larger than your supply chain data. So, it’s just that in terms of software engineering, you end up with something that was designed to deal with the supply chain and its scale, and then you realize that if you want to deal with the weather data, you need to handle literally 100 times more data. It’s a lot of friction in that. I mean, obviously a decade ago, it was kind of a problem of its own to just get access to the weather data. This problem nowadays, with many cloud providers that sell you the data directly on the cloud, has become much easier. But crunching that data remains very complicated, especially because, again, this is not about crunching the weather data to do weather simulations. It’s about crunching the weather data to do something that is still very niche, which is trying to correlate this data with sales patterns. And again, just remember, weather is super local. For example, if you have a market that attracts people from a wide area, like 30 or even 50 kilometers around, you don’t have only one weather to consider. You might have to consolidate data on a larger geographical area, but you will have to figure out all the necessary heuristics on your own.

Kieran Chandler: But we’re starting to get very complicated here, looking at different altitudes and different granularities. Is there not something a bit more simplistic we could introduce, maybe something more reactive, like instead of a min/max, you could have an automated purchase order as soon as the temperature reaches a certain amount of degrees? Is that not something interesting?

Joannes Vermorel: I mean, it is interesting indeed. With temperature, you have the first measurement of interest, absolutely. And indeed, you can start by focusing just on temperature and average temperature during the day. Yet, it’s very far from reflecting everything you need to know. If we go back to the example of having a spike of demand for meat because people expect to have a barbecue if the next weekend is very hot, but also very rainy and windy, you might not have such a good spike for your barbecue products. So, it’s again, yes, the temperature is important, Weather forecasts make it clear that this weekend or the weekend after, with high probability, it’s going to be a very nice weekend. Then everybody knows it, and thus you have clients that will start buying the weather-sensitive products. If you just analyze the data with only like one or two days of lag, you have also information about the weather. Remember that weather forecasts are not very accurate beyond about ten days from now, so you can’t optimize with weather forecast decisions that are going further into the future than ten days. After ten days, you pretty much revert back to the seasonal average as if you didn’t know anything. It’s still something of interest, but it’s also very short-term. Just consider that when you, for example, enter a heatwave like in Paris, the heatwave may last for a couple of weeks. The period of time where the forecast actually matters is not the three weeks of the heatwave, but rather the three days where you are entering the heatwave and then maybe the two or three days where you’re exiting the heatwave. So, it’s a very narrow point of time during the year where you even have an edge, truly.

Kieran Chandler: Okay, so the window is very small, and collecting all this data doesn’t really give us that much of a benefit. If we look ahead at some of the supply chains of the future and envision that one day we’re going to have future weather forecasts that are incredibly accurate and can look further into the future, and we’ve got greater control of our supply chains as a whole, could you envisage that day when a weather forecast would be more useful for trying to leverage the weather data to forecast better?

Joannes Vermorel: It’s possible. There are companies, like electricity providers, who are doing that with very high efficiency. For them, the weather data is an input that is very valuable and helps refine their accuracy by a significant margin. But the interesting thing is that the most useful application of weather data is not looking into the future; it’s actually looking into the past. For example, if you are selling ice creams and you trigger a new commercial campaign for your products and it sells very well, does it sell well because the campaign was good or just because there was a heatwave in Paris and pretty much everybody selling ice creams had a good period during that part of the summer? Looking back at the weather data can be very useful for explaining the demand for your products. Here, you don’t have to deal with the intricacies of granularity. You can aggregate data over longer periods of time or extended geographies, and it will give you something that has strong explanatory power.

Kieran Chandler: That’s probably why people, including many supply chain practitioners, are so enthusiastic about the idea of using weather data to refine the forecast. Even if that part doesn’t work out so well, it’s because weather data is very useful for explaining the past. How about things like climate change? We’ve got this understanding that the world is gradually getting warmer, and that’s going to affect our seasonal forecasts going forward. Can we leverage this insight? Is there anything of use?

Joannes Vermorel: Unfortunately, there are orders of magnitude at play. Even the most pessimistic climate change predictions are considering something that, over the course of many years, is still relatively small compared to the daily fluctuations we see in weather.

Kieran Chandler: Of course, during the 21st century, we’re talking about a couple of degrees of difference. By the way, if you think of geological time, it is a very steep evolution of the climate. If the earth globally heats up or cools down by two degrees over a century, it’s a lot. It’s very significant. Nonetheless, we are talking about a couple of degrees over a century. Even in a very mild climate like Paris, the temperature typically fluctuates between the highest point during the day and the lowest point by about 20 degrees. We have a very mild climate. There are many regions in the world where temperatures fluctuate by 40 degrees between the highest point during the day to the lowest point during the night.

Joannes Vermorel: The bottom line is that having a climate change of maybe two degrees over the course of a century – when you think of how much change it will bring from one season to the next – is literally vanishingly small. Most consumer products enter the market, go from zero to their peak in something like twelve months, and then disappear from the market two or three years later. You realize that the impact of average climate is very small for supply chain optimization, just because we are not talking about the same timeframe. Supply chain is about optimizing decisions for the next couple of months, maybe next year, or the next couple of years if you have really big plans. But, I don’t think there are many companies who are thinking centuries ahead, except for being leaner and consuming less energy or producing less waste, which is a win for the future. However, it’s not something that you really plan with a forecast.

Kieran Chandler: So, what kind of other data should we be looking at? What’s more important than the weather data right now?

Joannes Vermorel: We have reached a stage where most companies are not even correctly using their transactional data. Most companies we talk to, when we start working with them, did not even quantify their stock-out costs in euros or dollars. Stock is a trade-off between the cost of stock and the cost of stock-out. If I’m simplifying, basically, you have to balance two risks, and you’re not even assessing them.

Kieran Chandler: Financially, one of those two risks, it’s very hard to optimize anything. So the first stage is to do the most of your transactional data that is super reliable and has exactly the granularity that matters for your supply chain decision. The second stage is to use all the data that is not transactional, but that is easy to collect and still very relevant to your own supply chain. So for example, that would be web traffic data if you’re a brand. You can look at how many visits every single page that you’re posting on your website gets, and particularly if you have pages about your products, you can look at the web traffic. And again, when the web traffic evolves, you will get this information that is tightly connected to your own products, and that’s something that you can use.

Joannes Vermorel: Maybe a third layer of data would be competitive intelligence, you know, what your competitors are doing, their pricing, etc. It’s data that is more complicated to collect, but still very intimately connected to your supply chain activity. And then beyond that, you have things like weather data and social data that are possible. We have examples of companies making use of social data, but they tend to be super tech-driven companies who have already gone through all the previous stages of making the most of their transactional data, web traffic data, and competitive intelligence data. And then you can venture into weather data and maybe social data, and those are sort of external sources.

Kieran Chandler: Okay, so to conclude today, maybe weather data is useful and can be kind of interesting, particularly when looking into the past. But the main conclusion is basically that there are other things that are more important.

Joannes Vermorel: Yeah, and if you don’t already have a very large data science team, I mean, you’re not ready. That would be my message: you will know that you’re ready when you’re hiring your 24th data scientist, and you don’t exactly know what sort of new things to explore. Then it would be a good time to start having a look at that.

Kieran Chandler: Okay, let’s wrap it up there for today. Thank you, Joannes. Okay, that’s everything for Lokad TV this week. We’ll be back again next time with another episode, providing we survive this heatwave. We’ll see you again next time. Bye for now.