00:00:04 Intro: Forecasting challenges for high-growth firms.
00:01:16 Company growth types and related forecasting issues.
00:02:18 Difficulty of predicting new product success.
00:04:07 “Hit or miss” scenario and responses.
00:05:43 Other forecasting problems and stepwise growth concept.
00:08:01 Predicting business growth timing challenges.
00:12:18 Strategies to mitigate forecasting unpredictability.
00:13:46 Supplier negotiations role in ‘hit or miss’ situations.
00:14:13 Identifying indicators of upcoming growth.
00:15:16 Google search stats: a forecasting tool?
00:16:23 Strategies for rapidly growing businesses.
00:16:46 IT changes implications during growth.
00:18:14 Businesses’ tendency to over-forecast.
00:19:03 Quantitative supply chain and swift growth reactions.
00:21:11 Closing thoughts.
The interview between Kieran Chandler and Joannes Vermorel focused on supply chain optimization challenges for high-growth companies. Vermorel stressed the complexities of forecasting in such scenarios, identifying two types of growth: organic and “hit-or-miss” from new products. The latter, he argued, complicates forecasting due to its unpredictability. Recognizing the fine structure of growth is crucial to differentiate between these scenarios, informing better inventory decisions. Vermorel suggested solutions like quantitative risk modelling, end-to-end automated decision mechanisms, and flexible supply chain strategies adapted to the growth structure. He also advised anticipating constant changes in the company’s IT landscape to cope with growth-related complexities.
In the interview, Kieran Chandler, the host, discussed the challenges of supply chain optimization for high-growth companies with Joannes Vermorel, founder of Lokad, a software company specializing in this field. The conversation focused on the dilemmas that these companies often face in anticipating and responding to demand, and the types of growth that can complicate forecasting efforts.
Chandler introduced the problem of forecasting for high-growth scenarios, citing the example of companies buying excessive stock to meet anticipated demand spikes, only to find themselves stuck with surplus when the growth doesn’t sustain, a situation that resulted in a loss of 4.3 billion dollars for H&M earlier that year.
Vermorel explained that companies facing high growth often encounter a series of difficulties, stemming from both internal disorganization due to rapid growth, and the statistical complexities tied to the nature of the growth. The founder emphasized that the source of a company’s growth significantly impacts the forecasting process, making it more or less complicated.
Vermorel described two types of growth that businesses might experience. The first, organic growth, happens when a company grows uniformly, such as in the case of an e-commerce business that sees a steady increase in website traffic, resulting in a linear growth of product sales. This scenario, while relatively simpler to forecast, still contains uncertainties about the sustainability of the growth, exemplified by the H&M situation.
The second type of growth arises when a company introduces new products to the market. This scenario becomes more challenging for forecasting as not all new products will contribute to growth; only some might be successful, creating a “hit-or-miss” situation. The growth, in this case, is driven by these hit products, while the rest might not show significant sales increases. Vermorel mentioned fashion companies as an example, where occasional products become major hits, contributing to overall growth, but the majority of products maintain flat demand forecasts.
This hit-or-miss scenario presents a particular forecasting challenge. Companies cannot respond to it by simply increasing the stock for all products, as only a fraction will likely be successful. An undifferentiated answer, as Vermorel put it, would lead to wasted resources and puzzling biases in the forecast. He elaborated that while a business might be growing, the forecast at a disaggregated level might not reflect this growth, primarily because it’s impossible to predict which product will be the next big hit. As a result, the average forecast can be much lower than the actual outcome.
Vermorel begins by acknowledging the inherent unpredictability in forecasting for businesses, especially when it comes to launching new products. Not only is it uncertain which products will be successful, but it’s also unclear how many products a company might launch in the future. This uncertainty extends to products that haven’t yet been fully developed or considered for launch, making it challenging to forecast their demand accurately.
Despite this understanding, Vermorel notes that predicting the exact moment when growth will occur remains elusive. Companies may anticipate growth spurts but struggle to pinpoint their timing. This is particularly true when expanding to a new country, which introduces a new set of variables not suitable for traditional statistical forecasting.
Addressing the solution to these challenges, Vermorel suggests that although it’s difficult to improve average outcome forecasts due to the “hit or miss” nature of product success, companies can gain insights by understanding the statistical structure of their growth. This understanding helps differentiate whether growth is driven by individual product success or company-wide factors. In turn, this allows for better inventory decisions, such as adopting conservative inventory policies early on and being reactive when a product appears to be a hit.
Vermorel asserts that the key strategy in managing inventory risk versus expected reward is to capture the hit when it happens. He warns, however, against aggressive inventory responses to fast-growing products in businesses that usually see organic growth over extended periods. These situations may not represent a “hit” but might be a temporary spike, making aggressive inventory decisions potentially detrimental.
In instances where forecasting is blurred, Vermorel recommends adjusting for variables like negotiating better minimum order quantities (MOQs) with suppliers. This strategy allows for purchasing smaller quantities initially and responding aggressively if a product proves to be a hit, offering a practical way to accommodate the inherent uncertainties of the “hit or miss” scenario.
Vermorel emphasized the inherent uncertainty in growth prediction. While there may be indicators, he argued that there isn’t a foolproof way to anticipate growth definitively. Any such method would be exploited by companies to generate limitless growth. Therefore, it’s vital to acknowledge this irreducible uncertainty. Using Google search statistics as an example, he explained that such data might provide short-term insights, but it falls short in helping businesses anticipate long-term trends, which are critical for supply chain planning.
Given these uncertainties, when a company finds itself in a period of exponential growth, Vermorel suggested a few strategies to be better prepared. He emphasized the importance of anticipating that the company’s IT landscape would be in constant flux. As a company grows, typically every two to three years, it may need to overhaul its entire IT infrastructure. This continual change can cause complications in accessing and reconciling data from old and new systems, thus further complicating growth forecasting.
Vermorel also discussed the tendency of companies to over-forecast their growth due to emotional attachment. This bias towards optimism can lead to inadequate risk management. He proposed a more quantitative approach to risk modelling, arguing that growth represents a specific type of risk that requires a swift and efficient response. For such responses to be effective, he advocated for end-to-end automated decision mechanisms rather than relying on human input, which could introduce delays and errors.
Lastly, Vermorel recommended that businesses consider the fine structure of their growth. Whether growth is steady and organic, hit or miss, or due to new markets, the implications for the supply chain and inventory management differ significantly. He emphasized the need to adapt supply chain strategies to the specific growth structure the company is experiencing. This approach can help businesses manage their supply chains more effectively during periods of growth.
Kieran Chandler: Joannes, in terms of forecasting high-growth scenarios, it seems to be a fairly difficult task. What sort of problems do companies typically face in these situations?
Joannes Vermorel: There are actually a series of problems. Firstly, when companies are growing internally, there’s often a large amount of disorganization that’s just a result of the growth itself. This complicates everything, including forecasting, which is made more challenging by the growth. But on a statistical basis, it really depends on what is generating the growth. There are many different ways of growing that can make the forecast more or less complicated to accommodate for the growth.
Kieran Chandler: What sort of growth are we talking about here?
Joannes Vermorel: Probably the easiest type of growth is organic growth. For instance, if you’re an e-commerce business, every month your website gets a bit more traffic and all your products are seeing an upward trend. This growth is linearly correlated with the increase in your website traffic. That’s the easy situation because it’s uniform. The challenge arises when trying to anticipate if this growth will continue forever. If you anticipate a significant company-wide growth that doesn’t materialize, like the case of H&M, it can hurt pretty badly. But still, from a pure forecasting perspective, this is a simpler situation.
A different way of growing is when you launch new products. That means you have items that may or may not be growing, and then you get better at introducing new products to the market. These products may be a hit or miss. Your growth is driven by new products, but not all of them. This is typically when companies get better at optimizing their assortment and identifying products that are going to be most attractive to the market. This can happen, for example, with fashion companies that get better at launching products that occasionally become massive hits. This scenario is much more difficult in terms of statistical forecasting.
Kieran Chandler: In this hit or miss scenario, where you have one product out of twenty that’s really successful, how should a company respond to that? Is it viable to put extra stock in all of the items in your catalog?
Joannes Vermorel: Exactly, that strategy doesn’t work. It can’t be an undifferentiated answer. Also, it means that you will end up with a lot of puzzling biases in your forecast. For instance, your business is growing, but if you look at the forecast at a very disaggregated level, it doesn’t appear to be growing. This is typically for two reasons. First, you don’t know which product is going to explode, so the average is much lower than what you will actually get.
Kieran Chandler: Sometimes, a product explodes unexpectedly, but we also don’t know exactly how many products we’re going to launch in the future. These unanticipated products, which aren’t yet in our pipeline and ready to be forecasted, are absent from our forecasts. We basically forecast the demand for existing products or the ones we’ve already planned to launch, but the products still under consideration aren’t part of the statistical forecast. Other than not knowing which new products are going to really grow, are there any other problems that we face when forecasting growth?
Joannes Vermorel: Yes, there is also the fact that the growth of a business is not always steady. It can be a stepwise function where growth is slow, then suddenly there’s a large increase, and then no growth for a while. This is common in many businesses, both online and traditional brick-and-mortar.
Kieran Chandler: Why do these stepwise increases in growth occur?
Joannes Vermorel: In e-commerce, you can see dramatic impacts from search engine result pages. If Google ranks you higher or lower, it can have a substantial effect on your business. It can double your business size or cut it in half overnight. There have also been instances where the first player in a market to do national TV ads has seen significant growth. For example, people weren’t used to buying car parts online until a company decided to launch a massive TV campaign. Even companies that weren’t paying for TV ads benefited from this market uplift because people started comparing prices online.
Kieran Chandler: So you’re saying the key problem is a company might know they’re going to grow, they’re going to make that extra step at some point, but the real problem is they don’t know when that’s going to happen?
Joannes Vermorel: Yes, they know there’s a chance it will happen, but they might not know exactly when. For example, they might have a good guess if they’re expanding into a new country. When they start to open this new country, they can expect a step in growth, but it still isn’t eligible for a statistical forecast because there isn’t a statistically significant set of events representing the opening of new countries.
Kieran Chandler: It sounds like it’s quite messy. You only know that maybe one out of 20 of your products is going to succeed, but you don’t know which one or when that’s going to happen. So what’s the solution here? Can we actually forecast these scenarios?
Joannes Vermorel: It’s very hard to get a better forecast on the average outcome. The average outcome is going to be extremely fuzzy. When you have a hit or miss pattern, it means you don’t know which products will be the hits. If you did, you’d probably be incredibly rich.
Kieran Chandler: You suggest that we could statistically understand the growth pattern of a business. This understanding could differentiate between a business where the growth is driven by hit-or-miss at the product level, or by a step-by-step growth at the company-wide level, or even by another kind of growth. Can you elaborate on this?
Joannes Vermorel: Absolutely. There are different types of growth. Some products gradually grow over time to become very large, and it’s not all products but only certain ones. By capturing the nature, the statistical nature of the risk, you can take better inventory decisions. For instance, if your growth is driven by a hit or miss pattern, you’d want to have conservative inventory policies at the outset. However, you’d also want these policies to be incredibly reactive and aggressive when you start to detect that a product appears to be a hit. This strategy balances inventory risk versus expected reward, which is crucial in capturing a hit because such occurrences are frequent in your business.
Kieran Chandler: But, what if a business doesn’t exhibit these hit or miss patterns, where products grow organically over an extended period?
Joannes Vermorel: In a business where products grow organically over time, a product that starts to grow very fast is probably not a hit. It’s probably just a fluke. Therefore, it would be a poor inventory decision to aggressively respond to this growth.
Kieran Chandler: So, we can produce some sort of forecast, but it’s going to be quite blurred. How do we mitigate for that blurriness, especially in a hit-or-miss kind of scenario?
Joannes Vermorel: You’re right, it’s challenging. There are a few businesses where you can more predictably predict the future. For example, in aerospace, if you’re servicing parts for aircraft, the growth you’ll see in your supply chain is typically related to the number of aircraft you need to serve. You typically know months in advance if you will have to serve more aircraft. However, most businesses do not have the luxury of such an indicator. One strategy could be to negotiate better minimum order quantities (MOQs) with your suppliers. This could mean buying smaller quantities more often, allowing you to respond aggressively if you detect a hit, instead of purchasing sizable amounts of inventory and ending up with overstocks on all your misses.
Kieran Chandler: Earlier we talked about those steps and making the jump. What sort of clues should a company look for to identify they’re about to enter a growth period?
Joannes Vermorel: The problem is there is no guaranteed way to predict growth. If there was a method to leverage statistics to accurately predict growth, at Lokad we wouldn’t be doing supply chain optimization, we’d just be playing the stock market. It’s fundamentally irreducible uncertainty. If you had a way to be absolutely sure that you were going to grow, some company would be exploiting this trick to grow indefinitely. But as we’ve seen, even the best companies can only grow for a long period until they become exceedingly large.
Kieran Chandler: Are there limits to the market, and is it possible that growth can just stop? For example, can we include things like Google search statistics into forecasts for some insight?
Joannes Vermorel: Yes and no. The issue is that, let’s say you are selling a product, like an over-the-counter remedy for a common cold. You need to source weeks, if not months, in advance. Google search might give you a few hours’ head start when people start to look up ‘I have a cold, what can I take to sleep better?’ However, they will only start to search that at the last moment. So, while you may be able to slightly improve your forecast for the next day with a few hours’ notice, in supply chain management, you typically need to look at least a few weeks ahead. Therefore, even real-time search results don’t help you anticipate events that are weeks away.
Kieran Chandler: If I’m seeing my company grow exponentially, what can I do to be best prepared? You mentioned improving relationships with suppliers and possibly changing minimum order quantities. What else can I do to be as prepared as possible?
Joannes Vermorel: One counterintuitive measure is to anticipate that your IT landscape will be in flux for a long period. This complicates things because accessing data becomes even more difficult. Companies that grow significantly every two or three years often need to change everything in their IT landscape. The existing ERP, website, WMS, etc., may not be good enough anymore. This constant change means you risk not being able to reconcile your historical data with the data from the new system. You may end up with a new ERP or WMS and then face the challenge of reconciling data from the old system with data from the new system. This can cause numerous inconsistencies and make forecasting under growth more complex due to these IT disruptions.
Kieran Chandler: In the real world, would you say that people often overestimate how much they’re going to grow because of an emotional attachment to their companies?
Joannes Vermorel: Yes, there is a tendency to be overly optimistic. The idea should be to hope for the best but prepare for the worst.
Kieran Chandler: For the worst and typically, at least frequently, people do the exact opposite. But fundamentally, what happens is that very few companies try to model their risk, I would say, quantitatively.
Joannes Vermorel: That’s what we try to do at Lokad with this idea of quantitative supply chain. Growth is a certain class of risk. It means that there is a certain class of uncertainty where your business, or some products, can suddenly sell a lot more and you need to have processes in place where you can react very swiftly to this sort of situation. Swift action typically means not having people in the loop because if you expect that coping with the growth depends on supply chain managers that need to look at the data on a daily basis, make their mind up about the fact that what they’re seeing is not noise but really the result that this product has entered a new stage of sales volume. Compared to the previous stage, it’s now selling 20% more or whatever.
People will want to wait a bit to be confident in what they see. You can easily add a couple of weeks of lag just because you have humans in the loop to implement decisions. So, probably one way of coping with growth is not actually to be that much better in anticipating the growth. It’s just that when the growth is statistically noticeable, to act on it much faster. This means to have an end-to-end automated decision mechanism that basically triggers reorders faster, as opposed to having someone that is going to wait until they’re unfortunately very confident in what they’re seeing.
But that also means typically that you’re very confident in facing stock-outs because if you wait until you’re very confident that the product is a hit, you’re nearly guaranteed to have waited so long that you have near guaranteed a stock out because you’ve not placed a reorder early enough to accommodate for this emergence.
Kieran Chandler: So, to wrap things up, the core message to take away from today is basically if you’re a business and you’re experiencing huge amounts of growth, it’s better to be more reflexive and responsive rather than to store up stock and act with the future in mind. Would you agree with that?
Joannes Vermorel: Yes, and it’s important to really think about the fine structure of your growth and to differentiate that so that you can deal with the growth very differently, whether it’s business-wise organic steady growth, hit or miss, or any kind of other alternative growth structure such as opening countries. Because the consequences for your supply chain and for the way you size your inventory are very different.
Kieran Chandler: Okay, great. We’re going to have to wrap things up there, but thanks for your time. So that’s everything for this week. We’ll be back next week with another episode, but until then, thanks for watching.