The Estée Lauder Companies Inc. (EL) Earnings Call Transcript & Summary

Right? Do you stick to the version of the truth that it got to be and then take a hit on the forecast accuracy? Or do you just follow the curve and then say, you know what, I expect inventory to be pushed towards the end of the quarter. That's the behavior I might as well forecast it, right? So there are several such trade-offs that makes the forecasting initiative like this, be it for supply chain, be it for merchandising, successful or not but very, very often, we talk a lot about the technology and the feature function of the technology. And that's why I think there are so many projects out there even with good technology, doesn't end to be a success because people often overlook the complexities involved in these decisions. So tell us a bit about how do you measure whether it is a good forecast or a bad forecast? How do you measure a plan and adoption? If the plan is willing to take the numbers? Or do they have better numbers?

Yes. I wanted to come back on a point that you just made is I do believe that this sell-through-based strategy can only be delivered through technology, right? Because finally, our company needs a sell-in forecast to set inventory targets, right, and guide the markets on where our revenue expectation is, right? But we believe that sell-through is the purest form of consumer demand, right? So making this translation from sell-through to sell-in can only be done through technology. Manually it's almost impossible, right? So we had to implement this technology to make this shift, first of all. Now second, this transformation, together with our -- we do this together with our local partners and our local sales organizations, I would say, that do the execution. And it's a big mindset shift to forecast sell-through versus sell-in, right? We've always, as a company, forecasted sell-in. And now we're moving to a sell-through signal. So there's a big change that we need to manage there. Then to your question, Siva, how do we measure if we are adopting stat to the right level, we've had to implement numerous but also simple to understand metrics so that planners can actually know what to adopt out of this stat and where do they need to enrich it themselves to achieve the best possible result for the company. And I think that's really the journey where we're on today as we're growing the accuracy of the stat together, right, my data science team together with Siva's team. We're also working on where do we enrich that AI forecast, right, so that we can elevate the forecast accuracy for the company as a whole.

That's an important point. And I think it's a good opportunity to also talk to the audience a bit about the technology, how it works and then how it integrates with it. So you have a planning system that you use, right, which is the consumer of this forecast, right, and then which does subsequent inventory and other things. And the forecast itself is generated by Zebra's demand intelligence software and then passed on to the planning system. And here, there is an important question of forecast value add by the planners, and we get this question often, right? Do we want our planners to go through the millions of combinations that are sitting there and then rewet the forecast that's generated by an AI model. It looks counterintuitive for someone to do it until the AI model comes with a stupid answer, and then you end up making a wrong decision, right? Then it becomes incredibly hard to say, do you look at all million combinations or 50 million combinations in your case versus some selective part of it, right? So how did you go about deciding which part of it is no touch for the planners and which part of it, there is an opportunity to enrich to come to a better forecast?

Yes. So one reason that we implemented this technology is because we had to save time for the planners, right? So our initial approach was just to adopt the stat for all our C&D items, our low runners. They're adding the bottom part of our revenue, right? The stats in the human forecast are or manual forecast, I would say, the accuracy is more or less the same, and they're by far the largest part of our portfolio. It's the entire TIL. So why spend effort manually forecasting these if you have a machine that can do it for you with the same result. So I think that was the first step that we made in our adoption journey. The second step that we made was we started to measure obviously, the difference between the automated AI forecast and the manual forecast. And wherever we saw that we were adding cycle-on-cycle and negative enrichment, right, or we were making this stat worse, we decided or we implemented a process that planners could simply just adopt the stat forecast with one click. And we would adopt the stat forecast in our final forecast and with that, eliminating the negative value add, as you called it, in our forecasting process.

Great. So we would like to keep a few minutes for the questions, at least a couple of questions. So walk us through the results that you have seen through the year and tell us the significance of the numbers that we are seeing because a number can be misleading, right? You will -- when you walk down the aisle here, you will see people saying 30% better forecast, so on and so forth. Walk us through this a bit.

Yes. So over the last 2 years, we've seen our demand forecasting accuracy. It's a mixed accuracy, we call it WFA. It has grown with 9%, and we measure this at a LAG-3, which is, I think, an industry standard. So the mixed accuracy at LAG-3 has grown with 9%. A lot of this growth came from our Western markets, North America, Europe, Middle East, Africa. These are also typically the markets where we are able to transition to a sell-through-based forecast, whereas, as I mentioned in the beginning, we're still in our road map ahead, we're heavily focused on predicting, let's say, the rebound of the Chinese traveling consumer. And I think a couple of other things that I'd like to mention here is that new product forecasting is always a challenge, especially when you try to forecast items that are completely widespread, right? So you don't have a predecessor that you can take the trends from and it's completely widespread. So besides forecasting for emerging markets like China, we're going to put a heavy focus on new product introductions, right, and get a right forecast there. Our company brings about or turns over about 30% of its portfolio every year, right? So new products are a big part of our mix. And as we just mentioned, we're going to further drive our forecast evaluation, make sure that we only enrich the stat forecast there where it adds value and make sure that we stop detracting value from the stat or from the AI forecast. And we already see some very promising results there. For instance, in North America, that was our first global market that went live with this planning solution. We achieved 66% mixed forecast accuracy just last month, right? So at like 3 with a huge SKU complexity, that's a massive, massive result. Our North Star is achieving a 70% forecast accuracy for the entire company. So North America is tracking well to that goal.

I want to say this in context. 66%, why are you celebrating could be for that industry with the sheer complexity where you have 50 percentage variability month-on-month on the demand, right? If the demand is 50% next week, it's -- next month, it's 25%, and then it rebounds to 37.5%. So there is literally no method to the madness and 66% is incredibly powerful while the North Star is 70. Thank you very much, Frank, for this. We have 4 minutes. Maybe we can take 1 or 2 questions. There is someone who is walking around with a mic or if you're loud enough, please put your hand up and then ask the questions, please.

My name is Adriana Kuri. I have a question because working in the beauty industry for a long time, and I can see the complexity for managing so many different brands with different innovation patterns and every brand manages their innovation differently. How do you plan or how do you start the implementation process for AI because I know you would use that on predictive analytics, but then the next step is to go on a brand by brand or category by category basis. And then perhaps it comes to you globally. How do you manage the complexity and how would you start working on that to get to that global umbrella?

Yes. Maybe I'll...

Can you repeat the question, please.

No, we will repeat the question. The question is how do you go about starting with the AI? What is the approach you will take? Do you take a brand-by-brand approach? Or do you take a big bang approach to rolling out AI in an organization given how complex the beauty industry is?

Yes. What I would say is start with the beginning and not with the end. And what I mean with that is we started with the end initially. So we wanted to go to sell-through forecasting from day 1, right, because we knew that was our endgame. For the organization that was way too difficult to make such a big step in one go, we're talking about hundreds of planners, right around the world. So we had to really scale it back and start with a forecasting solution that was as close as possible to what we were doing manual at that time. So we really started to forecast shipments, right, that we were making to our vendors. And only from there, we really had a pro walk/run approach that finally brings us to a sell-through-based forecasting for the most mature markets, I would say, start with that. Then the second piece is, I would really encourage region by region, I would bring in region, all regions bring them in, look at a -- we see great trends at a channel level and at a major category level, right, and govern this at a -- through a global organization, right? So we built all our processes globally, which we then implemented locally. And the strength of a global forecasting process is that you really drive results much faster than when you would implement this at a local level, region by region by region.

I think the important point there for me is, Frank, particularly when a technology is as complex and transformative as this do not bring about too many changes at the same time. Do not look at this as one big project, let's get to the end state, and we consulate a whole lot of problems. So we had the technology work for where we are and then went about the process transformation, which is a much bigger leap. Otherwise, people are going to mix the technology issues with the process issues and we're going to get more a lot of pushbacks. Any other questions? We have one more question here.

So my question for you is, I know that it has typically been a challenge to work with retailers on getting their point -- it's been a challenge to work with retailers on getting their point-of-sale data for does it sell-through forecasting? How have you gotten their engagement? And what are some of the challenges that you faced?

Yes. So good question. So if I look at our global markets, right, I would say that, in North America, we have a couple of really big retailers that we work with, and they were ready to offer the POS, point-of-sale data, and we actually had this already ongoing in our organization. In China, the business is quickly transitioning to a more online model. And there also, they're very willing to share their point-of-sale data. In Europe, where the retailer landscape is much more disintegrated and spread, I would say, is much more challenging. What I do often find is that they're willing to provide this information as part of a negotiation, right, either for money or for something else. And what you can also think about is service levels, right, that are improving priorities and allocations, those types of things.

Excellent. Thank you very much, Frank. Really appreciate your attendance today. Thanks for being a great audience, appreciate your time. And we are going to be here for a few minutes. If you have any questions, if you want to grab either of us, we'll be happy to talk to you. Thank you.