WEBINAR: Master Demand Planning with AI: Turn Retail Chaos into a Clear Vision

Hello everyone, welcome. We are going to wait just a couple minutes for folks to um join and then we'll get started. All right, let's go ahead and get started. Thank you so much for everyone uh for joining today. Very excited to talk through the principles of uh demand planning and forecasting in the retail space. Uh my name is Victor Hart. I'm senior account executive here with Leafio um based in the US and uh on the call with me today is uh Helen who is our product director for the demand planning and inventory optimization solution. And so what we'll be walking through today is a little bit of background about Leafio and our company uh some of the clients we work with. Then I'll be handing things over to Helen to talk through the really specific uh details of our demand planning process that um she knows much much better than I do. Uh so just a

little background um as we go through the process here. Um, so again, Helen, COO and the inventory optimization expert here with Leafio. Again, my name is Victor Hart. I'm senior account executive. Um, a little background about Leafio. So, we've been in business for over 15 years and uh, we have implemented in hundreds of uh, retail clients all over the globe. everything from uh clients with 5 to 10 locations up to several thousand locations. And so we've got a full suite that we'll talk about here in just a second. Uh but the uh software is very very uh scalable depending on what the client's actual needs are. And here are just a list of our great clients that we've worked with over the years. And

what you'll see is that we've got a uh a lot of experience in very different verticals within the retail space. So we have groceries, supermarket clients, um liquor stores, hardware stores, pet stores, everything in between. And so regardless of the vertical that you work in, um we our team has experience in that. and we can help leverage the experience we have with these clients to help you optimize your uh processes and your solutions. Some of our recent wins uh these are again spread all over the world uh but Spar is one of the largest uh grocery chains in Eastern Europe. Ki 4 you might be familiar with um in Western Europe. is a grocery store based in Southeast Asia. Lint is one of the largest um

chocoliers around the world. Philip Morris um etc. So we have a lot of experience in all of these different types of clients and would love to work with you to apply our learnings from there and help you improve your processes. Um what we offer is a single unified platform for end toend merchandising and supply chain management. Um so just to give you an idea of the different modules within the solution. If we start in the upper right hand corner here, we have our assortment uh performance module which is all about driving our assortment strategy based on the sales that are actually coming in for each individual solution. Uh so based on what's selling where uh what products should we be offering and then that ties into our shelf efficiency solution which is about maximizing the use of space

within each one of our retail locations. So we know what's selling where. Then we need to be able to uh assign the optimal amount of facings for each one of those products. That then ties into inventory optimization, which is what we're going to be talking about today. So, we know what we're selling where, we know how many facings. We then need to know or need to make sure that we have the product available to place it on the shelf um accordingly. So inventory optimization is going to take care of our demand forecasting, our replenishment, managing the optimal amount of inventory within each node in our supply chain. Be that at a the actual retail location, be that at the uh central distribution center, central warehouse, however your supply chain is structured, we can um help accommodate

that. uh we can take into account promotions. So learning how um based on past promotions uh what's the price elasticity of products things like that so that we can help drive the accuracy and get the forecast more accurate for future promotions. Um there's a customer loyalty program as well and a uh application on your mobile phone so that you can for example uh when planagrams get updated um from the central merchandising team uh get tasks sent down to the store level. So the whole idea here is to as effectively and efficiently as we can manage the entire uh retail

process. So with that I think I'll hand things over to Helen. Yeah I will share the screen if you don't mind. Basically the same one. Do you see my screen now with the webinar agenda? Yes. Great. So regarding the agenda of our today's conversation, first of all um we are going to talk just a little bit about the departments that are utilizing the forecasts uh and from the perspective of the sales and operation planning uh how the forecast is utilized uh in terms of different uh departments and what is uh influencing uh what factors are influencing the accuracy in each specific case. Um, of course, we are going to uh talk about just a couple of principles of the SNLP, but in general and a small spoiler, we are going to have a separate webinar um

on the topic of the sales and operation planning in details. We're going to talk about the methodological points uh in terms of uh building the processes in terms of enhancing and increasing the efficiency of the SNOP processes in retail. So uh keep track of uh our uh agenda of the upcoming webinars not to miss it. Uh, of course we are going to talk deeply about the machine learning models uh uh that can be utilized for the um for covering uh the tasks of the demand forecasting uh and not only the demand forecasting just to talk about how this um tasks and problems and issues can be covered with the help of the machine learning uh technologies. Um, of course the factors that are influencing the accuracy of the machine learning models uh because really the machine learning is really about the

data and the accuracy of the data and based on our experience uh uh success is really depending on heavily depending on the uh data and the accuracy of the data. the smart ways to measure the forecasting success. Uh, of course we will talk about the measurement of the metrics for the forecasting accuracy, the forecasting mistakes. uh but furthermore we will talk about the financial metrics uh the outcomes um that uh are influencing the financial results of each and every retail company that is depend that are dependent on the forecasting accuracy. So let's start um if talking about the departments uh that are uh utilizing the forecasts and uh uh what kind of forecast they rely on. We have uh all the retail company

basically that are dependent on the forecast in uh different ways and manners but uh the most dependent departments are the commercial departments. Of course they are uh operating with the sales forecasts uh mostly with which are compared with the sales plans and that is how the SNOP starts basically uh the operational departments here we're talking about more the demand forecasting uh basically the forecasting quantities the purchasing department heavily dependent on the demand forecasts and sometimes uh uh They uh rely on the sales forecasts for them to understand uh and plan their promotional activities properly together with the commercial departments. uh logistics and transportation uh really dependent on the demand forecast to understand the capacity and here we're

talking about both the capacity of the central warehouse and the capacity of the transportation and marketing uh and development department. So they really rely on the mostly uh sales forecasts as they are taking part uh in um the consideration let's say of the promotional activities if talking about uh the marketing department and about the overall planned revenue in terms of the development department as they're influencing that uh of course so um as I have mentioned we are going to talk about the um principles and methodology of the SNOP but uh just to clarify uh one of the core principles of the SNOP is the unified forecasts that are uh cross-cheed through all the departments and when we're using the same approach when we're using the same

uh understanding of how the forecasts are built which factors are considered uh when the demand forecast and sales forecasts are calculated on the same data. This can help the retail business to have the unified forecasting approach uh to to the like the general understanding of the future and the plans that are going to [Music] um going or not going to happen. let's say um in terms of the practical application of uh machine learning and in general artificial intelligence um when going back to the leaf platform in general um we are covering currently four and even five it it will be a small spoiler as well from my side. So of course uh we are utilizing the machine learning in terms of the demand forecasting I was talking about

previously. So when we are talking about the baseline uh forecast uh the um regular demand forecasting uh short mid and long-term forecast of course the promotional demand forecasting and uh it was the first step for liio to utilize the machine learning as we understood and saw uh much better results in terms of the forecasting accuracy for the promotional demand forecasting And currently we are utilizing the demand forecasting for the outer facing uh calculation uh that uh is based uh basically on the future demand and uh uh when the shelf efficiency solution uh is calculating the planagrams for uh our customers the automated uh facing can be based on the uh future vision let's say rather uh than on the past u sales data

uh identifying the promotional analogs. So this is the task that is more connected with the promotional activities as well but um it is more about like analysis and understanding how accurate the promotional forecasts are. uh and we have a separate block of functionality that is covering uh this analysis uh and comparison between the um actual SKUs that are participating in the future promotional activities and the SKUs that were uh participating in the past promotional activities with different factors and the combination of them. uh clustering of stores uh to optimize the assortment performance. Basically uh yes this is one uh of the features of the assortment performance solutions uh solution that uh is based on also on machine learning technologies. uh and uh this is the

smart approach of clustering of stores in terms of each and every category we're introducing uh the solution to uh and it helps to automate the clustering and understand to which specific clusters do we need to introduce this or that new item uh based on the specifics of the cluster and product clustering for sales forecasting uh in terms of new items Definitely this is a very important feature of the inventory optimization solution that helps to um calculate the initial forecast for the new introduced SQU in case if they don't have any kind of similarities uh between the existing and new introduced SQU. Um when we are talking about the machine learning models uh there is a huge hype among the utilizing of um AI and uh

machine learning in particular uh in all the spheres of business and especially in retail. Uh that is why and we uh as a company uh as practitioners in uh utilizing of the machine learning technologies for more than five years definitely understand the difference between uh the uh traditional or statistical methods of forecasting and machine learning methods. Uh let's start from the uh cost of the implementation uh as it is really influencing the results. So um from the standpoint of the cost of course we understand that machine learning technologies are a little bit more expensive than the traditional methods. uh and uh but on another hand we understand that uh the accuracy of the forecast the adopting to changes in the data uh some uh especially which is the most important for me is the complex interdependencies between different

types of data and different categories that are taken into consideration by uh machine learning the scalability automation flexibility and uh explainability of results uh there can be some issues or questions about the explanability of results because we uh are used to have like a straightforward formula uh in statistical methods where we clearly understand everything. We see the dependencies, we see and understand the expected results and it is very visual. But in terms of uh machine learning we um have special like another type of approach for visualization of the results and uh explanation of these results. So that's why here uh everything is covered as well. Um at the same time we understand uh that the what is the measurement what is the difference in terms of the forecasting accuracy but just like to understand

that it is higher. Okay. But uh how uh how um is it really crucial percentage uh of the forecasting accuracy and due to our results we see that uh usually of course it depends on the data the W map uh is um 7% and great uh and greater um in terms of the statistics model that means that the forecasting accuracy is at least 7% higher higher than the traditional or statistical meth methods or models for forecasting. And it is important to highlight that our approach uh is um in terms of balancing between the cost uh and efficiency. uh and we are trying to implement the machine learning uh and AI technologies where they are uh supporting and giving greater results if

to compare with the used or traditional methods. Uh that's why we are like utilizing these technologies in the spheres I was highlighting before. uh when we are talking about the models uh that can be used uh especially and in particular for demand forecasting now as the main topic of our today's webinar was the demand forecasting and utilizing machine learning technologies in specifically demand forecasting tasks. Uh, of course that can be there can be like a range of different models starting from simple moving average which of course obviously is not machine learning technology. Uh, some kind of um, ARMA uh that are including the regressors that are not including regressors. uh some newest models like NBIT um some um well-known models like the Facebook uh phonet uh and uh the decision tree models like

random forest boost and light GBM. So there is like a huge variety of course um there there is much bigger quantity of different models that can be used for the demand forecasting purposes but there is just like some examples of models that uh uh can be used uh for this task. And when it comes to machine learning, here is the range of examples of this specific machine learning uh models that are uh used by our company and uh are utilized uh uh for covering of this task with the high level of uh accuracy and uh uh just now on the biggest quantity of projects we're using like GBM uh due to our own research and due to um even international competitions in

terms of the forecasting and uh demand forecasting in particular the light GBM model is given the greatest results. So basically it is fast and efficient. Um when it comes to machine learning, we are talking about uh big batches of data uh big batches of historical data and uh uh not only the accuracy is important here but accuracy as well but at the same time it is extremely important for retail business to have uh fast results. That's why the speed of the usage of uh the model is uh really crucial uh if we are talking about the task of the demand forecasting and at the same time due to the results of the Kaggle competition. Um the light GBM model is recognized as one of the leading model for forecasting tasks. Um when it comes to the factors that are

considered and must be considered uh for the purposes of the demand forecasting when it when we are talking about just the regular demand forecasting for different range of times of course we uh when we're talking about shortterm demand forecasting the number of factors can be different. So we have additional factors when we are talking about the long-term forecast there be there will be some additional factors but in general we are taking into consideration the demand factors the calendar factors uh the most important thing is the price factors or promotional factors uh as we are utilizing the price elasticity because it really changes a lot in terms of the demand forecasting even if we're not talking about the promotional activities. Just the fluctuation of price can be crucial for the demand forecasting. Uh and also taking into consideration the canibilization of sales that is also uh connected with the price elasticity and the SKU

characteristics um is also considered when we are talking about the demand forecasting for new introduced SKUs. when it comes to the promotional demand forecasting uh of course additionally we need to consider some other uh important factors that are really influencing the promotional demand. Uh of course this is price and the price elasticity here is really crucial. Uh also the system considers the type of promotional campaign, the type of advertising, the type of the promotional mechanics, uh the additional layouts, the quantity of locations that are taken into consideration in terms of the promotional campaign uh and some other relevant and important factors. Um of course uh when we are uh considering the demand forecasting like regular demand forecasting and promotional demand

forecasting uh it is worth saying that one more important thing uh need to be covered here and it is seasonality and if talking about the regular seasonality um um like different periods of time when we have different demands due to the demand patterns um that are influenced by weather for example or by some holidays some other external factors. It is one thing and uh usually it is covered but another thing is high seasonal categories and uh it is uh sometimes very hard to identify these items. uh I mean uh some high spikes of demand during Easter, during Ramadan for example. Uh and it really depends on the model and how uh these items first of all need to be identified uh for the future forecast and secondly need to be forecasted

um based on uh uh on the special methods of forecasting and special factors that are utilized here. uh and also of course uh there we have a lot of new introduced squs and this is a factor that is taken into consideration in this case as well. Um it is the most common question uh on our meetings during the demonstrations of inventory optimization solution. What is your forecasting accuracy? Uh what is the uplift of the forecasting accuracy? uh we can expect after uh the project will be implemented. To uh directly answer the question uh we need to understand different um angles of uh the measurement. So uh of course the assessment horizon. So it really and the result of the forecasting accuracy will

really differ if we are comparing for example the daily level and monthly level. Of course on the monthly level we will have much higher forecasting accuracy if to compare it with the actual data. The aggregation level the higher aggregation level we use for the measuring of the forecasting accuracy the better results we gain. Uh and uh the matrix as we have a range and it is not a full range for calculation of the forecasting error. Usually we are talking about the bias, W mapper, MCMI, percentage error but of course the measurements uh the criterias how to calculate and the KPIs uh there is a huge list of it but um in general um what we are seeing and what are the benchmarks in terms of different industries not only in retail uh if to compare the weighted map. Uh we see the

range of starting from 10 to 25 size of weighted MAP um on the level of the company in general uh depending on the horizon but like in general it is weekly or monthly result. So it's better to compare um the actual results and the b with the benchmarks and to understand whether we have some uh um uh some gaps with forecasting and it can be improved with the help of the latest technologies. Uh as I have mentioned at the very beginning, the financial metrics are really important here as well. uh because we are um um we are considering the project for the implementation of LEA inventory optimization solution not just like a tool for forecasting or providing some figures for planners uh

but uh we rather consider it as a tool that helps to uh maintain um better results in terms of financial efficiency and we are financial responsible and interested in the improvements of results, we are ROI oriented company. That is why except of the forecasting accuracy and measuring the forecasting accuracy which obviously is an important part uh inside of the solution we are more convinced with the uh sales plan with fact. So whether we are meeting the plan or not uh the level of the overstocks and loss sales uh inventory turnover and the percentage of availability uh usually we are keeping track of these indicators and the influence of the forecast uh on these indicators but from our experience we see even if we have very high level of the forecasting accuracy let's assume that it will be

like something 95% of the forecasting accuracy on some level uh we still can have bad results in terms of u execution let's say and uh uh as uh we understand that as we have implemented more than 100 projects um we understand that we need to dive deeper into the processes uh not only to stop with the forecasting accuracy but also to understand how this forecast is utilized and what outcomes financial results uh we are g uh gaining uh with the usage of the solution uh and I think uh that's it from my side and I will pass my word to Victor. Thank you Helen. Give me one second and I will share my screen again. So um Helen uh talked through a bunch of

details about the forecasting process that we um leverage within the Leafio platform. Let's talk about some of the advantages of that. So first and foremost, we have industry-leading forecasting models and approaches just like Ellen uh talked through there a moment ago. Um so you're getting the best possible uh uh processes and logic that can be used. Um our technology uh can process large volumes of data. So like I mentioned at the start of the call, we have some clients that have uh thousands of locations. Let's say they're in a supermarket industry where they've got 40,000 SKUs as an example. um on average in each of their locations. And our solution can help um process all of that data very very conveniently. Um and it's consistently

being uh trained and adjusting based on new uh new data that comes in. So, um, our sales patterns are changing, our, uh, assortment ranges are changing, and the system is going to consistently, um, adjust to that based on the data that it's, uh, that it's being fed automatically. So, not requiring a user to adjust their strategy, as an example, as a merchandising manager or something like that. the system is going to pick up on that data automatically. Um, customuilt dashboards. So, one of the benefits of the solution is that um, for example, as a supply planner, I don't have to be doing manual uh, calculations in Excel or something like that. I can let the system take care of that for me. So I can then focus on more value ad uh tasks, more monitoring of the uh health

of our system overall. And then I can build my own dashboards based on the data that's important to me because what's important to me and my role might be different than what's important to Helen and her role, etc. And so we can leverage that. Um again we talked about the expertise of our consultants across various retail sectors. So whether um you are in grocery convenience uh liquor stores, hardware stores etc. We have a deep uh team to help um manage your processes based on the challenges that are unique to each one of those sectors. um which again we can accommodate those different types of clients and then a very quick integration with your existing systems. So typically our solution would integrate with um an ERP

system or point of sales system or what have you. Uh but based on our hundreds of implementations over the years, we've integrated with pretty much any system that's out there. And so it's very very easy and quick to get those implementations going. Um, one of the things that Helen mentioned is that we're very very focused on improved financial results. We don't want to implement software just for the uh purpose of implementing software. We want to help uh our clients achieve better performance in their business. Um and in the inventory optimization space, one of the things that we see is typically a really big reduction in our overall inventory levels with a increase in the uh availability of product. And so you can see here um we have a big reduction in

our loss sales because the product is more available while at the same time we're reducing the overall stock um which makes a big difference on the uh the uh balance sheet for each client. Um and this is achieved through better and uh more comprehensive data and the optimization and uh the the uh the the better uh features that are available as compared to what a typical ERP or point of sale system or what you have uh might offer. And just to give a realworld example, one of our great clients um is Nova Supermarkets. Um they're one of the largest supermarket chains in Europe. And you can see the results that they've achieved. Increase in sales with a big reduction in their overall inventory

levels. um better inventory turnover rates and a uh a much higher availability for each product. And so um all of this really comes to uh comes together well in practice and um helps our clients achieve much better bottom line results for their business. So with that, I wanted to say thank you so much for attending the webinar. Um you can see the uh the contact information here. So if you're interested in achieving these kinds of results in your business, please reach out to us. We'd be very very happy um to help. Thank you for your attention and have a beautiful day ahead. Thank you so much everyone. [Music]