Welcome to CONF42 Incident Management 2024. I'm going to talk about data driven retail, leveraging forecasting models to enhance customer experience and operational efficiency. There's a lot of, jargons here, but on a simple terms, I'm going to talk about how various forecasting models helps achieve, enhanced customer experience, in this fast paced world. Before getting into the topics, I would like to introduce myself. I'm, I'm Sijo Walekadmanikantan. I'm a data scientist with, over seven years of experience. my specializations have been primarily, with a huge forecasting, background. and when I talk about forecasting, it's forecasting at scale, dealing with, millions of time series and, with a lot of historical data. I'm also, active in the researching research space, authorship, and as well as mentoring junior data scientists and analytics professionals. educationally, I have a master of science from University of Texas at Austin, and a bachelor of engineering degree, in mechanical engineering. I would like to walk you through what I'm going to talk today. so, At the very beginning, I'm going to talk about the need for forecasting, with respect to retail. so first I will go into the nuances of, high level retail value chain. and then, I will introduce you to the. Retail landscape and the necessity of why, why advanced forecasting models are very careful consideration of forecasting models need to be, considered, in the retail space. in the next section, I will go over forecasting challenges, specifically regarding, Horizon Green and, the challenges brought about by COVID 19. I will also touch upon other, other aspects, other challenges, relating to these, and, and specific forecasting challenges. and, and once we understood the value chain as well as, the forecasting challenges, I will touch upon on a high level the three different classes of, forecasting techniques that's being used in the industry and that has shown, tremendous results, with respect to performance, accuracy, and so on. Then I will introduce you to some case studies, that a couple of case studies on how, certain retailers, global retailers have managed to, use forecasting to, to solve their supply chain issues. and towards the end, I would discuss where the future of forecasting is and what are the generic trends that's being seen in the retail space. as of as of today and it's a it's a very very bright future ahead as well as filled with a lot of challenges So without further ado, i'm gonna get started, with why do we forecast right? So first thing before even understanding why we need to forecast why need to make predictions about anything like it could be sales. It could be demand it could be revenues or it could be outages and so on any any You disruptions and so on. we need to understand the complexity of the retail value chain, right? So retail value chain is on a very, very, very high level is the chain. It's the, series of events that captures the flow of goods from manufacturers to consumers. Okay. So talking about consumers, we are all consumers. I'm sure everyone is aware of Amazon, Walmart, Kroger, and so on, right? these are the multinational corporations. I'm sure there are other, different sizes. I mean, retailers come in different sizes and scales, even, even, a neighborhood grocery chain will have its own retail value chain. So, we cannot, generalize, generalize. retail value chains, it's it's different. according to the companies and everything so but typically, we see these components in like ranging from manufacturers to consumers in in every supply chain, right? And sometimes it could be like a different order or different configurations, or, or it could be, each component might be tons, hundreds, might, it may be one, and it might be having a different sort of networks and so on, right? So in here we have, the manufacturer, right? So manufacturer is the one who manufactures goods. if we take, let's say, a Walmart store, for instance, any, or you can take any grocery store, for instance, right? They have tons of inventory, tons of stocks, tons of products, right? it's, you know, called SKUs, stock keeping units. each, each product is identified, each SKU code and, and, and all other things. Right. So if you walk into a Walmart, any store, you, you're, you're dealt with like, I don't know, hundreds and thousands of products, if not millions. Right. Like, a milk from a particular brand is, is, is a SKU. And then even if you look within a category of something like milk, right. There could be maybe, fifties or hundreds of hundreds of different SKUs comes in different sizes, different shapes, different packaging, and so on. Right. So these, so if you take a simple case of, one such product, it's being manufactured at some facility. let's say if you take a perishable goods, generally, like, like milk, for instance, it comes from the. same country or same vicinity as, as the places, but it's not guaranteed. Let's say something like a plastic product or a kitchen product might come from as far as China. Right. So we all know like China is the biggest manufacturing hub. So a lot of products come from China and then they need to be transported to, how does it reach consumers? They must be transported across the continents to reach the consumers. now the problem is we don't know how many products a customer want, right? So that's a problem. But when you talk about transport, I would like to go into the transport first. So transport could be I put an image of truck here, but it could be ship. It could be multi modal. It could be trains It could go from let's say ship to a train to a Car and then to a person in a bike to store and then consumer so it could be any combinations, right? So usually the the the manufacturer the goods move from manufacturers to distribution centers, right? distribution centers are warehouses where all these products are stored and depending on the You demand from the stores or consumers, the products move from distribution centers to particular stores and then later on consumers but there are several questions that question marks right like from a manufacturer to distribution center how many products do we stock up in the distribution center, right? So as we talked about like we have hundreds and thousands of products and we need to allocate the space in the distribution center According to according to the demand because the space is limited. We don't have unlimited space. So we should be able to Understand how much space is required in the distribution center and then we need to know You How big of a transport that's required and how frequently the transport vehicle should go from a distribution center to a store and how many products should be stocked up in a store, right? Like as a consumer, you don't want to go into a store and then the store associate telling you that, Hey, we have the product, but it's in a distribution center. I'll get it to you in a while. These days, people expect, deliveries and products on the same day. I'm sure like in the U. S., Amazon has same day deliveries. Most of the e commerce giants are trying for same day delivery, two hour delivery, three hour delivery, and so on. And in places like India, there's even 10 minute delivery, right? So we expect products to arrive. As quick as possible. So for that if we as retailers need to Anticipate customer demand, and then act accordingly. But the problem comes because when we, when the products move from manufacturers to consumers, these demands need to be anticipated in advance. For example, like, we need to know that, let's say, six months or one year in advance that the customers might need, in this area might need 10, 000 units of a particular jacket because that's a demand, right? So Anticipation is crucial here because the same thing if consumers get happy unhappy They might leave you for another retailers and and they will look for products elsewhere. Then they're not going to wait correct. So in summary, There are so many missing questions and missing, variables that could affect, different pieces. So, it's, it's, I would say there's like tons of variables, right? Like, like even, later I'll get into the challenges, right? Like COVID and things like those will shut down these retail value chain, which would cause issues, but consumers still want, products and stuff. Right. So that's why we need forecasting and forecasting at several grains, and several horizons and make for making, different decisions. Correct. So if you look at the landscape, right, the retail landscape and forecasting, need. There is long manufacturing lead times, right? Like, imagine that you're coming up with a new product, there is a product design that's involved. And then once the product is designed and then you send it to the manufacturing facility, let's say if it's in China, and then, you enter a contract and then, the products get manufactured and then they need to be shipped. So it takes a long time. So it's, companies probably are, are probably, you know, discussing about what product might come out, let's say one year or two years or even three years and four years from now, right? So you need, you need forecasting. And another thing is consumers are complex, especially, consumers as in, the day to day, uses, right? The retail users, people change, like, especially in this world, like you're all influenced by, social media trends and, and advertisements and so on, right? Like a celebrity wearing a particular, a jacket would shoot up the demand for a particular jacket. But these are, these are, but these are, unexpected demands. but still like the point is their behaviors and preferences change depending on a lot of, lot of, factors. Right. And another thing is with with a lot of retailers becoming omnichannel, like blending online, in store and hybrid experiences, it's, it's tougher, right? Like there are so many retailers offering order online and pick up in store and as well as hey, go to the store and place the order in the store. So. it's getting complex because, back in those days, the retailers used to estimate the traffic, a mall traffic, and like, it was a traffic based, systems and traffic used to be a great input for the transactions and essentially revenue and sales. But these days that traffic could be, could lead to no transactions and they could go into a store and place an order, or they could go into a store and then look up their phone and then. place an order from the phone and then, then, that would, you know, throw away the projections of the roof, right? And retailers need a real time inventory optimization for so and so reasons because things change rapidly and also right like competition is a big factor. with let's say Amazon for instance are are running prime day they every every retailer has their own one or two days of high discount high promotion events and then that would You know, lead to, loss of sales or maybe like, competition, taking away market share and things like those. So we need better decision making tools considering the competition in mind as well. So forecasting is, is extremely critical, because they can act as so many things, right? Like controlling costs, like we don't want to like, let's say, build up a lot of inventory in our, warehouse and take up space where we could, might as well like, Stock up on profit maximizing products, right? So forecasting is a very, very, very critical competence for all the retailers for maximizing sales, maximizing profits, and so on. So now let's look at, forecasting challenges, like why it's hard and why do we need different, scalable and ever evolving forecasting systems. so I'm not going to go into all the challenges here. I'm going to touch up on and give a sense of two main aspects, with respect to, the forecasting. So let's say, so I'm going to look at horizon, right? So horizon is nothing but, how long into the future are we looking at? let's say, if you're talking about, I don't know, one to two, three weeks, like what sort of decisions are the companies making, right? So let's say three weeks, like generally, generally those decisions are around. Store replenishment, sales and operations, execution, smooth delivery, distribution center replenishment, inventory allocation, end of season, clearance, seasonal promotions, and so on, right? So these are very short term decisions that, that the, store managers and, Location planners or inventory specialist or supply chain planners make to ensure that, the products get sold, the targets are met, inventory is sold out and so on. And, and the other one is, is around, like, let's say, 3 to 12 weeks. That's, we are talking about somewhere around, 3 weeks to, close to three months mark a quarter, right? So here the decisions would be mostly around workforce optimization capacity management And sales and operation execution smooth delivery flows distribution center replenishment and inventory allocation, right? so in these even there are a lot of overlaps because the forecast generally, gets more and more accurate, as we go closer, right? So it's easier to predict whether it will rain three hours from now or even tomorrow than one month from now, right? Because a lot of things are not in our control and it's harder to make predictions. especially when, like, there are so many things driving the demand as well as the supply. And at a three to 24 months mark, it's then it's a different ball game. Like we are moving a bit about from day to day operations to a high level strategic planning and things like those, right? So we talk about like, at this this A quarter out to let's say to up to a two year mark. It's mostly about assortment planning like what sort of assortments do we need to put in a particular store as well as not just store at a at a company level right let's say I don't know like a Milk company, let's say they will have not just milk. They will have dairy based products, right? So What what should be the optimal allocation of budget funds to create let's say an assortment of 50 products so that and as well as the units to like Maximize profits maximize sale and also meet the demand of all the consumers, right? So and space optimizations like as I talked about before You the retailers have limited space, at a given point in time, right? So, they need to ensure, ensure which products get what space. And so they need to, or if there's a new product development, a new product coming, they need to make sure that they need a lot of space and, and, and so on. Right. So. And also a long lead time purchasing, right? So companies usually like enter contracts with manufacturers and and Shipping partners way in advance. It could be as it could be long time more than 24 months as well, but like generally they enter into agreement based on the price that's that's being realized based on the future oil prices based on transportation costs based on manufacturing cost based on product cost based on material planning and so on right and also like sales and operation planning Like do we have enough workforce? to to function to to Drive, let's say store operations or it could be corporate workforce or it could be planners and so on right? And anything beyond 24 months Or even up, it could be, depending on the retailers could be like three years, but usually most of the retailers, 99 percent of the retailers have like a, two, two years cycle, with respect to, long horizon planning. so, but Anything beyond 24 months, even that's possible, right? Like companies are making five year, 10 year or even 20 year, strategic, planning. So companies still want to know what happens beyond the scale and to make strategic decisions on which directions to go and also to influence product design and development. product design and development, decisions as well as to enter into contractual obligations so that, the retailers can make profit. Like if you think some materials price is going to go up by, let's say 50%, it's better to lock in that contract right now to get it at a much cheaper price. So things like those, those are strategic decisions, that, that could be made. So every Decision is, is, should be happening at a very different time. So that means, let's say there's something like a demand forecasting system, and this, based on the use case, like, based on the teams, for instance, the, all the, use cases that we are seeing on the screen is handled by different departments, like, for instance, product design and development, versus, let's say, assortment planning. These are all different Teams, so these teams would need inputs at different points in time. So let's say, something like one to three weeks, the teams requiring these short term forecasts expect a very high level of accuracy, and they might, require it at a very, very, very small granularity. Let's say they would get and download, okay, this particular SKU of milk, what is its, I don't know, sales or demand or, or anything, right? So there, they track a lot of, inputs, like could be sales. It could be transactions. It could be, even stock outs, right? Prediction stock outs. So versus let's say, when we get into like strategic planning and product design, it could be on a high level, like, okay, maybe the grocery chain or, or even like a dairy, what is it? This or milk as a whole like, okay, how many gallons of milk or it could be at a very high level as well Right, so it's different every forecasting use case is different and sometimes we are forecasting for the same thing, but depending on the team different, granularities are required to ensure that the customer experience is elevated, the the Issues are solved at different, points in time. So that's, that's, about horizon. but there are so many other challenges as well, right? Like, as I said before, forecasting at a SKU level or a store or a region or a channel level, like for instance, Walmart has both, e commerce store and they also have physical stores. And so these are different, like, so maybe. for the same product, we need channel level sales. It could be, let's say a region level, forecast. So these are all, and they are all interconnected. They're all hierarchical problems, right? Like, let's say if the, the, any, retailer has like thousands of stores and then, Every state has let's say 50 stores and then the sum of all these store forecasts must align with the state level forecast and as well as region level forecast and then Ultimately country level and then global level, right? So That's challenging to make it work, as well as, short term versus long term, right? Like immediate was a strategic decision. So immediate is more around, yes, we need to add a high granularity and like the accuracy needs to be precise to make that final step in the process, as perfect as possible because it, it, it has the highest impact. I'm not saying the other one that does not have an highest impact, but still has a very high impact making, like, some mistakes in, in that level of the value chain can cost, a lot for, for the retailers. And, another thing is, since retailers operate on both e commerce and the store level, the data can come from, several sources, right? Like, POS data might come from the POS systems involved, installed on the stores. And these are physical systems. So there could be issues with the physical system. That's a different story. Like sometimes the store won't work, like some PO systems won't work and they might be faulty. Or, so then, then the other case of, yeah, we have POS systems and then, some people make purchases online with credit cards. Some people use cash So we don't have information about cash payment, but if you pay with a credit card you have you get much more information about the consumers and so on so That's data fragmentation and then there is an issue of Demand volatility, right? That's that's a tougher problem to handle. people people's intentions their purchase behavior customer behavior changes because of Several reasons we cannot predict all of them, but we can kind of look at the patterns and and Predict their behaviors and predict the demand based on or with some volatility and variance, of course You on the on the sales demand and other other metrics that retailers care about right and also There's a big trade off between accuracy and operational efficiency we can always work with High complex systems to, create highly accurate forecast, but sometimes training them generating forecast might take a huge hit on operational efficiency. one thing that I want to touch upon has the impact of COVID 19, not just with respect to impact of COVID 19. the reason why I brought this up is because, is to show that, Such things can happen. Unexpected things can happen in the world of, retailers, right? Like it's, it has been a wake up call on being resilient on, such events. And for instance, when COVID hit, the retailers had to make a lot of, a lot of, store level and like other changes like safe shopping and six feet distance and mask providing hand sanitizer offering, touchless payments and, and so free mask and like, yeah, ensuring some people, some retailers even like, install temperature control systems in front of the retailer stores to ensure safety for their customers. Right. And also people started working from home a lot and they started staying at home a lot. So they were shopping Online a lot. So the consumer behavior changed they were looking for And the retailers started facing challenges with respect to some products might look good on a screen now the products are our product visibility depend is dependent on how it looks on a particular smartphone or whether people are shopping on a iPhone versus android or is it on a laptop or I don't know right like and also a lot of sharing like if you purchase something online you tend to like Ask somebody close to you or or you're going to share it with a friend and so on. So Tons of consumer behavior change happened during, the pandemic, as well as there was a lot of supply chain disruptions, right? Like when, in the pandemic started in, in Wuhan in China, a lot of factories, manufacturing factories were based out of Wuhan and that caused them to shut down these, factories and it was affecting the inventory management. Like, they couldn't manufacture a lot of things. So. Even though there was demand the supply was going down supply was Just what is available in? warehouses or whatever is there in transit? So there was a lot of shortages for a lot of items and it cost the prices to shoot up Costing inflation and so on right? So even now, there is an influence of COVID 19 on forecasting systems with respect to data handling and like, like having a different consumer behaviors. A lot of retailers have to, you know, rethink their forecasting systems. And it actually, you know, thought about, brought in a change with respect to having, resilient and flexible forecasting systems in the industry. So, how do we forecast these demand and the metrics that are key for retailers, right? So, on a high level, there are three classes of, actually, three classes I'm going to talk about, which is, these are the most common, but there are more. First, we have statistical models, right? Which includes methods like, ARIMA or exponential smoothing. And these models work really well for, detecting simple trends, seasonality, and stability. develop stable environments, but they struggle a lot with rapid and complex shifts in the data because they rely a lot on historical data and linear patterns, right? Next, we have machine learning models. Unlike statistical models, these are data driven and they can capture nonlinear relationships between variables. we can also incorporate external factors such as weather, promotions, even competitive behaviors, into the forecast, and that makes it, more robust compared to statistical models, right? Then we have deep learning models, specifically neural networks like RNN, LSTMs, and so on. These models are particularly effective, for time series forecasting in complex and dynamic environments. I mean, they can process vast amount of data and find hidden patterns that other, models can't. but, one caveat here is as we go for more complex models, like deep learning models. We need more and more data, right? Like, if you have only let's say 50 data points deep learning models might not be The right approach even a statistic statistical models or a simple models might even naive models could perform better than deep learning models just something to keep in mind. So talking about statistical Forecasting techniques and there are many i'm just i'm going to touch upon. the to common and most popular ones that's, being used in the industry, which is the ARIMA, which stands for autoregressive integrated moving average for time series forecasting. And then there are, different flavors of exponential smoothing that is, right for detecting trends and seasonality. so there is the exponential smoothing is in different kinds. They can be a simple exponential model to like seasonal ones like holds winter and so on. but they're extremely useful in detecting trends and seasonality. we could also be using regression analysis to understand relationship between variables. just, so I've seen people using regression analysis for forecasting as well, like. it needs to be carefully examined because of the assumptions of the regression analysis. So the, these models are simple, interpretable, and they're really good for relatively stable demand. one of the problem here is, in highly volatile or nonlinear environment, they may not be the best choice. then comes the data science and machine learning techniques. Moving beyond traditional statistics, data science and machine learning have transformed how we approach forecasting in retail. Machine learning models such as let's say decision tree, random forest, and gradient boosting can process enormous amount of data and make sense of complex non linear patterns. So one of the key advantages of machine learning is its ability to handle large data sets, incorporate multiple variables, including external factors like promotion, weather, or even social media trends. for instance, a machine learning model might predict increased demand for a product if it detects a sudden surge in online searches or social media mention, right? Machine learning models are also adaptive, unlike statistical models that rely on predefined rules. Machine learning models learn and adjust as new data becomes available. This means that they can quickly detect shifts in consumer behaviors and adjust forecasts accordingly. one example is, how some retailers use machine learning to predict the success of a new product, product launch by analyzing social media trends, historical sales data, or even, influencer mentions, right? this was an interesting case study by one of the retailers. However, there are trade offs. machine learning Models, require more data than statistical models, and also they require greater computational power compared to traditional methods. And they're highly flexible, but at the same time they're also more complex and harder to interpret, especially when, it comes to understanding why the model made up data. A specific, prediction, right? There are explainable models as well, but like, they're all, they're all still considered black box. We can, make, explain, the model, features at a high level, but still the, stakeholders and the business, folks who make operational decisions, still don't understand at the core. And there's a trust issue when it comes to machine learning models. And these things need to be carefully implemented, when productionizing. The other class of model is deep learning models. I've mentioned two of them, there are different others as well, but, RNNs, for time series data, like if you have sequential patterns, RNNs, RNNs picks it up very well, same thing, LSTM networks specifically, for capturing long term dependencies, right? So, like even, even, so if you. model, like even like they can capture like external, what do you call it? factors as well, like events, promotions and so on, along with capturing the sequence based, patterns. and they are usually really good for long term forecasting. and what I've seen is like a retailer could use LSTM model to predict for a new product category over the next 12 months, taking into account sales, data, competitor, competitor actions, and even economic trends. however, deep learning models comes at a very high cost. They have very high training times, complexity, and they require significant amount of data, computational resources, making them more suitable for larger retailers with big data sets. One of the bigger drawback, according to me, is they can be challenging to interpret as their decision making process is often seen as a black box compared to more transparent methods like regression or statistical methods. So this is a few tips on, implementing, time series models, in the industry or for specifically your, use case. this is very, extremely critical, right. On, on trying to understand which models to start with, right. In the first place, I've seen like people jumping right into, complex models, but, they don't achieve the results that you're looking for. Right. So, according to me, these findings. Super important starting with EDA. EDA is extremely important to Understanding the task and exploring your data It is crucial to first conduct EDA to get a sense of data's patterns trends and outliers This ensures that you're not missing any key insights and helps you verify that you're solving the right problem with the correct approach. It also solves a bunch of other issues like, Oh, looking at the data quality and like getting ideas for your forecasting problems and so on. Also check for gaps and stationarity in the time series. Time series models often require stationarity and gaps in the data can seriously disrupt forecasting accuracy. Addressing these issues early on will save you trouble later. Now if you're talking about evaluation, select the right evaluation metrics for your problem. Depending on your goals, you might use MAPE, WMAPE, or any other metrics, but, don't forget to use simple benchmarks like Naive Forecasting set a baseline. If your advanced models can't beat this, there is something wrong, right? And when it comes to models, start simple, avoid diving straight into deep learning or LSTM unless you have plenty of data, simpler models can often be more effective and interpretable, especially when data is limited, right? And another thing is about interpretability. It is extremely essential to know your models, right? You should be able to explain why you selected a particular model and understand how it works under the hood If you can't justify your choice, it is it's a sign that something's off, right? there are many other aspects you need to consider while implementing such a system forecasting systems, but Generally, i've seen people making mistakes with respect to these right, so Next, let's quickly go over some case studies on like some practical ways. Some retailers have Fixed used forecasting to solve some real world challenges The I don't have specifics on how they on certain aspects certain details, but like this will give an overview about how Companies are using it, right? So in this case study, we are looking at a global retailer with over 10, 000 stores across different regions. Their primary challenge was frequent stockouts, especially during peak seasons. This led to lost sales, customer dissatisfaction, and inefficiency in managing supply chains. The company's incident management was largely reactive. Whenever stock odds occurred, the supply chain team had to rush to make last minute replenishments with significantly increased costs. To address this, the retailer implemented a machine learning based demand forecasting system. This model didn't just rely on historical sales data, but also incorporated External factors like weather, regional events, to predict demand more accurately. they saw, due to this, they saw a 30 percent reduction in stockouts during peak seasons. And they also managed to cut down on emergency logistics by 25%, leading to more cost effective replenishment of the inventory, right? Most importantly, the incident management process shifted from being reactive to proactive. instead of waiting for stockouts to occur, the forecasting model provided early warnings so that they could address the issue before it impacted the store. This not only reduced the operational stress, but also improved customer satisfaction as products were consistently available when needed. This is a great example of, of how. proactive incident management can lead to, extremely, well balanced and, highly efficient, system and how it can lead to customer, improvement in customer experience. Right. the next one is around, QSR chain, QSR stands for, Quick service restaurants like fast food restaurants like McDonald's and KFC and and so on right? So their main challenge was supply chain disruptions which led to ingredient shortages And, and it caused major service delays when, let's say, certain ingredients, ran out. Customers were unhappy because popular menu items weren't available. Like, imagine eating a, a burger when, without lettuce. I mean, they won't sell it, but still, like, if the lettuce was, not available, the companies or the stores couldn't make the, such products involving lettuce, right? So even, same thing, their incident management team was mostly reactive and constantly responding to these disruptions rather than preventing them. This resulted in higher costs and more importantly, a negative impact on the customer experience. And later the QSR chain introduced AI driven demand forecasting models. The model analyzed a wide range of data, historical sales, weather, And eventually they were able to achieve 20 percent shortage in ingredient, in ingredient shortages, right? And the forecasting model allowed the chain to plan ahead, ensure that supplies are delivered on time in the right quantities. and the forecasting system helped, reduce the food wastage by 15 percent as they could better align this, supply with actual demand, right? So this is a move towards achieving sustainability goals as well. So as a result, their incident management became proactive with better supplier coordination. So this is how you can at several companies, are using forecasting to drive, customer experience and reduce. become more reactive, become more proactive, supply chain and develop proactive incident management systems, right? Now what's up, about the future for, forecasting in the retail world? There are several key trends. we see an increasing use of IOT data for real time inventory tracking and management. Right. So these are more data coming in at a very high pace, which will help us, make more, real time decisions. And then, there's another, trend that's being seen in the industry, which is to incorporate social media and external data for demand sensing. when talking about external data, there are data from events, like, let's say a concert or, or sports events and so on, as well as, with a lot of consumers being online, we have more and more data about their presence, their shopping habits, their trends and so on. So companies have started, incorporating such things into, into the demand sensing systems. Then, predictive analytics for customer personalization and marketing optimization. it's self explanatory. like we see, all the e commerce and like shopping websites giving us personalized recommendations, sending marketing emails, and so on. also there's a lot of, ethical, aspects, right? because we Retailers, retailers collect a lot of data, like when you enter your, there's your payment information. There is, yes, transactions, your, your interest. So over a period of time, the companies have a huge, huge, huge, data, And they know a lot about you, right? They collect data and they see the trends and they have millions of such customers online for these big retailers and they know what you like and what you don't like. And some countries, some regions, they have their own privacy rules and data rules and so on. So that's, having an impact these days and companies are getting more, taking more actions on their ethical friends as well. So now also. A lot of companies have achieved this the fully automated demand sensing forecasting systems driven by AI there's a huge trend right now with llms generative ai and personally in the industry I think it would be a great great Tool, you know to synthesize a lot of natural language with respect to let's say Forums social media forums and like social media data and text data and synthesize that As, features into the demand sensing models to, capture those peak peaks in the demand, right? So, and as I said, like, there's a lot of uncertainty volatility, with respect to retail and retailers must adopt, AI powered models to remain competitive. That's that's that's there's no doubt about that. The forecasting has come a long way, but there is a need for continuous innovation right So demand forecasting cannot just rely on traditional statistics with customer preferences and marketing conditions shifting so rapidly. the businesses need models that can learn, adapt, predict in real time. And Then this means that they need to invest in scalable forecasting systems that are capable of processing large data sets quickly and generating accurate predictions. the other key point is collaboration. retailers need to work closely with data scientists and technology providers to ensure their forecasting models are not only accurate, but also flexible enough to respond to new challenges. The complexity of today's retail environment, means that no single team or department can tackle these issues alone. One example of, continuous, innovation, is, is, with the Retailer X. I mean they have a I powered forecasting system that predicts the demand for products to individual store location by integrating multiple sources including store traffic weather online behavior so they can this retailer can predict the demand shifts in real time and adjust their inventory accordingly like even at a Weekly or like one week cadence, right? So ultimately the future of retail depends on mastering predictive analytics for short term and long term gains so those who can harness the power of AI and machine learning to forecast demand accurately, will not only enhance their operational efficiency, but also gain a critical competitive advantage in this ever changing and ever evolving market, right? So, so, so we, we talked about how, we talked about the retail value chain and we talked about how there could be a lot of incidents, in the retail value chain that could disrupt the businesses. so it, we, we went through how forecasting, can, you know, help, fix those issues and improve customer experience and just building, smart, resilient, robust, supply chain systems, supply chain businesses, operations to improve customer experience. Right. So there's a lot, Of topics covered today, and some of them are very high level, some of them, I try to go as in depth as possible. but if you have any questions or a feedback, for this particular talk, please feel free to reach out to me on, on my email. you can also reach out to me on LinkedIn, if you can find me there. but I'll try to respond as soon as possible and I enjoyed presenting it here, and thank you so much everyone.