View Recording: Custom Demand Planning for Retailers and Manufacturers

Brandon Dey Hey, everybody, happy Friday, junior. 0:0:12.602 –> 0:0:13.752 Brandon Dey Otherwise known as Thursday. 0:0:15.652 –> 0:0:18.402 Brandon Dey Amy, did you want to kick things off or do you want me to go? 0:0:18.712 –> 0:0:19.982 Brandon Dey I know we just spoke about this. 0:0:19.42 –> 0:0:19.962 Amy Cousland You go ahead and get started. 0:0:20.612 –> 0:0:23.402 Brandon Dey OK, everybody, thanks for taking the time. 0:0:23.632 –> 0:0:24.842 Brandon Dey My name is Brandon. 0:0:24.972 –> 0:0:27.402 Brandon Dey I am head of engineering at concurrency. 0:0:27.602 –> 0:0:34.322 Brandon Dey It’s great to connect with you guys digitally or at least with your circular initials. 0:0:35.312 –> 0:0:41.672 Brandon Dey I’m gonna be walking through custom demand planning options specifically for manufacturers and retailers. 0:0:42.382 –> 0:0:45.92 Brandon Dey Umm and I am super pumped about it. 0:0:45.102 –> 0:0:57.412 Brandon Dey This is a a really popular use case for for manufacturers, for retailers and for folks either looking to get started with AI or just improve their business. 0:0:57.422 –> 0:0:59.412 Brandon Dey So it’s a good sweet spot. 0:0:59.642 –> 0:1:2.872 Brandon Dey I have 3 specific goals and they’re pretty basic. 0:1:2.882 –> 0:1:4.882 Brandon Dey The first is you learn something useful. 0:1:5.762 –> 0:1:12.852 Brandon Dey There’s a lot of noise and a lot of misinformation online about demand planning and and AI in general. 0:1:13.542 –> 0:1:14.902 Brandon Dey Would love for you to learn something useful. 0:1:16.232 –> 0:1:18.42 Brandon Dey #2 is you. 0:1:18.92 –> 0:1:25.472 Brandon Dey My goal is for you to stop thinking about build or buy and I’ll get into what you should be thinking about instead. 0:1:26.262 –> 0:1:32.922 Brandon Dey And then lastly, I hope you have a path forward whether you’re thinking about AI or or demand planning in particular. 0:1:35.602 –> 0:1:40.752 Brandon Dey So the reason demand planning is is the topic of choice is because it’s super valuable. 0:1:40.922 –> 0:1:50.82 Brandon Dey It’s often low hanging fruit from a wanna get started with an impactful set of technologies standpoint. 0:1:51.502 –> 0:1:53.622 Brandon Dey It’s relatively easy to automate. 0:1:53.742 –> 0:2:4.742 Brandon Dey There are certainly some challenges, but and it’s not a a set it and forget it, but relative to other AI use cases, it lends itself well to full automation. 0:2:6.362 –> 0:2:10.392 Brandon Dey It also has synergy with another use case, which is price optimization. 0:2:10.832 –> 0:2:12.632 Brandon Dey We’re not gonna get into price optimization. 0:2:13.102 –> 0:2:20.542 Brandon Dey That’s a topic for another webinar, but demand planning models play very, very nicely with that application, and together they can drive a lot of value. 0:2:23.992 –> 0:2:40.892 Brandon Dey Demand planning also produces these models that allow you to understand your business from a from a causal standpoint, and even if not from a causal standpoint, you still at least learn the useful correlations that allow you to take more effective action. 0:2:41.302 –> 0:2:54.762 Brandon Dey So it’s really one of one of the use cases that allows you to understand reality at a higher level of resolution and the better you understand reality, the more effective your your actions will be so. 0:2:55.352 –> 0:2:55.882 Brandon Dey Umm. 0:2:56.352 –> 0:2:58.322 Brandon Dey Super useful for that for those reasons. 0:3:1.432 –> 0:3:5.722 Brandon Dey Like I mentioned, it’s also a gateway to higher impact AI projects. 0:3:7.352 –> 0:3:11.522 Brandon Dey Now the goal of demand planning though is is pretty simple, right? 0:3:11.532 –> 0:3:21.552 Brandon Dey We wanna make more money and we wanna do that by forecasting the future demand for your product catalog. Right. 0:3:21.562 –> 0:3:28.72 Brandon Dey And you, you really make more money by either reducing your overbuy costs or reducing your under buy costs, right? 0:3:28.82 –> 0:3:39.362 Brandon Dey So on the overbuy costs side, that’s things like increasing your inventory turns, decreasing your carrying costs because you’re more effectively allocating capital to your inventory. 0:3:40.592 –> 0:3:44.432 Brandon Dey And on the reduction of the underbuy side, as this stockouts. 0:3:44.442 –> 0:3:44.582 Brandon Dey Right. 0:3:44.592 –> 0:3:46.292 Brandon Dey If you’re a retailer, they what you call them. 0:3:47.62 –> 0:3:54.642 Brandon Dey That allows you to stop foregoing revenue because you didn’t have product to sell, right, increasing your revenue margin. 0:3:55.482 –> 0:3:56.292 Brandon Dey And hopefully I know why. 0:3:59.262 –> 0:4:0.512 Brandon Dey But who are we anyways, right? 0:4:0.522 –> 0:4:1.212 Brandon Dey Who is concurrency? 0:4:1.222 –> 0:4:7.472 Brandon Dey So we’ve helped a number of organizations with custom AI demand planning on the slide. 0:4:7.482 –> 0:4:19.52 Brandon Dey Here is an example of how we helped a a large restaurant chain save a bunch of money, basically by better forecasting demand for their for their entrees. 0:4:19.982 –> 0:4:26.792 Brandon Dey So essentially the big problem that they had was, you know, they’ve got, you know, almost 700 stores across most of the country. 0:4:28.882 –> 0:4:36.862 Brandon Dey The store managers at each of those stores are responsible for creating schedules for their employees and those schedules. 0:4:36.872 –> 0:4:38.242 Brandon Dey You know when they’re going to work. 0:4:40.262 –> 0:4:44.692 Brandon Dey Those are reliant on expectations of how many people are going to show up for dinner that night, right? 0:4:44.702 –> 0:4:50.992 Brandon Dey Or a particular hour, and so the the managers weren’t as good as they could have been in doing that. 0:4:51.2 –> 0:4:53.902 Brandon Dey And it led to a lot of excess wasted labor. 0:4:54.922 –> 0:4:55.482 Brandon Dey Umm. 0:4:55.682 –> 0:4:56.702 Brandon Dey Or so. 0:4:56.712 –> 0:5:7.922 Brandon Dey Basically, people were staffing empty restaurants or on the flip side, they had longer customer wait times because you know, they’re they’re there weren’t enough people to so staff. 0:5:9.632 –> 0:5:10.712 Brandon Dey So we came in and we helped. 0:5:10.722 –> 0:5:16.192 Brandon Dey We essentially built them a a system to predict how many people were going to eat out. 0:5:17.582 –> 0:5:21.412 Brandon Dey At a particular hour for each of their product lines, right? 0:5:21.422 –> 0:5:26.192 Brandon Dey So, like eating in dining out, etcetera, for every one of the restaurants. 0:5:28.2 –> 0:5:36.682 Brandon Dey What was unique about this project is that we used local weather Demand and local event data in the predictors, right? 0:5:36.692 –> 0:5:48.42 Brandon Dey So if there’s a big football game coming up, that’s certainly going to affect or to what degree is it going going to affect people going out to eat umm and then whether whether it was used as well? 0:5:49.972 –> 0:6:2.822 Brandon Dey Now it was it fully automated solution in that basically going end to end from raw data to useful prediction of how many entries are gonna be eaten at a particular hour. 0:6:3.112 –> 0:6:4.752 Brandon Dey That was all all automated. 0:6:5.352 –> 0:6:15.362 Brandon Dey There were certain cases where you want to take it offline and do some do some work, but by and large was automated and as you can guess, the result of this was that, you know, saved him a bunch of money. 0:6:15.592 –> 0:6:24.962 Brandon Dey Basically, the umm ROI was backed into by comparing store store manager forecasts with the new system forecasts. 0:6:25.362 –> 0:6:29.582 Brandon Dey And you know the system that we built was. 0:6:31.122 –> 0:6:37.752 Brandon Dey Not over forecasting as often and not under forecasting, resulting in a Mist revenue opportunity. 0:6:40.782 –> 0:6:49.72 Brandon Dey Also, before I continue, I should say feel free to reach out in the chat with any questions you have and if I see them, I will. 0:6:49.112 –> 0:6:51.622 Brandon Dey I’ll answer them right away or get to them at the end. 0:6:57.952 –> 0:7:8.762 Brandon Dey OK, so similar use case we we helped a uh basically a large furniture retailer umm forecast. 0:7:8.772 –> 0:7:10.592 Brandon Dey How many couches were gonna be purchased? 0:7:12.572 –> 0:7:15.172 Brandon Dey In the future, save them a bunch of money right? 0:7:15.182 –> 0:7:34.482 Brandon Dey For the for the same reasons this one was unique in in that we were the forecast horizon was so far out, it was really a 10 to 26 weeks, which is right on the cusp of as far out as you can go for even relatively stable businesses and we’ll get into to more of that in a minute. 0:7:36.362 –> 0:7:53.492 Brandon Dey So I’m not going to drain this whole slide, but what was unique about this was how we were able to help this furniture retailer address really stubborn problems that historically they hadn’t been able to address right specifically intermittent demand, right? 0:7:53.782 –> 0:8:1.432 Brandon Dey When consumers buy things in sort of fits and starts, it makes it really hard to predict and we didn’t have a perfect solution. 0:8:1.442 –> 0:8:4.152 Brandon Dey That perfect solution to intermittent demand does not exist. 0:8:4.662 –> 0:8:12.582 Brandon Dey You can ask Walmart or Amazon or any of the large retailers that but we are able to we were able to go a long way in in addressing that. 0:8:13.382 –> 0:8:20.42 Brandon Dey There are other things like erratic or lumpy demand and I’ll get into some of the the fun words for that in a bit. 0:8:22.342 –> 0:8:24.732 Brandon Dey But in general, we save them, save them money. 0:8:24.742 –> 0:8:26.712 Brandon Dey They have over 100 stores. 0:8:27.402 –> 0:8:42.662 Brandon Dey It’s more accurate than the human supply chain analyst forecasts, and I should say the other thing that was unique about this project was how we were brought into automate this decision, right. 0:8:42.672 –> 0:8:58.942 Brandon Dey This this decision of should we buy a more couches right now or not, in part because the supply chain analysts weren’t using the existing technology that they had access to, so they already had tools and forecasts, but they didn’t trust it, right? 0:8:59.332 –> 0:9:1.382 Brandon Dey They didn’t trust it because they didn’t understand how it worked. 0:9:1.392 –> 0:9:11.22 Brandon Dey And so this was a big part of what we did on this project too, is surfacing why the system is making a prediction so that people can can trust it. 0:9:11.32 –> 0:9:13.392 Brandon Dey And and reason about if it’s worth trusting or not. 0:9:16.162 –> 0:9:21.772 Brandon Dey Alright then the last thing I’m going to save share is essentially how we we helped Kroger out with. 0:9:24.292 –> 0:9:32.522 Brandon Dey Uh, their labor schedules by adding whether information into their forecasts of grocery Demand. 0:9:33.572 –> 0:9:35.442 Brandon Dey They’re obviously a huge scale. 0:9:35.452 –> 0:9:44.762 Brandon Dey They’re the largest traditional grocery retailer in the country, and so any percentage point that you save them in excess labor is a very big number. 0:9:45.732 –> 0:9:47.942 Brandon Dey So this was this was a very fun project. 0:9:52.172 –> 0:9:53.382 Brandon Dey All right, so I’m going to dive into. 0:9:55.172 –> 0:10:1.642 Brandon Dey What we actually mean when we talk about Forecasting: case, there’s a lot of different flavors of Forecasting:. 0:10:2.612 –> 0:10:5.182 Brandon Dey There’s really execution forecasting. 0:10:5.192 –> 0:10:7.422 Brandon Dey This is on the the short short end. 0:10:7.432 –> 0:10:8.382 Brandon Dey The near term side. 0:10:8.392 –> 0:10:32.222 Brandon Dey So if you’re forecasting anywhere you know sub two months, this is sometimes called demand sensing as well, but it really supports things like price optimization and very local inventory replenishment and doing things like order promising as soon as we shift beyond and we look beyond that eight week Mark and we’re forecasting from 9 weeks to up to a year. 0:10:32.332 –> 0:10:34.342 Brandon Dey We’re doing more operational forecasting, right? 0:10:34.352 –> 0:10:37.322 Brandon Dey So it’s similar to execution forecasting. 0:10:37.332 –> 0:10:46.352 Brandon Dey It does support inventory allocation, some procurement promotion planning, but these are higher order things, right promotion planning that need. 0:10:48.272 –> 0:11:1.542 Brandon Dey Needs stability and sort of foundation of expectations that execution Forecasting: just can’t support and anything beyond that is more strategic forecasting and we’re not really, we’re not gonna get into that. 0:11:1.552 –> 0:11:24.662 Brandon Dey But these are things like umm or this is really an activity that supports things like, you know, rationalizing your your product offerings or figuring out what what markets to expand into from a product portfolio standpoint, doing things like capacity planning, all super important activities, but not in the purview of the forecasting that I’m gonna spend most of our time talking about. 0:11:28.112 –> 0:11:32.932 Brandon Dey Alright, so there are three things. 0:11:33.12 –> 0:11:37.812 Brandon Dey Three applications of demand planning models, and they’re often lumped together. 0:11:37.822 –> 0:11:43.82 Brandon Dey So I’m gonna mince them apart and spend a little time on some of the nuance. 0:11:43.92 –> 0:11:48.132 Brandon Dey Because it’s an important part of value creation, with these demand planning models. 0:11:48.142 –> 0:11:54.132 Brandon Dey So the first application that Demand Planning unlocks is what you think about this is inventory planning, right? 0:11:54.202 –> 0:12:1.632 Brandon Dey It addresses things like, hey, how many couches do I need in the next 26 across the next 26 weeks, or how many widgets? 0:12:1.902 –> 0:12:3.572 Brandon Dey And we gonna need in the next 6, right? 0:12:4.562 –> 0:12:14.822 Brandon Dey This is all inventory planning allows you to uh by new stock in expectation of satisfying imminent demand. 0:12:14.892 –> 0:12:15.112 Brandon Dey It. 0:12:17.592 –> 0:12:21.142 Brandon Dey The next application of demand planning, though, is market response analysis. 0:12:21.152 –> 0:12:29.852 Brandon Dey This is a little little more rare for a variety of reasons, but umm, it’s essentially uh. 0:12:30.82 –> 0:12:31.692 Brandon Dey Elasticities of demand, right? 0:12:31.782 –> 0:12:42.882 Brandon Dey If if I charge X percent more for for this couch or for this component, you know what’s gonna happen are more people are gonna buy it or less? 0:12:44.12 –> 0:12:48.372 Brandon Dey Umm, so this is really about figuring out how elastic your product portfolio is. 0:12:49.812 –> 0:12:54.302 Brandon Dey And of course, as you can imagine, relies on pricing information, right? 0:12:54.312 –> 0:13:5.402 Brandon Dey So it relies on observing different prices for the same product at different points in time, or even at the same point in time in order to observe different levels of demand at those different price points. 0:13:6.402 –> 0:13:12.322 Brandon Dey Not everybody has access to that because not all products go on sale and that might not be a part of your your strategy. 0:13:14.502 –> 0:13:20.472 Brandon Dey The last thing that or really the third application that Demand Planning locks is is price optimization like we talked about. 0:13:21.42 –> 0:13:28.212 Brandon Dey So given that you now know why you’re demand moves in the ways that it does, how can we adjust our pricing, right? 0:13:28.222 –> 0:13:30.592 Brandon Dey This sounds really similar to market response analysis. 0:13:30.702 –> 0:13:37.492 Brandon Dey It’s really market response analysis and demand planning are both inputs to price optimization. 0:13:37.502 –> 0:13:43.312 Brandon Dey So, umm, Demand plan is often a precursor to your pricing optimization work. 0:13:47.122 –> 0:13:54.472 Brandon Dey I know I’m going through stuff kind of fast, but like I said, feel free to that information or ask your questions in the chat. 0:13:55.912 –> 0:14:0.432 Brandon Dey OK, so the the core tasks of demand planning. 0:14:1.432 –> 0:14:1.792 Brandon Dey Umm. 0:14:3.312 –> 0:14:9.182 Brandon Dey Really depend or sorry rather our our point forecasting and and probabilistic forecasting, right, so. 0:14:10.892 –> 0:14:13.732 Brandon Dey Both are important. 0:14:13.802 –> 0:14:17.482 Brandon Dey Most people forget about the second one probabilistic forecasting. 0:14:17.492 –> 0:14:24.952 Brandon Dey This is hey, I predict that anywhere between, you know, 12 and 15 couches are going to be demanded tomorrow, right? 0:14:25.682 –> 0:14:27.652 Brandon Dey There’s an upper and a lower prediction interval. 0:14:29.782 –> 0:14:40.132 Brandon Dey Point forecasting is just the number in the middle of that that bound, not playing anybody mind here, but with this, but you can also do scenario evaluation, right? 0:14:40.142 –> 0:14:46.172 Brandon Dey So what happens if X happens right where X is something within the demand model, right? 0:14:46.182 –> 0:14:47.422 Brandon Dey So if you’re accounting for. 0:14:49.42 –> 0:15:2.232 Brandon Dey Things like price variation, for example, you can say well, what if price goes up by experts and what’s gonna happen to demand or if you have your marketing campaign information or sometimes like called marking effort. 0:15:2.342 –> 0:15:10.872 Brandon Dey If you have Marketing effort as a driver of demand and you say, well, what if I exert less marketing effort in the next time period, what’s gonna happen? 0:15:11.202 –> 0:15:17.272 Brandon Dey That’s all scenario evaluation and then decomposition is really just breaking you. 0:15:17.282 –> 0:15:22.732 Brandon Dey You take a say you forecast a wave right of demand over some time horizon. 0:15:22.982 –> 0:15:30.92 Brandon Dey Decomposition is basically just deconstructing that wave into constituent parts that helps you understand demand. 0:15:30.102 –> 0:15:31.782 Brandon Dey So what are the seasonal parts? 0:15:31.792 –> 0:15:33.582 Brandon Dey What are the trends and so on? 0:15:38.272 –> 0:15:57.492 Brandon Dey OK, so each of these are important to to do and I wanna point out that each of these tasks often need to be performed at different levels of granularity across different different time horizons, different levels of accuracy or demand patterns, and even downstream use cases. 0:15:58.542 –> 0:16:4.92 Brandon Dey And so it’s conceptually possible to do all of these things with with one. 0:16:5.62 –> 0:16:11.782 Brandon Dey It’s common really to build multiple specialized models for different use cases, and I’m gonna say more on that in a bit. 0:16:14.892 –> 0:16:17.332 Brandon Dey OK, So what goes in to Demand Planning? 0:16:20.532 –> 0:16:38.642 Brandon Dey Typically there are four different arc types of inputs or in machine learning land, we call them features, but those four inputs are pricing information, historical sales information or historical demand information. 0:16:38.812 –> 0:16:44.752 Brandon Dey Whatever The thing is that you’re trying to predict in the future, an input is the history of that input, right? 0:16:46.32 –> 0:16:47.882 Brandon Dey The third thing is product taxonomy. 0:16:48.132 –> 0:16:50.962 Brandon Dey You know what category does this skew belong to? 0:16:50.972 –> 0:16:51.902 Brandon Dey What subcategory? 0:16:51.912 –> 0:16:52.502 Brandon Dey And so on. 0:16:52.992 –> 0:16:59.822 Brandon Dey And then lastly the the the 4th ARC type of input is temporal information or stuff you get from the calendar, right? 0:16:59.832 –> 0:17:1.602 Brandon Dey So what hour the day is it? 0:17:1.612 –> 0:17:2.542 Brandon Dey What’s the day of the week? 0:17:2.552 –> 0:17:3.272 Brandon Dey Day of the month, etcetera. 0:17:5.722 –> 0:17:11.52 Brandon Dey Now, pricing is typically among the most important factors that influence demand on products and services. 0:17:11.62 –> 0:17:23.902 Brandon Dey So, umm, super important to include that, but it’s often the input that people don’t have data on in general or reliable data. 0:17:24.12 –> 0:17:33.362 Brandon Dey In the case that they do have it, and there’s a lot of good reasons for that, but it’s often one of the things that you just don’t include because it’s data quality stuff. 0:17:35.442 –> 0:17:35.722 Brandon Dey OK. 0:17:35.732 –> 0:17:44.242 Brandon Dey So typically you can expect to see the same pattern in terms of which of those inputs are more important than the other. 0:17:44.352 –> 0:17:48.582 Brandon Dey And this isn’t like a rule, but it’s just a a general sort of heuristic. 0:17:48.592 –> 0:17:52.952 Brandon Dey For what you can expect so centrally, price tends to be the most important. 0:17:53.872 –> 0:17:58.152 Brandon Dey Umm, historical demand is is also super important. 0:17:58.442 –> 0:18:8.42 Brandon Dey And then you get some calendar features like the week before moving into product catalog and then you get sort of lower level calendar feature says you’ve been less important. 0:18:10.272 –> 0:18:13.922 Brandon Dey Now, this is obviously a general abstraction for what’s important there are. 0:18:15.852 –> 0:18:26.622 Brandon Dey Lots of Custom, very bespoke drivers of your business that can and often do get included into models that would also show up in in a graph like this. 0:18:30.772 –> 0:18:36.162 Brandon Dey OK, so I mentioned some funny words before lumpy or attic. 0:18:36.172 –> 0:18:37.962 Brandon Dey Intermittent smooth. 0:18:37.972 –> 0:18:39.172 Brandon Dey These are all or. 0:18:39.182 –> 0:18:43.482 Brandon Dey These are rather the four ways of characterizing demand, right. 0:18:43.572 –> 0:18:55.92 Brandon Dey And so you essentially have smooth, smooth demand, which is up here in the upper left, right, where it’s you get this nice, very stable, very smooth. 0:18:55.782 –> 0:19:0.392 Brandon Dey Umm to me demand levels and that’s obviously pretty easy to predict, right? 0:19:0.402 –> 0:19:5.812 Brandon Dey There’s a lot of consistency there and it’s easy for our models to learn the patterns. 0:19:5.822 –> 0:19:17.552 Brandon Dey Now imagine though you get a little more variability in your timing and and you get you get Demand that’s called erratic, right, so. 0:19:19.372 –> 0:19:20.542 Brandon Dey You or yeah. 0:19:20.552 –> 0:19:26.832 Brandon Dey So essentially it’s it could be stable if the the peaks and valleys weren’t quite as high or weren’t quite as low. 0:19:28.272 –> 0:19:30.502 Brandon Dey It makes predicting demand tricky. 0:19:32.142 –> 0:19:47.152 Brandon Dey If we move down to the lower left though, and it’s very variable in terms of demand and but it’s relatively ohh fixed the magnitudes you get what’s called intermittent demand. 0:19:47.442 –> 0:19:48.812 Brandon Dey This is also super thorny. 0:19:48.822 –> 0:19:50.392 Brandon Dey It’s it’s a quite common. 0:19:51.242 –> 0:19:56.472 Brandon Dey In fact, I’ve never worked with the product catalog that didn’t have, umm, any of these. 0:19:57.332 –> 0:19:58.682 Brandon Dey Sort of unwanted patterns. 0:19:59.872 –> 0:20:3.362 Brandon Dey And then lastly, you get lumpy demands kind of mix of erratic and intermittent. 0:20:5.392 –> 0:20:16.642 Brandon Dey And these are all important to address, because the degree to which your data exhibits these patterns will drive so much of the project complexity. 0:20:16.652 –> 0:20:26.902 Brandon Dey So much of the project cost will go into handling each of these demand patterns in a responsible sort of managed way. 0:20:29.792 –> 0:20:34.212 Brandon Dey And the first way that you can sort of manage this is by classifying them right? 0:20:34.222 –> 0:20:35.462 Brandon Dey So you got a bunch of input. 0:20:35.572 –> 0:20:49.802 Brandon Dey You’ve got your historical sales or historical demand information and now you need to classify what product belongs to what kind of demand over what time frame and you can do that through through something called coefficient of variation. 0:20:51.532 –> 0:20:59.272 Brandon Dey And and Adi, these are just technical ways of sort of creating that, that taxonomy excuse me. 0:21:2.52 –> 0:21:7.422 Brandon Dey I should say to you there, there is a very rich literature on different ways of classifying demand. 0:21:7.492 –> 0:21:14.352 Brandon Dey This is a very basic way, but it’s it’s often very effective, so no need to you know if it ain’t broke, don’t fix it. 0:21:17.522 –> 0:21:24.172 Brandon Dey OK, so this is a depiction of what you do if. 0:21:27.12 –> 0:21:28.52 Brandon Dey Let me back up a step here. 0:21:32.392 –> 0:21:53.272 Brandon Dey OK, so on this graph on the right, each one of those dots you is a skew that you want to predict or a yeah skew you wanna predict right there are you know 12 or 13,000 dots on this picture notice and it’s been the the the quadrants here represent the demand pattern, right? 0:21:53.422 –> 0:21:59.112 Brandon Dey Notice how tiny smooth is down here and demand is huge. 0:21:59.202 –> 0:22:12.972 Brandon Dey You’ve got lumping erratic now what’s happening here is you take 13,000 products and you say, well of those, how many can I predict very reliably because they’re demanded very consistently. 0:22:13.502 –> 0:22:23.292 Brandon Dey Well, represents a pretty small subset of your overall product catalog, and also by the way, because of the most stable, they’re probably the least valuable to forecast, right? 0:22:23.352 –> 0:22:31.332 Brandon Dey You can take simple averages and and get a pretty good and and sufficiently good estimate without doing anything fancy. 0:22:32.912 –> 0:22:42.792 Brandon Dey Now, the vast majority of your product catalog is going to be lumpy, intermittent or erratic, and that’s trickier, right? 0:22:43.542 –> 0:22:51.532 Brandon Dey And so you have to do a number of things or you can do a number of things to expand your forecast coverage, right. 0:22:51.542 –> 0:22:56.412 Brandon Dey There are going to be products that you just cannot forecast for because they are too tricky, right? 0:22:56.502 –> 0:22:58.252 Brandon Dey You can think of a new product, right? 0:22:58.262 –> 0:22:59.552 Brandon Dey Something that you haven’t sold before. 0:22:59.562 –> 0:23:2.772 Brandon Dey Maybe it’s a new it’s a new lazy boy. 0:23:2.782 –> 0:23:3.692 Brandon Dey It’s a new couch, right? 0:23:3.702 –> 0:23:4.252 Brandon Dey New chair. 0:23:4.882 –> 0:23:10.152 Brandon Dey Umm well, how do you forecast demand for something that you haven’t had history to learn? 0:23:10.372 –> 0:23:11.632 Brandon Dey Sort of patterns on. 0:23:12.422 –> 0:23:15.92 Brandon Dey Well, one way of of doing that is saying, well, how close. 0:23:15.102 –> 0:23:20.222 Brandon Dey How similar is this new lazy boy to other lazy boys and how well have those lazy boys sold? 0:23:21.902 –> 0:23:26.892 Brandon Dey You can basically draw relationships like that and attempt to forecast it. 0:23:26.902 –> 0:23:32.172 Brandon Dey Now it’s introducing a lot of complexity for a a good. 0:23:32.182 –> 0:23:32.942 Brandon Dey Maybe it’ll work. 0:23:33.272 –> 0:23:37.402 Brandon Dey Umm, but often it does work, but there are diminishing returns, right? 0:23:37.412 –> 0:23:39.172 Brandon Dey And so essentially. 0:23:41.342 –> 0:23:43.852 Brandon Dey This graph shows that tradeoff of. 0:23:44.142 –> 0:23:45.722 Brandon Dey I can forecast everything. 0:23:47.702 –> 0:23:48.452 Brandon Dey Or sorry I can. 0:23:48.502 –> 0:23:48.892 Brandon Dey I can. 0:23:48.902 –> 0:24:1.392 Brandon Dey Yeah, I can forecast everything, but the quality on average would be pretty low or I can forecast a, you know, very small or a smaller subset of the product catalog, but have a very high accuracy, right. 0:24:2.342 –> 0:24:3.762 Brandon Dey And so there are tradeoffs here. 0:24:4.652 –> 0:24:15.992 Brandon Dey Each of these is zones represents a different technique you can do to expand the fraction of your product catalog that you’re forecasting for right. 0:24:16.2 –> 0:24:19.762 Brandon Dey So fast moving products, attribute expansion and similarity expansion. 0:24:21.182 –> 0:24:22.682 Brandon Dey These are essentially. 0:24:24.822 –> 0:24:29.282 Brandon Dey The variance of what I mentioned before with hey, this is a new product. 0:24:29.292 –> 0:24:30.482 Brandon Dey I haven’t sold it before. 0:24:30.552 –> 0:24:39.962 Brandon Dey Draw relationship between it and existing products and then use the history of those existing products or forecasts of those existing products to forecast this new thing. 0:24:40.792 –> 0:24:44.602 Brandon Dey The difference here is you’re either doing it for similar attributes. 0:24:45.282 –> 0:24:45.912 Brandon Dey Umm. 0:24:46.502 –> 0:24:50.282 Brandon Dey Or more generally, similarity across everything. 0:24:57.922 –> 0:24:59.862 Brandon Dey Alright, so just to kind of break this down. 0:25:2.762 –> 0:25:14.112 Brandon Dey This is a fairly technical side, so I think I might just kind of cruise through it, but the terminology here is is kind of important as you get into the weeds and so imagine. 0:25:17.352 –> 0:25:18.992 Brandon Dey The most obvious thing, right? 0:25:19.212 –> 0:25:23.382 Brandon Dey If we’re right here, this Gray line, this is our the time of forecasts, right? 0:25:23.392 –> 0:25:32.312 Brandon Dey Our forecasting time and we’ve got this whole wiggly line called our forecast horizon and we make a forecast and that’s our domain forecast suite. 0:25:33.782 –> 0:25:53.322 Brandon Dey You know, so if you look in the review mirror, we’ve got this look back frizon umm, not really forms the basis of of what we’re doing now at the time of forecasts, we’re basically taking all of the data available to us that drives demand and we’re piping it into our model and it’s producing the forecasts suite. 0:25:54.562 –> 0:26:11.652 Brandon Dey It gets more complicated though, because you can pull those data points that you’re using to make your forecast from different points in time in the past and at different levels of granularity depending on what kind of feature it is. 0:26:11.662 –> 0:26:22.742 Brandon Dey And so you can think of this as essentially taking, umm, you know, historical values that do affect sort of future values. 0:26:23.162 –> 0:26:28.532 Brandon Dey And just figuring out how long back you need to look to to find something useful. 0:26:34.132 –> 0:26:46.332 Brandon Dey OK, so if we were to put all of that into a a bad slide with two small of text, it would look like this, it’s essentially an end to end flow of one of these solutions, right? 0:26:46.342 –> 0:26:50.532 Brandon Dey So at the top you basically have stuff you wanna forecast. 0:26:50.542 –> 0:26:53.712 Brandon Dey At the bottom you get all of the value from it, right? 0:26:54.72 –> 0:27:5.492 Brandon Dey You know you got your inventory planning, use case, market response analysis, UMM, SCENARIOS evaluation etcetera and let’s say, yeah, you wanna pipe it all into a pricing optimization platform suite. 0:27:5.502 –> 0:27:12.452 Brandon Dey Well, in between you have all this stuff that you might go through to convert the data into something something useful, right? 0:27:12.462 –> 0:27:18.372 Brandon Dey So the first would be obviously finding those items that you want to forecast that are actually active right now. 0:27:19.182 –> 0:27:31.422 Brandon Dey Not everything that you have sold, you still sell, so obviously important to find the active ones if they are inactive, you know put them off to the side. 0:27:32.222 –> 0:27:36.572 Brandon Dey If they are active though, you wanna first classify that demand, right? 0:27:36.582 –> 0:27:40.132 Brandon Dey You put it into this first blue box and you figure out what’s intermittent. 0:27:40.142 –> 0:27:41.22 Brandon Dey What’s smooth and so on. 0:27:42.832 –> 0:27:47.982 Brandon Dey Then you’re going to basically take different actions depending on what kind of demand it is, right? 0:27:47.992 –> 0:27:57.672 Brandon Dey And so if it’s lumpy, you’re gonna, you’re gonna put those items into a specialized model to handle the lumpy stuff if it’s not lumpy, then you’re going to. 0:27:59.792 –> 0:28:4.642 Brandon Dey Essentially, dump it into the next step and see if you can. 0:28:8.42 –> 0:28:41.582 Brandon Dey Basically, if the intermittent and erratic stuff is sufficiently similar from a product standpoint to treat them the same or or differently, umm, if they’re different, you know you put the subtly different ones over to the right in the same bucket you you move the inactive ones and then for those that remain you do what’s called feature engineering, which is essentially connecting with your business stakeholders and understanding what might plausibly Dr demand for various things and and building a data set around them. 0:28:41.692 –> 0:28:47.612 Brandon Dey And then move sort of piping them into the model and you finally get your model and outcomes. 0:28:48.12 –> 0:28:48.562 Brandon Dey Useful stuff. 0:28:50.212 –> 0:28:51.542 Brandon Dey Alright, that’s super high level. 0:28:55.42 –> 0:29:1.722 Brandon Dey Now I said I want you guys to to leave not thinking about build or buy. 0:29:2.332 –> 0:29:7.982 Brandon Dey Instead, I think it’s more effective to think about what to build and what to buy. 0:29:7.992 –> 0:29:13.62 Brandon Dey Not everything is created equal, and it’s important to know what’s what, especially for really. 0:29:13.72 –> 0:29:18.442 Brandon Dey For all technology solutions, but, but especially AI technology solutions like demand planning. 0:29:19.432 –> 0:29:27.692 Brandon Dey And So what I like to do with our customers is work through a map. 0:29:28.852 –> 0:29:31.942 Brandon Dey Umm, that’s shown here that can help them. 0:29:31.952 –> 0:29:43.42 Brandon Dey Reason about which components under the hood they should be buying or they they should be building and if they do build them, should they build them in house or with the an external partner? 0:29:43.572 –> 0:29:46.502 Brandon Dey Now this map kind of has a lot going on, but. 0:29:49.372 –> 0:29:55.562 Brandon Dey At the top, you start by pointing out the business value. 0:29:55.572 –> 0:29:59.772 Brandon Dey The reason behind all this technology, right? 0:29:59.912 –> 0:30:2.732 Brandon Dey Well, demand planning, this is for healthy working capital, right? 0:30:2.742 –> 0:30:9.52 Brandon Dey We set at the top of the call that demand planning is to make more money right by reducing overbuy or under buy costs. 0:30:9.62 –> 0:30:20.312 Brandon Dey That’s all about healthy working capital in order and then basically everything below that is a component required to make it happen, right. 0:30:20.812 –> 0:30:24.32 Brandon Dey And then you basically position each of those components. 0:30:24.642 –> 0:30:25.252 Brandon Dey Umm. 0:30:25.782 –> 0:30:34.852 Brandon Dey On on the graph where you’re mapping that components position and that components movement, right position and movement. 0:30:34.862 –> 0:30:41.632 Brandon Dey Closely related, very important to talk through basically the vertical axis is the that that value chain. 0:30:42.342 –> 0:30:52.572 Brandon Dey So the higher up is the more valuable it is to users and the business and the horizontal axis is the maturity of the component, right in terms of technology, right? 0:30:52.582 –> 0:30:57.382 Brandon Dey So on the on the far left, you’d have like brand new ideas. 0:30:57.392 –> 0:30:59.162 Brandon Dey Nobody’s kicked tires on it yet. 0:30:59.272 –> 0:31:1.992 Brandon Dey That’s your POC kind of stuff on the very far right. 0:31:2.172 –> 0:31:3.272 Brandon Dey These are utilities, right? 0:31:3.282 –> 0:31:5.122 Brandon Dey This is like electricity, right? 0:31:5.162 –> 0:31:9.812 Brandon Dey Or electricity is like very, very much utility and nobody ever thinks about utility. 0:31:9.822 –> 0:31:39.772 Brandon Dey So it’s all the way down here now, going back up to healthy working capital in order for that to happen, you need to decrease your carrying costs in order to do that, you need better inventory turns and in order for that to happen, first of all, you need something to track our ROI and you need a crud app to change inventory levels right in order to change inventory levels in the ERP integration, which needs a platform which needs a compute which needs energy and so on. 0:31:40.282 –> 0:31:43.472 Brandon Dey Now I’m going to point out a few different things, right. 0:31:43.482 –> 0:31:47.572 Brandon Dey So all of this in red is this stuff that’s not AI related, right? 0:31:47.792 –> 0:31:52.752 Brandon Dey A huge portions of this system that you do not need a data scientist for, right. 0:31:52.762 –> 0:31:53.612 Brandon Dey Nor should you hire one. 0:31:54.472 –> 0:31:56.602 Brandon Dey It’s very sort of classic. 0:31:57.522 –> 0:32:0.172 Brandon Dey Umm, so crud, that’s a great question. 0:32:0.242 –> 0:32:8.712 Brandon Dey Stands for create, read, update and delete any software basically that lets you interact with it is a CRUD app. 0:32:9.622 –> 0:32:10.932 Brandon Dey It’s got an unfortunate acronym. 0:32:12.432 –> 0:32:14.722 Brandon Dey But that’s that’s what it is. 0:32:15.272 –> 0:32:15.902 Brandon Dey Great question. 0:32:15.912 –> 0:32:22.962 Brandon Dey So the stuff in red is the AI disabled portions ROI tracking basically UI to change the Inventory. 0:32:23.412 –> 0:32:26.432 Brandon Dey An ERP integration which needs business logic. 0:32:27.372 –> 0:32:31.932 Brandon Dey The ERP integration obviously needs a platform that platform needs. 0:32:31.942 –> 0:32:34.172 Brandon Dey ETL to get some historical data. 0:32:35.522 –> 0:32:37.932 Brandon Dey And that’s, you know, obviously all being computed. 0:32:38.742 –> 0:32:44.852 Brandon Dey Umm, alright, so this is important to talk about but it’s the non not as interesting stuff. 0:32:45.102 –> 0:32:46.762 Brandon Dey This stuff in red is the. 0:32:48.692 –> 0:32:52.102 Brandon Dey Stuff related to the to the AI demand planning portion of the system. 0:32:52.112 –> 0:32:52.322 Brandon Dey Right. 0:32:52.332 –> 0:33:10.652 Brandon Dey And so obviously what’s very visible and often not mature at all, that you need for demand planning, those are the forecasts themselves, right, not the forecast model, but the output right, in order to for that output to be useful, it needs to be explainable. 0:33:10.662 –> 0:33:13.192 Brandon Dey This is its own section really. 0:33:13.202 –> 0:33:17.112 Brandon Dey In order for those forecasts to be produced, you do need some business logic. 0:33:18.62 –> 0:33:20.532 Brandon Dey You also need the actual forecasting model. 0:33:20.592 –> 0:33:24.742 Brandon Dey The forecasting model then needs a few different kinds of data points, right? 0:33:24.752 –> 0:33:33.712 Brandon Dey Those four really that I talked about before, which would be product, sorry, taxonomy data, historical demand, data, price data, all of those things needs. 0:33:33.942 –> 0:33:37.452 Brandon Dey ETL and an ML OPS platform. 0:33:37.462 –> 0:33:40.102 Brandon Dey Write something to make this automated and and while governed. 0:33:40.592 –> 0:33:42.682 Brandon Dey And both of those things need a platform. 0:33:45.112 –> 0:33:46.392 Brandon Dey Now there are two other. 0:33:48.232 –> 0:33:51.762 Brandon Dey Sources of data up here there’s weather data and calendar data. 0:33:52.152 –> 0:33:58.192 Brandon Dey Notice how each of these data sources and each of these components they’re in different they’re in a different position in space. 0:33:58.952 –> 0:33:59.442 Brandon Dey Umm. 0:33:59.832 –> 0:34:5.772 Brandon Dey And this is really important because it’s gonna make or break if you buy them or build them. 0:34:6.742 –> 0:34:9.802 Brandon Dey And so kind of to help illustrate this, I put some colors here. 0:34:9.812 –> 0:34:15.932 Brandon Dey Basically just to highlight the the different zones of tech maturity. 0:34:16.892 –> 0:34:24.52 Brandon Dey So, uh, you know the brand new stuff called Genesis, then you’ve got sort of this custom built lane. 0:34:24.62 –> 0:34:29.132 Brandon Dey Then you move into more configured product world and then finally you’re in utility space. 0:34:30.962 –> 0:34:31.312 Brandon Dey All right. 0:34:31.322 –> 0:34:39.732 Brandon Dey So you you kind of, you wanna every organization is different and I’ll kind of get into how to think through some of this or how I’ve found it useful to think through some of this. 0:34:40.172 –> 0:34:40.702 Brandon Dey Umm. 0:34:41.52 –> 0:34:47.662 Brandon Dey But typically, you’d want to build the Genesis and the custom built stuff right? 0:34:47.832 –> 0:34:51.932 Brandon Dey That is sufficiently sort of high value in the value chain. 0:34:53.782 –> 0:34:57.772 Brandon Dey There are components though that may not be worth building, right? 0:34:57.892 –> 0:35:11.832 Brandon Dey And those are the things that you might want to configure right there are there are components that aren’t totally off the shelf that you can buy, but maybe they’re open source and you can stand on the shoulders of giants and and just kind of configure them, saves you a bunch of time and money. 0:35:12.122 –> 0:35:19.592 Brandon Dey And then lastly, you’re in utility space where it is not worth anybody’s time to configure. 0:35:19.602 –> 0:35:20.552 Brandon Dey Build this stuff right? 0:35:20.562 –> 0:35:23.632 Brandon Dey So just buying off the shelf, that’s fine, right? 0:35:23.642 –> 0:35:28.182 Brandon Dey Like nobody’s gonna go out and build their own generators to produce their own electricity, for example. 0:35:28.192 –> 0:35:29.752 Brandon Dey Right, that does makes no sense. 0:35:30.452 –> 0:35:31.582 Brandon Dey Same with compute. 0:35:31.852 –> 0:35:33.122 Brandon Dey Calendar data is basically the same. 0:35:34.972 –> 0:35:35.282 Brandon Dey OK. 0:35:38.602 –> 0:35:40.852 Brandon Dey So this maps out. 0:35:42.762 –> 0:35:43.382 Brandon Dey Really. 0:35:43.432 –> 0:36:0.122 Brandon Dey What you what you should buy, configure build but it doesn’t really say much about who should do that, and it feels like a basic question, but there aren’t a lot of good resources that I found to help you walk through. 0:36:0.132 –> 0:36:11.592 Brandon Dey And so I made this decision tree to really map how I think about and how I guide customers on their build versus buy. 0:36:12.342 –> 0:36:12.902 Brandon Dey Umm. 0:36:13.82 –> 0:36:13.692 Brandon Dey Journey. 0:36:13.702 –> 0:36:16.392 Brandon Dey And so the first preference. 0:36:18.272 –> 0:36:20.72 Brandon Dey Is buying off the shelf right? 0:36:20.752 –> 0:36:27.122 Brandon Dey And there are exceptions to this, but a good starting default would be buy things off the shelf if certain things are true. 0:36:27.352 –> 0:36:30.552 Brandon Dey The second preference would be to build internally, right? 0:36:30.632 –> 0:36:33.982 Brandon Dey It’s good to own stuff, especially the important stuff. 0:36:35.892 –> 0:36:37.922 Brandon Dey The last preference would be to build with a partner. 0:36:38.452 –> 0:36:57.312 Brandon Dey Now each of these things happens under different scenarios, and so the first question that’s important to start with that helps inform which component should be built by or bought by which party is, is this component a competitive advantage? Right? 0:36:58.182 –> 0:37:5.252 Brandon Dey If it’s not, and then the next question is, does it need customization or flexibility now? 0:37:5.262 –> 0:37:7.382 Brandon Dey If it doesn’t buy it off the shelf. 0:37:7.432 –> 0:37:7.692 Brandon Dey Right. 0:37:7.702 –> 0:37:8.362 Brandon Dey Not worth it otherwise. 0:37:10.12 –> 0:37:21.882 Brandon Dey In the event that it needs customization and flexibility, umm, the next question here is does it impact really strategic stuff or the customer, right? 0:37:22.882 –> 0:37:27.432 Brandon Dey If it doesn’t, but the shelf if it does, the next question is we’ll show, can I build it in house? 0:37:28.582 –> 0:37:39.72 Brandon Dey And that’s a question of do you have the people with the skill set to to build the thing if you don’t build with a partner now if you can build it in house, there are cases where you. 0:37:41.202 –> 0:37:41.692 Brandon Dey Were you? 0:37:41.702 –> 0:37:48.292 Brandon Dey You were you build internally, but there’s also a a a point in time where just because you can doesn’t mean you should, and we’ll get to those. 0:37:49.142 –> 0:37:51.972 Brandon Dey So the next question is basically do you need it fast, right? 0:37:51.982 –> 0:37:52.952 Brandon Dey Do you need it really fast? 0:37:52.962 –> 0:37:59.2 Brandon Dey Time to market where you can’t afford really to have folks kind of like learning how it’s done. 0:38:1.782 –> 0:38:4.652 Brandon Dey If you don’t need it fast, you can. 0:38:4.662 –> 0:38:8.692 Brandon Dey You can choose to build internally, and that’s not to say it’s definitely going to be slow, of course. 0:38:11.652 –> 0:38:27.82 Brandon Dey If you do need it quickly and you prefer just sort of trade the long term costs with investment up front and you have a culture of innovation, build, build with a partner, right? 0:38:27.272 –> 0:38:32.112 Brandon Dey If you don’t have a culture of innovation but those other things are true, then you can buy off the shelf. 0:38:34.72 –> 0:38:46.422 Brandon Dey So this is the a set of heuristics that I found to be useful for organizations who are still thinking through like what to build and and what to buy. 0:38:48.392 –> 0:38:50.492 Brandon Dey It’s not right all the time, but it’s a it’s a pretty good start. 0:38:55.372 –> 0:38:56.962 Brandon Dey I just checking in the other questions. 0:38:57.312 –> 0:39:1.142 Brandon Dey Alright, so some challenges with demand planning systems. 0:39:1.312 –> 0:39:2.572 Brandon Dey It is not without challenges. 0:39:4.362 –> 0:39:4.752 Brandon Dey Umm. 0:39:6.502 –> 0:39:12.612 Brandon Dey Click this so first of all out of stock events do bias demand data right? 0:39:12.622 –> 0:39:15.112 Brandon Dey So if you can’t observe, if you don’t observe. 0:39:17.92 –> 0:39:17.462 Brandon Dey Umm. 0:39:18.442 –> 0:39:26.352 Brandon Dey If you don’t have the opportunity to see how many things would have been purchased because it wasn’t available, we’re not going to know how to treat. 0:39:26.642 –> 0:39:31.502 Brandon Dey Treat that right like just cause you didn’t have demand doesn’t mean, umm, there wouldn’t have been. 0:39:31.702 –> 0:39:36.532 Brandon Dey And so that’s a that’s a tricky thing that needs to be addressed with some fancy statistics. 0:39:37.442 –> 0:39:41.572 Brandon Dey Umm, the next thing is really reconciliation of the hierarchy, right. 0:39:41.582 –> 0:39:43.372 Brandon Dey You’ve got a product catalog. 0:39:43.422 –> 0:39:53.222 Brandon Dey You’ve got things that belong to different categories and subcategories, and if you forecast that the skew level right, I need fourteen of these couches and seven of those. 0:39:55.322 –> 0:39:56.612 Brandon Dey They need to roll up. 0:39:56.622 –> 0:40:5.632 Brandon Dey They need to be able to be aggregated to that category or subcategory or whatever your preferred level of abstraction is in a way that’s compatible with like. 0:40:7.512 –> 0:40:9.322 Brandon Dey Your your constraints basically right. 0:40:9.372 –> 0:40:27.632 Brandon Dey Sometimes you can forecast at the category level, and it’s often something that we do and then you have to spread that group level forecast down to specific SKUs and you only need to do that in cases where the skew itself doesn’t have sufficient signal or volume to weren’t just really warrant forecasting it on its own. 0:40:28.672 –> 0:40:31.252 Brandon Dey But anyways, this is a a tricky thing that needs to be dealt with. 0:40:32.182 –> 0:40:34.612 Brandon Dey Now the challenge is what we talked about before with. 0:40:36.452 –> 0:40:37.742 Brandon Dey Limited catalog coverage, right? 0:40:37.752 –> 0:40:38.462 Brandon Dey You don’t have enough? 0:40:39.592 –> 0:40:47.632 Brandon Dey Umm, high quality signal to forecast for a sufficiently, you know, large fraction of your product catalog. 0:40:47.702 –> 0:40:48.402 Brandon Dey And So what do you do? 0:40:51.182 –> 0:40:53.982 Brandon Dey There’s the forecasting of new products, right? 0:40:53.992 –> 0:40:55.452 Brandon Dey This is called the cold start problem. 0:40:55.852 –> 0:41:3.922 Brandon Dey I mentioned it before, there is a lot of things that you can do to to handle that and umm and it’s it’s deeply technical. 0:41:3.932 –> 0:41:5.92 Brandon Dey So it didn’t didn’t wanna get into it? 0:41:5.882 –> 0:41:9.92 Brandon Dey Umm, another challenge is, is data related? 0:41:9.102 –> 0:41:9.672 Brandon Dey Of course, right. 0:41:9.682 –> 0:41:13.212 Brandon Dey So quality data is is a challenge here. 0:41:13.622 –> 0:41:17.742 Brandon Dey Typically, most often with price and for a variety of reasons. 0:41:17.752 –> 0:41:17.962 Brandon Dey Right. 0:41:17.972 –> 0:41:33.822 Brandon Dey Pricing is often one of the most complicated, very Custom activities that an organization goes through and it results in Umm, suspect data that you may want to think better of if you include it into into your data forecast models. 0:41:35.752 –> 0:41:38.872 Brandon Dey Umm, it’s often it’s related to price. 0:41:39.102 –> 0:41:46.822 Brandon Dey Sometimes you don’t have uh price variability, especially if we’re if you’re B to B organization, right? 0:41:46.832 –> 0:41:57.472 Brandon Dey So by price variability, I mean you don’t have cases where you sold a skew for price X and price X minus. 0:41:57.742 –> 0:41:58.912 Brandon Dey You know why, right? 0:41:58.922 –> 0:42:5.992 Brandon Dey You you didn’t sell the same thing at different prices and so you can’t draw relationship between changes in price and changes in volume. 0:42:6.322 –> 0:42:10.412 Brandon Dey You just see changes in volume, but you have the same price, so not much you can do about that. 0:42:11.592 –> 0:42:14.972 Brandon Dey Umm, now the challenge is there. 0:42:16.22 –> 0:42:26.782 Brandon Dey This is really a challenge from my point of view as a third party coming into another organization and seeing what what could be done to create value. 0:42:26.982 –> 0:42:35.892 Brandon Dey Low adoption of price optimization price optimization is is is, not surprisingly, super important, right it it drives really fundamental things of your business. 0:42:36.562 –> 0:42:41.392 Brandon Dey And despite this, there are there’s often low adoption of this very, very valuable technology. 0:42:43.262 –> 0:42:45.82 Brandon Dey All right, so a little clickbaity title. 0:42:45.92 –> 0:42:45.422 Brandon Dey Mine. 0:42:45.432 –> 0:42:47.232 Brandon Dey Uh, some stuff you should know. 0:42:48.282 –> 0:42:57.892 Brandon Dey Often folks want to know up front how much of their product catalog they can forecast, right for reasons I talked about here. 0:42:57.942 –> 0:42:59.752 Brandon Dey You we can’t know that up front. 0:42:59.842 –> 0:43:5.972 Brandon Dey It’s impossible to know how many of your products can be forecasted before we forecast them. 0:43:6.782 –> 0:43:12.212 Brandon Dey We can make reasonable judgments about it, but really this is the last thing that we can conclude. 0:43:13.922 –> 0:43:21.882 Brandon Dey And so there are certain heuristics that that we can go through to to help make an informed and conservative sort of judgment on this. 0:43:22.992 –> 0:43:24.412 Brandon Dey But empirically, you can’t know until the end. 0:43:26.412 –> 0:43:26.662 Brandon Dey Now. 0:43:29.142 –> 0:43:35.12 Brandon Dey Similar to what I mentioned, with that use case with the the furniture manufacturer. 0:43:37.752 –> 0:43:42.352 Brandon Dey Demand planning forecasts are super easy to veto by people. 0:43:44.282 –> 0:43:50.322 Brandon Dey It’s demand planning tools are often not built to be autonomous decision making tools. 0:43:50.332 –> 0:44:5.222 Brandon Dey There are decision support tools and so it relies on a human to read its output and take action on it, and for reasons related to poor implementations or distrust, often people ignore them. 0:44:8.252 –> 0:44:9.642 Brandon Dey So it does require change management. 0:44:9.972 –> 0:44:11.942 Brandon Dey Super important some projects don’t need it. 0:44:11.952 –> 0:44:16.562 Brandon Dey Some projects do demand planning is one where definitely does cause it’s a. 0:44:16.612 –> 0:44:18.782 Brandon Dey It’s definitely a people, people product. 0:44:20.652 –> 0:44:22.772 Brandon Dey It also demand planning is often neglected. 0:44:24.322 –> 0:44:25.802 Brandon Dey It’s not sexy. 0:44:25.812 –> 0:44:29.232 Brandon Dey It’s been around forever and I think that’s unfortunate. 0:44:30.242 –> 0:44:35.802 Brandon Dey Also what happens is these projects can fail even if it’s not a technical thing. 0:44:36.242 –> 0:44:45.402 Brandon Dey If they lack a champion and by champion, I don’t mean just a sponsor, but I do mean somebody who is gonna go to the mat and really evangelize for this technology. 0:44:46.132 –> 0:44:46.962 Brandon Dey That’s super important. 0:44:49.932 –> 0:44:51.982 Brandon Dey Also, it’s not a panacea solution. 0:44:52.502 –> 0:44:59.782 Brandon Dey You can Demand plan is only worth it if the future is expected to look like the past, right? 0:44:59.932 –> 0:45:6.902 Brandon Dey Demand planning is a tool that learns the past so that it can predict the future that only works if they’re expected to be the same. 0:45:7.12 –> 0:45:11.192 Brandon Dey So if your business isn’t stable, forecasting is not not great. 0:45:14.32 –> 0:45:18.912 Brandon Dey OK, so stuff that affects the scope of these projects, there’s a lot. 0:45:18.922 –> 0:45:23.132 Brandon Dey So really, the state of the data, how complex is the data processing? 0:45:24.162 –> 0:45:25.972 Brandon Dey What does the pricing data look like? 0:45:26.262 –> 0:45:26.802 Brandon Dey Where? 0:45:26.812 –> 0:45:27.752 Brandon Dey Where is it all at? 0:45:27.762 –> 0:45:30.882 Brandon Dey Does it require a lot of, you know, integrations and so on? 0:45:32.322 –> 0:45:37.2 Brandon Dey Umm, another driver here is basically the demand patterns, right? 0:45:37.792 –> 0:45:40.832 Brandon Dey How lumpy and erratic and intermittent is this? 0:45:43.152 –> 0:45:51.492 Brandon Dey This can be backed into relatively easy just by looking at a single data source, often, which is the historical demand. 0:45:55.342 –> 0:46:3.962 Brandon Dey Another thing that’s tricky is deciding if and how you are going to address that cold start problem and slow moving items problem right? 0:46:3.972 –> 0:46:8.662 Brandon Dey So the cold start is that you have a brand new product for the first time and you wanna sell it. 0:46:8.832 –> 0:46:12.422 Brandon Dey How do you forecast something you haven’t sold before and slow moving items? 0:46:12.432 –> 0:46:15.462 Brandon Dey These are related to lumpy, bumpy products. 0:46:15.732 –> 0:46:23.422 Brandon Dey The the more technically you try to address these problems, the wider the scope. 0:46:25.382 –> 0:46:33.72 Brandon Dey Umm, another thing is how human readable do you want or need the output? 0:46:33.922 –> 0:46:36.732 Brandon Dey If you don’t need it human readable, then that’s less work, right? 0:46:36.742 –> 0:46:39.12 Brandon Dey If you need a very human readable that’s more work. 0:46:39.942 –> 0:46:43.452 Brandon Dey The good news is there’s lots of things that can be. 0:46:45.372 –> 0:46:50.462 Brandon Dey Basically built and recycled to to facilitate this effort. Umm. 0:46:51.962 –> 0:47:0.492 Brandon Dey And then another thing, I think that’s important is uh for organizations to think about when they undertake one of these projects is how much do you wanna learn about your business? 0:47:0.502 –> 0:47:4.682 Brandon Dey How much do you want to kind of roam and explore different things that you might find interesting? 0:47:6.422 –> 0:47:14.352 Brandon Dey This is an often and understandably overlooked section of these projects, but one of the higher leverage activities. 0:47:16.642 –> 0:47:18.632 Brandon Dey Alright, So what comes next? 0:47:20.22 –> 0:47:20.942 Brandon Dey We are. 0:47:20.952 –> 0:47:28.92 Brandon Dey We’re offering a an AI in visioning session with with myself and our CTO to do an AI envisioning session right? 0:47:28.102 –> 0:47:35.922 Brandon Dey So this is zooming out of just demand planning and it’s aligning the organization on strategic objectives. 0:47:37.162 –> 0:47:39.2 Brandon Dey Fielding different pain points. 0:47:39.12 –> 0:47:47.552 Brandon Dey Different opportunities to improve and sharing the state of what could be possible with various AI solutions. 0:47:47.682 –> 0:47:56.252 Brandon Dey This has been super popular and it’s a very, very good high leverage activity to get started with AI stuff for brand new. 0:47:56.702 –> 0:48:2.522 Brandon Dey Also, we’re doing complementary to our deep dive sessions into demand planning. 0:48:2.532 –> 0:48:15.512 Brandon Dey In particular, if what you learned today was was interesting, or if you have specific questions that could be something, check out and then lastly, umm, we are offering a complimentary Microsoft funding review. 0:48:15.522 –> 0:48:23.662 Brandon Dey So you know what kind of funding from Microsoft would be available should you decide to go down this this path? 0:48:24.992 –> 0:48:27.502 Brandon Dey So good next steps. 0:48:28.432 –> 0:48:32.322 Brandon Dey Also, I should say you can reach out to me if you’re interested in chatting. 0:48:33.132 –> 0:48:35.842 Brandon Dey Otherwise it will be, umm, a survey. 0:48:35.852 –> 0:48:41.602 Brandon Dey After this, I believe Amy to if that’s right to to kind of capture your information. 0:48:40.522 –> 0:48:43.52 Amy Cousland Yeah, I put a survey and I put a link to a survey. 0:48:43.62 –> 0:48:45.732 Amy Cousland If you’re interested in these, if you want to just fill that out, we’ll be in touch. 0:48:47.822 –> 0:48:48.122 Brandon Dey Yes. 0:48:48.132 –> 0:48:50.192 Brandon Dey And with that, I want to thank everybody for your time. 0:48:50.202 –> 0:48:57.92 Brandon Dey Umm, appreciate you being generous with it on a Friday junior and I hope to see you around. Thanks.