Supply Chain Expert Video Series
GAINS ON is a video series featuring GAINS Co-Founder Bill Benton sharing his supply chain insights based on decades of industry leadership in supply chain planning. GAINS ON allows viewers to learn about the latest advances, processes, and insights that can make your supply chain planning function more impactful and effective. It’s a masterclass for anyone wanting to learn about supply chain planning.
The on-demand video series includes insights on topics including:
- Demand Driven Materials Resource Planning (DDMRP)
- Inventory Optimization
- Lead Time Sensing
- MRO & Service Parts Planning
- And more
Check back often for new videos to help you master the art of supply chain planning, or follow GAINS on LinkedIn to see the latest GAINS ON videos.
About Bill Benton:
Bill Benton, widely recognized as a visionary in the supply chain industry, is known for rolling up his sleeves and solving complex supply chain challenges. He is passionate about bringing practical innovation to supply chain planning using ML and AI. Bill has received several awards for his professional achievements and is a frequent speaker at the Gartner Supply Chain Symposium and other industry events.
MRO & Service Parts Planning
In this episode of GAINS On, join GAINS Co-founder Bill Benton as he describes the complexities of MRO and Service Parts planning, and how GAINS uses AI tools to manage this complexity.
(00:02): Hi, my name's Bill Benton, I'm the co-founder of GAINSystems. Today I'll be talking about MRO and service parts planning in this version of GAINS On.
What multi-echelon refers to is the multiple tiers in a, either distribution network or in a fleet support network. So you might have a global hub, where virtually everything's kept, regional DCs, you could have stores or branches if you're doing fleet service or operating bases, trucks or remote sites. So managing how broadly and deeply you wanna deploy inventory, what's optimum, where you wanna expedite, how critical is the part, how high margin is it? These all come into play in multi-echelon inventory optimization.
(01:12): For those people dealing with repair or rotables, you have the 360 degree supply chain. What you send out comes back in and it may or may not be repairable. Managing around mobile fleets of serviced machines, selling parts to end customers. Those customers are moving the assets around often, and the demand will follow those. You may not always be made aware of that. There can be a large number of locations that are widely dispersed. As those expand, it not only multiplies the number of SKU locations, but also the question of what's the right next to stock at each site. Dealing with an extremely large number of parts or SKUs, that sheer volume adds a lot of complexity. And moreover, those often have deeper vision change, where you might have four engineering redesigns of the same part across its lifespan.
(02:11): Unlike many other environments, usually there's gonna be multiple parts needed to fulfill a work order or repair order, or even just a sales order because the end user needs multiple parts to perform their job. How do you manage that mix at each and every site given the frequency of demand, given the proximity of supply, and then compound that complexity with the need to look potentially across conditions of the market, whether it's a core that it's in need of repair, it's unserviceable, it's repaired, or it's overhauled, and different customers might have different requirements for different conditions. So the optimization methodology has to consider that and try to get a near optimum mix of parts and conditions across parts where applicable. You need to do this in a financially feasible manner. So whether that's according to a budget for an internally sourced fleet. according to profit goals if you're selling the parts, or potentially just according to inventory turnover and other targets. So how do you reduce surplus of active inventory and also reduce or minimize obsolescence and obsolete part write-offs?
(03:29): In most cases, there's extremely high downtime, expedite, or lost sales risks. Often the need is urgent. And if it's not fulfilled in the most cost effective way high cost remediation is often needed. You're often dealing with both long and highly variable supply lead times. These can be very challenging, like variable, particularly with repair, where there's uncertainty about the turnaround time of the repair and the complexity, the availability of spare parts to perform that repair, et cetera. When you're dealing with an OEM situation where you're trying to provide parts to the field and end customers and your sister business unit is using those same parts producing new equipment, you've got competition for that supply. So one of the challenging things there is balancing and optimizing that.
(04:32): GAINS provides blended methods. Look not just at the historical demand or usage, but blend things in including fleet service information. How many units are in the field that service from this region that need to be managed? How is that fleet growing, shrinking, aging? How does that affect demand? And how can we build profiles around that? GAINS uses AI techniques like nearest neighbor, and others to say that this new part is very likely similar to these other parts where we've had the chance to observe, if not their full life cycle of demand, at least several cycles. How do you manage demand across different conditions? So GAINS has tools to both look at demand by condition, look at the ability to say substitute across those, and then plan for that which is optimal. We use AI tools to manage this complexity because there is no single solution, you can just get near optimum. We also combine that with bootstrap methods to determine what is the safety or what we call service doc needed for each part, particularly when you get into lumpy demand, which really breaks most traditional safety stock methodologies.
(00:00): Hi, my name's Bill Benton, I’m the Co-founder of GAINSystems. Today I'll be talking about MRO and service parts planning in this version of GAINS On.
(00:30): Each decision that GAINS doesn't automate and presents to the Expert Planner is assessed from two points of view. One is, how much risk does this action pose to our ability to fulfill service and what risk does it pose financially? And so when you look at that, it's looking at our overall goals and how much just one element would contribute to either reducing that risk or avoiding it, and ranks those things accordingly. So if you have low cost, low risk decision, those all get automated. We look at automation of different supply options, not like a traditional buy-from-a-vendor, receive, distribute, sell. In this case, there's a lot of redistribution that's available. You can use fill in alternative supply, you can repair, you can upgrade, you can use alternate parts. All these things are usually a manual decision, which adds a ton of complexity. GAINS will provide an optimal and feasible suggestion so that you need not review those all, although you always have the opportunity.
(01:37): What GAINS does is it records each change of the recommendations GAINS makes, and learns from that to determine what factors are correlated with one of those decisions. And here we have a crawl, walk, run approach because understanding why a decision was made, after say a negative event occurred, and trying to understand what you could have known beforehand or why the decision was made, we look at first saying, all right, what are the things you're likely to reject, right? So that's lower risk cuz you can always end approve it the next day, but that might eliminate 50% of the workload off the bat.
(02:36): Moving from preventive to predictive maintenance after a certain number of hours of use or miles driven, certain components of a machine are gonna be broken down and have a series of parts replaced, without regard to whether those parts might have some meaningful lifespan left. But predictive maintenance does actually measures the likely lifespan of those key parts to do two things: One is prevent scrapping parts that might be good until the next cycle. Secondly, determining those parts that might break before the prevent cycle and avoid an unscheduled maintenance event. In reality, not all parts are gonna have sensor data to do predictive maintenance, but what we can do is really tighten up those reliability estimates so that that estimate can enhance the model in a much more accurate fashion than today.
(03:30): Key performance indicators apply such as inventory turns, line fill rate, part availability, et cetera. But beyond that, there are special metrics that need to both drive the model and be measured and improved upon. First is something similar to First Pass Resolution Rate. They're going to make the repair and they can make that repair upon diagnosis. So when they look at it, geez, I have the parts available, I'm gonna replace parts A, B, and C to not only diagnose what I need but to actually execute that, and that might dramatically reduce the cost of that. Similarly, when you think about shipping parts, so if you're either a distributor or a retailer, splitting shipments is quite expensive. So if you have to send two parcels rather than one, that's sort of the rough equivalent of a first pass resolution, right? Do I have everything I need, in the place I need, at the moment of a request to fulfill that. Turnaround time of repair or T&T is a huge metric and achieving that or reducing that has myriad benefits. Last is looking at outer ground or expedite. Some of these things are highly costly that either erode your budget to fulfill your service goals internally, or dramatically reduce your profit margin. So these are other types of measures of effectiveness that are not only measured by GAINS, but built into the model to help optimize excellence.
(05:13): A high number of SKUs and or SKU locations, that complexity is key. High cost of stockout and/or downtime, or very high amounts of working capital, or parts inventory investment. If either or both of those are the case, usually that's because you're trying to buffer a large amount of uncertainty and/or very high service expectations. And that usually means there's, there's room for improvement without cutting any muscle. So the second aspect we focus on is from an industry perspective, if you're an OEM, you manufacture machines that are used out in the field and you have long-lived machinery and you need to provide spares over that entire lifespan, including the warranty period and end of life. If you’re a spare parts distributor and you're selling spare parts, not necessarily to consumers or retail, but business to business, those are very good fits. And lastly is fleet operators that are servicing essential equipment. So that ranges from airlines to oil field companies and mining equipment operators. These are the two sort of areas where we think we can help. Appreciate your time.
Lead Time Sensing
In these episodes of GAINS On, join GAINS Co-founder Bill Benton as he describes lead time sensing and prediction, empowered by machine learning through GAINS. We know there’s a lot of noise around AI and ML, everything from chatbots to social media ads, but what we focus on here is pragmatic ML and AI that can be operationalized to be resilient across teams and time, and has built-in bidirectional plumbing. So we’re excited to bring this.
(00:05): Hi, my name's Bill Benton. I'm the co-founder here at GAINSystems. I'm excited to talk to you today in this GAINS On segment regarding lead time sensing and prediction, empowered by machine learning through GAINS.
Lead time sensing or prediction is the estimation of the time from which somebody decides they need something to the time they receive it. Generally, these have been maintained poorly and infrequently sometimes set and then rarely touched. What we've learned through the pandemic and following is that these can be highly volatile. And if you were to look at any one parameter driving performance, this might be the single most important one.
(01:23): You set it and forget it, right? And then the user manually adjusts with each order based on their own intuition of what the current status is. So that's bad. Less bad would be that you use some dynamic broad brush parameters like, from this production center, I'm gonna expect this, or this vendor, I'm gonna expect this. When in fact, there's a lot of variance from those, but at least those get updated and if say, a routing from one continent to another through a port, and you can recognize that that's happening faster or slower. Good is using real data, observed data to calculate what the trend in historical lead time has been at a part location, ideally, or if there's insufficient data, slightly higher level like part family or vendor.
(02:21): So let's assume you're already at the good level and you are looking backwards. Well you really miss inflection points that way. So for example, as peak stress in the pandemic was observed, you would've assumed looking at history that the lead times would continue increasing. And once some of those capacity constraints were resolved, you actually saw lead times declining. And that was very hard to see upfront. That surge was not just caused by overestimating demand, but overestimating supply time. So you're buying in advance when it wasn't necessary. So what we've done is we've used machine learning methods and really focusing on what's ready, readily available to predict what the next lead time will be. And that uses data ranging from simple things like how many delayed orders exist from this supply source at that moment, and how does that compare to how many existed before, and how did those things correlate with lead time? Also looks at location. At minimum it looks at over 500 different potential correlates to determine which ones predict the next lead time well, not just measure the last one.
So if you're trying to buffer across a one-week lead time plus or minus four days, that's a very different question than if you're trying to buffer across a one quarter or a half a year long lead time. Merely getting that input correctly defined can reduce the amount of inventory you need significantly, or conversely, ensure that it's sufficient to meet the needs. So if you're understating lead time, that's very likely to lead stockouts and material problems. And if you're overstating it, you're gonna be tying up working capital.
(04:22): Accurate, available to promise: If you're manufacturing, people are asking you when things can be provided. It can help with long-term planning in terms of vendor collaboration or supplier collaboration so that that's aligned. It can help prioritize work in terms of expediting and de-expediting. So there's, there's many factors at play, and again, if you were to think of one parameter that might touch on 15 or 20 key elements of how much work it is, how effective you are, and how accurately you can predict, lead time is one of those key levers.
Thank you for your time today. We know there's a lot of noise around AI and ML, everything from chatbots to social media ads, but what we focus on here is pragmatic ML and AI that can be operationalized to be resilient across teams and time, and has built-in bidirectional plumbing. So we're excited to bring this. We think it's highly useful and easily implemented. Thank you.
(00:05): Hi, my name's Bill Benton. I'm the co-founder here at GAINSystems. I'm excited to talk to you today in this GAINS ON segment regarding lead time sensing and prediction, empowered by machine learning through GAINS.
(00:40): Data's noisy, right? So if you look at your supply history, there's gonna be returns, there's going to be negative amounts, there's going to be dates that weren't accurate. How do you work through that? We use some anomaly detection. So when we talk about machine learning for prediction, we also use it beforehand for detecting those anomalies and the data coming in, so that we do an initial cleansing. Secondly, we work tightly with practitioners that are trying to employ this with our GAINS Labs team to refine what we call the feature data. So this is input data, and we'll usually iterate that two to four times to get to the point where we're convinced that there's vastly more improvement than degradation.
(01:46) So, traditionally, that's been hindcasting. So when you look at your supplier performance, on time performance, split lines versus whole lines, etc., you might be punishing somebody who's been improving rapidly and is likely to do well in the future. Or conversely, you might be rewarding somebody who's very likely to have an imminent decline in performance. As you think about anything where you're trying to negotiate and you're looking at what they're going to deliver, it's much better to know what they're likely to deliver next versus what they've delivered last.
(02:29): One of the achilles heels of machine learning models and making sure that they're both operationalized and resilient across time and teams is refreshing. One of the reasons that we've built this into GAINS as opposed to simply importing other people's models, is to ensure that. So we collaborate with customers on the feature data, and as you learn more about AI and machine learning, what you'll learn is that the methodologies are less important than the data you feed it. And so what do is we ensure that there's a consistent refresh of feature data and looking at new types of feature data as conditions change, but then also that the model automatically refreshes and retrains with available data and an available target to put that output as opposed to requiring, say, a decision scientist to rerun; it's built to last.
(03:28): These would be environments where A: lead time is longer and more variable. If you are having to deliver parts or goods in far less time than you get advanced notice and you have a long lead time gap, this could be incredibly helpful. The more variable that lead time or lead time gap is across time, the greater the need for this. So if it's steady, if it's long, but steady lead time sensing and prediction isn't gonna be as valuable. If it's long and variable and variable cyclically this can be hugely valuable, as we mentioned to improve myriad operating performance measures as well as productivity for the planning team. Thank you for your time today. We know there's a lot of noise around AI and ML, but what we focus on here is pragmatic ML and AI that can be operationalized to be resilient across teams and time, and has built in bidirectional plumbing. So we're excited to bring this. We think it's highly useful and easily implemented. Thank you.
Optimization as a Service (OASIS)
In this installment of GAINS On: Co-Founder Bill Benton discusses GAINS Supply Chain Optimization as a Service (or OASIS) offering. What it entails, how most companies can benefit from its offerings, and why OASIS is able to offer greater abilities at a lower cost than doing it yourself.
(00:05): Hello, my name's Bill Benton. I'm the Co-Founder here at GAINSystems, and I wanted to talk about a concept we call OASIS.
(00:32): You might have heard the term pass planning as a service or BPaaS business process as a service. This is a flavor of that that's intended to provide maximum results with minimum time and complexity. And this alone can help, optimizing this can help reduce working capital by 15 to 20% and/or reduce expediting and/or increase service results. So, so the first thing here is what is my inventory policy? And that's gonna be in two parts. One is what is the optimal amount of service or safety stock that you should hold by product or SKU or part? And then within that, across each location, and then how much should you replenish? What's your order size or order quantity? The second thing we look at is combined, do you stock, and if so, at what service level. So there we're talking about what is your deployment strategy?
(01:35): Do you have every item, part, or product at every location, or do you say withhold that, and have some I items only at higher levels, right? And not at the end points or point of sale or consumption. And then if you do stock it, what is the optimized service level or availability of that item? So as you probably know, the, as you increase availability or service level, your need for inventory goes up at an increasing rate. At the same time as you increase service level, certain costs go down like expediting or penalties or loss sales or premium purchases from alternative suppliers, all that goes down. There's some point that's optimal. The last two elements that are delivered as part of OASIS, our flow optimization and what we call multi-echelon optimization, these aren't necessarily applicable to everyone. Flow optimization is applicable where you do have the ability to work around tiers in a multi-tier network.
(02:38): What flow optimization does is determines not just, you know, periodically after a design project once every four years, but continuously where you should change flows. Do you want to go direct from the supplier or the plant to the DC bypassing the hub? Do you want to go from the hub to the branch or point of sale or field service location bypassing the DC? When do you want to do that? Maybe it's seasonal, maybe it depends on what your minimums are, what your transportation dynamics are, frequency of delivery. We take all that into account and give you a simple output, simple to implement. Not simple to define, but simple implement, which is what are your sourcing rules and when do they change, right? So this can be simple, which is we just want to do it once a quarter. We don't need seasonal profiles or we wanna really look at it and truly optimize in depth.
(03:35): So the first of five benefits or advantages of OASIS versus a traditional implementation or transformation project is the ability to move at pace. So as I alluded to earlier, we look for very raw data inputs, simple demand transactions, supply history as it relates to work orders or purchase orders or transfer orders. And with these data, we'll mine all of that to create what we need to determine the optimum inventory policies. So this can be done periodically. Usually it can be pulled without IT resource and even allow it to be done through analysts and, and not IT resources. And then secondly, how do we minimize the amount of data going back? As I alluded to earlier, there can be as few as two fields that we're passing back and we can do a Pareto approach by looking at every period that can be monthly or even as infrequent as quarterly.
(04:39): What are the top X percent of items that give you 80% of the benefit so that we're not constantly doing a lot of updating for marginal improvement, but really focusing any effort on the maximum leverage changes. And in that fashion, again, we can defer or even avoid IT involvement given the scarcity of those resources in updating your current planning system. Or planning methods could be Excel, or reorder point, as simple as that, with these optimum values. The second advantage of a process like OASIS is benefits realization. And this comes in sort of three distinct elements. One is work out, right? So if this doesn't work, we know you're not gonna continue it. You're not, you haven't sunk into a long-term commitment for a solution that has to get integrated and work through several stages before you can use it. So we, we know we have to deliver not just output, but results.
(05:40): Secondly, it's extremely streamlined by virtue of the fact that we know how to do this. We've done this hundreds of times and know how to deliver quickly and focus, more importantly, focus on those things that are really gonna add value rather than, say, merging it all together in one large transformation. And then lastly, depth. So I mentioned multi-echelon, I mentioned optimizing service level. These, these are things that probably you simply aren't capable of quickly, but something that we could do faster, right? So that still might be a phase two of OASIS, but phase two might come in half the time of a phase one of a traditional deployment. Third element is skill sustainment or supplementation. So the skills required to optimize inventory, optimize flows, et cetera. These are fairly rare. They are sometimes hard to retain. And for medium and even larger organizations, it's hard to have a team that stays interested and does this month by month in the same context.
(06:59): So we at GAINSystems have a large and growing team of engineers and decision scientists that revel in this and, and have a great variety of tasks across industries and within industry, across companies. So we have a dependable team that has the skills that we have sustained that's ready to deploy rapidly. So this blends with some of the earlier benefits we discussed, but in terms of complexity, by preventing a long system acquisition and selection process, myriad approvals, by looking at this simple consultative service we can, in fact, streamline the point to making a decision and getting results started fast. On another complexity note, we minimize the IT integration and the amount of data because we can do a lot of data fill and we can estimate and we can use parameters and heuristics to avoid having to burden your IT organization or your planning organization with dozens of decisions before you get your first output.
(08:09): So with our immersed team, lastly, we'll talk about fifth and final benefit that we think is very significant and that's minimized disruption. So rather than the team having to do their day job in an old system and then implement a new system and coordinate across different groups, including IT and others, we bring that to the table. So this is not an additional job, in addition to your day job, think of us as supplemental teammates that are here to help you do your job better. Additionally, by improving inventory balance as a planning organization, you're dealing with less expediting, which reduces your effort significantly, and reduced disposal of surplus and excess. So we believe not only are we implementing a tool to get better financial and operational results, we're a net reduction in effort because we're gonna reduce your expediting and excess disposition. In summary, OASIS provides means to significant results, very quick fashion, with a very streamlined approach, both in terms of data and effort, and we believe is the fastest way to a quick win and a nice lead-in to subsequent advanced planning capabilities. Thank you for joining. Again, I’m Bill Benton, Co-Founder of GAINSystems. We look forward to talking to you about this and opportunities in the near future. Thank you.
In this Episode of GAINS On, join GAINS Co-founder Bill Benton as he guides supply chain professionals through the shortcomings and pitfalls of Demand Driven Materials Resource Planning (DDMRP).
(00:00): Hello, my name's Bill Benton. I'm a co-founder here at GAINSystems. Thanks for taking time to join us today. We'll be talking about DDMRP, that's an acronym denoting Demand Driven Materials Requirements Planning.
(00:32): So we're here to talk about a principle that some organizations, including APICS, are promoting for simplified means of improving planning. One is called DDMRP. Some advantages of things like DDMRP as they relate to improving over baseline MRP, which includes usually a fairly simplistic finished goods forecasting process. So sometimes forecasts are a bit optimistic. As well as you know, they often encapsulate simplistic inventory policy measures like certain weeks of supply for certain inventory classes. There are problems that are inherently complex and failing to embrace that complexity and manage it leads to inferior results in terms of lower than feasible inventory terms, more expediting charges, lower service level to your end customers, less accurate predictions for revenue and budgeting. And because of all this, we think that the slight increase in complexity of an advanced system like GAINS is well worth the investment.
(01:44): So the first point worth mentioning here is this idea of multi-horizon planning. So we have the concept of, for example, of frozen slushy and then highly variable liquid period. So in order to solidify transportation plans or solidify sequencing on the floor and manufacturing execution, you might want to freeze your supply plan over a one or two week horizon. Secondly, you might have a slushy period where it varies, plus or minus some amount. And then lastly, you're gonna have completely liquid. And within DDMRP, it's presuming that that these things are, are equally variable across time. So we think this is one area where advanced planning can be very helpful. Secondly, there's the concept of supply sensing. So there is lead time variability and it's important to account for that.
(02:46): DDMRP doesn't always do that. Base level core supply variability management, it's more based on demand variability. But even if that is included, supply variability can be parsed in the segments across time. So, for example GAINS has AI algorithms for supply sensing, and those algorithms can actually reduce variability over the short run and thereby reduce your need for working capital and therefore increase service or reduce inventory of both. So this is quite important element as well that DDMRP doesn't manage. Third, we have the concept of looking across different available inventory before executing more production or purchasing of the given item that appears to be below what's needed.
This can include things like redistribution of excess from other locations, where handling and transporting that excess costs less than holding it, or less than expediting into this particular site could look, look at alternate products where they might be slightly higher cost, but available and much lower cost in acquiring new material, rather than using what's available. These things are very important. And they're related to the fourth element here, which is potentially looking into alternative supply sources. So you might have other production lines that could produce something, you might have different suppliers. It might be higher cost, but shorter lead times, it could provide fill and demand for purchasing. All of this nuance and opportunity really is not typically available in any kind of DDMRP model at all is to be done supplemental, usually in Excel.
(00:00): Hello, my name's Bill Benton. I'm a co-founder here at GAINSystems. Thanks for taking time to join us today. We'll be talking about DDMRP, that's an acronym denoting Demand Driven Materials Requirements Planning
(00:33): Particularly in seasonal environments, but even just with general variability across time. There's gonna be times where you need to pre-build to avoid having less than needed to capacity in the future. One of GAINS' practitioners, for example, has over 50% of their demand in 10% of the year between Thanksgiving and Christmas, they’re a jewelry manufacturer. This type of pre-building, pre-buying, etc., really isn't feasible in the DDMRP construct. Some demands are large project based demands. These are things that are known in advance, are large increments, which will increase variability unless they're recognized and filtered as one-off events that are scheduled in advance. And secondly, there are things like outliers, one-time exports selling off excess, you know, at a low price. These are things where you don't want to plan more inventory just because you have these one-off events that you can see as outliers and/or filter as incremental scheduled demands. So that's the first.
(01:39): Secondly there are simple things that cause variance that can be managed like seasonality. So in a DDMRP environment, fluctuations across the year are seen as variability, no different from gaining and losing customers in many cases, new product launches, right? So that can be seen as not a step function upward because you've now adding a new way of new products that might use the same componentry as existing products, but seen as another false source of variability. And then of course, things like promotions. All of these things can be managed, recognized, and filtered out so that you understand that these aren't random variability that require costly management methods or working capital investments, but things that can be dissected and managed. Now, that said, you do need to manage around that core variability that can't be managed and discerning between the two is essential.
(02:37): Prioritizing given scarcity. So if you're trying to fulfill as much demand is feasible where you can't fulfill all the demands and you have to make decisions about which components cure or produce or which items to expedite for distribution or sales order. Understanding the interdependencies between items that go into a same bill material or interdependencies between items on the same sort of perspective sales order isn't at all managed by DDMRP. So again, that's, that would be a large manual effort to say, I have these shortages. How do I combine shortages of these items across other items in a BOM or bill of distribution or bill of sale into a complete perfect order. I want to thank you for your time today and would love to chat further about all of these topics. Thank you.