This article is brought to you by Retail Technology Review: Improving your demand and fulfilment processes.
RetailTechnologyReview.com spoke with leading representatives from the analyst and vendor communities to discuss some of the most important key talking points and areas of innovation taking place within demand forecasting and planning technology.
With a more fragmented and volatile playing field becoming the norm, companies need to be able to make better more accurate decisions to manage anticipated and actual demand as well as execute and fulfil orders as efficiently as possible by avoiding stock outs and delivering on time and to spec regardless of whether the order is direct to store or direct to the consumer.
Fortunately, the technology available is keeping up with these challenges. It’s just a case of being able to navigate your way around what’s available and what could fit your own specific needs and available budget. So, what precisely are some of the key challenges the demand forecasting and planning-related solutions that can make them less of an onerous process?
Connection between demand and supply is key
In terms of supply chain planning solutions, Tim Payne, research vice president, Gartner, makes the point that the demarcation between demand planning and supply chain planning has largely disappeared now. “So, the connection between demand and supply is key and increasingly we’ve seen technology being able to cover both the demand and supply planning side of the equation because that’s really important,” he says. Payne adds that everything is on cloud these days. “Nobody brings a new planning solution to market that is purely on premise.”
Payne also explains that more vendors are now adding more AI and machine learning into planning solutions. “Changes in business models means companies need a lot of flexibility in the planning solution,” he says. “For example, consumer goods companies traditionally sold to brick-and-mortar retailers and that was their main channel. Now, with e-commerce these companies also have to look at the demand for online sales. This is a business model change.
“You’re still applying essentially the same principles in terms of you’ve now got to look at the demand for your e-commerce channel in the same way as you have done for brick-and-mortar demand, although the demand for online sales will be driven by different factors than for a brick-and-mortar retailer. So, companies need to consider how much inventory to hold and whether to have separate inventory locations or stocks for the two channels – in store and online – or combine them because you want more flexibility. So, the supply chain planning solution has got to be able to cope with this omnichannel model.”
Improving the quality of decision making
However, Payne believes the biggest focus for companies, whether they are retailers or high-tech, pharmaceutical or industrial manufacturers etc, is being able to improve the quality of the decisions they make. “So, there’s a lot of focus on processes – the demand planning process, the supply chain process. the sales and operations planning process etc. However, we often get fixated on processes – are we following the demand planning process, is everyone conforming to our standard S&OP process? However, the point of planning is fundamentally to make decisions. Planning is a form of decision making and we have to decide how much we think we’re going to sell, move, make and put in inventory. So, the outcome of planning is the decision, and the outcome of good planning is making good decisions – what I describe as higher quality decisions.
“If we make higher quality decisions then we’re able to reduce value leakage and create opportunities to increase value because we get the right resources in the right place at the right time, and we can take advantage of disruptions and events that are happening in the market. So. there’s a switch happening, particularly now there’s so much digitisation going on. With all the digitisation and digital transformation work that companies are doing a big focus area we find for that is supply chain.”
Combination of different analytical techniques
Within the supply chain, Payne comments that the top focus area is supply chain planning because digitisation is about using lots of data and analytics. and particularly machine learning, which is all about prediction and planning is about predicting. He adds that automation of decision making is also a key focus. “So, there’s a lot of focus going on from manufacturing companies in terms of how they can improve the quality of the decisions that we make,” he says. “That’s driving many of the technological changes, not to take out optimisation approaches but to add in additional analytical techniques such as machine learning in all its various forms, deep learning and natural language processing etc. So, it’s becoming a combination of the different analytical techniques that helps to improve the quality of the decision-making.”
The impact of omnichannel
Bryan Ball, industry analyst and consultant, ex-Aberdeen Strategy and Research, makes the point that Covid applied a lot of pressure on the ability for many companies to fulfil orders, largely due to the growth in omnichannel. “It meant many companies needed to fulfil orders from different points to what they had initially planned to fulfil from,” he says. “For example, in the food and beverage industry, if a company’s usual distribution delivery points were grocery stores and restaurants because people were eating in restaurants and shopping in grocery outlets, it suddenly had to rethink everything because restaurants closed during the pandemic and everything either went through the brick-and-mortar grocery channel or through online orders. So, companies serving this sector had to adjust very quickly and move things around in a different way.”
Re-thinking fulfilment
So, Ball explains there were new challenges on the demand planning and forecasting side relating to the inbound information received. “In other words, there were new issues concerning where the demand was coming from, the timing of the demand and the volume of the demand as well as questions around the levels of data accuracy and demand volatility and so on,” he says. “However, with the huge growth in home delivery, for example, largely due to the pandemic, companies also had to rethink how they repositioned themselves on the fulfilment side, the execution side, and think more about where the products should be located in order to fulfil orders more quickly and cost effectively. Historically, goods would normally be stored at traditional distribution centres that the company had established, but because of the shift to direct-to-consumer model, some companies, particularly some of the larger ones, started to think about how they could use store sites as fulfilment distribution points because they were nearer to where much of the direct-to-consumer orders were coming from.”
Ball continues: “Historically, they might have relied on regional DCs covering large regional areas, Now, because of the major growth in the direct-to-consumer model, they might decide to position them in a major city or large locale closer to the point of delivery – maybe New York, Philadelphia, Atlanta, Houston or Los Angeles for example. Previously, companies might not have considered this as an option due to the local logistical challenges due to traffic congestion, but because these types of locations are now a hotbed of people ordering online, delivering to residences and condos has become more of a norm so companies are increasingly using their store sites as fulfilment points. So, it’s now not only important to capture as accurate data as possible on the inbound demand and forecasting side, but on the outbound side there’s a necessity to create as intelligent a model as possible telling you where best to stock products to minimise your cost and delivery.”
Whereas demand planning and forecasting used to be more of a front-end piece related to what you did in the supply chain, Ball explains that it has now become a very vibrant piece of what you need to do for effective execution and the fulfilment in the new omnichannel world – direct to consumer or brick and mortar stores. “Most of the planning model is based on inbound information relating to how you can better itemise certain items and get better specificity about the best location to send them to,” he says. “Although overall demand for a particular type of item might be fairly stable the type of demand might vary depending on where the customer is located.
“For example, think about small, medium or large clothing. The percentage of sales in small, medium and large might not differ a great deal in total, but the percentage of each size could vary a lot depending on locale. It might be the case that larger sized clothing is more in demand in cities, it might be the case that lighter clothing is more in demand in the South where temperatures are more consistently hotter over the year. So, the demand forecasting and planning solution should offer a greater level of sophistication at the fulfilment end. You probably won’t need snow ploughs in the South, so if you have a plant that makes snow ploughs it would be best to put it in a location that has snow and maybe mountains, Tennessee for example. It’s a good point of distribution to customers and it also offers competitive-cost manufacturing.”
The impact of social media
Steve Murphy, director – client services, Panorama Consulting Group, observes a number of key areas that are changing the face of demand forecasting and planning today. “One is the evolution of omnichannel to satisfy consumer demand, and the choice consumers have now between in-store purchases and online orders,” he says. “Online sales have exploded over the past few years, particularly since the pandemic. On social media, we’re all bombarded now with targeted ads based on tight tracking of your online activity today. Only a few years ago, the ads you would see pop up would be from four or five main companies that were targeting that type of advertising. Today, if you visit an online page within hours, you will start seeing ads pop-up related to that company and its products. When you sign up to your landing page, whether it’s Google, Yahoo or whatever the case might be, you’re going to see ads or stories about that retailer or that product.”
Murphy believes this is not only changing due to technology today, it can also change due to major events, particularly the pandemic. “The pandemic was a ‘one-time’ event, but it changed everything,” he says. “It changed how companies operate their supply chains and the large transportation companies had to rethink how they were going to deliver goods. Today, in the case of ocean freight, for example, you can now check-in at any time and see exactly where a shipment is on GPS.”
AI and machine learning
Another major development within demand planning and forecasting today, according to Murphy, is the evolution of AI and machine learning. “Leading ERP vendors such as Oracle, SAP and Microsoft as well as the specialised demand forecasting and planning solutions providers can, for example, use AI to take the trending economic patterns over the past three months, pull it into the system and accurately estimate what the demand will likely be for the next month. The level of accuracy of these systems has improved by leaps and bounds.”
Murphy adds that even though machine learning provides more and better data, one of the key points to remember is you still need a human being to have overall control. “In the case of major events that could have an impact on product sales, such as the Super Bowl, people who understand demand forecasting and planning based on years of hands-on experience might say I think the stock levels should be bumped up 1% above what the data suggests or pump it down by a similar level. This often can work out to be more accurate than the machine learning data suggested. So, you still need that human factor based on demand forecasting and planning experience rather than purely relying on the numbers that come out of the machine.”
Mukul Krishna. global research practice leader – supply chain and logistics, Frost & Sullivan, reflects that it was only a decade or so ago that the industry was just beginning to digitise, and people started looking at collecting data and creating data reports. “A lot of valuable data started coming out of that in terms of increased forecasting accuracy,” he says. “Then, more recently, the pandemic hit and this made many companies re-think how they manage demand forecasting and planning.
Moving on from historical data
“Someone in the apparel industry told me that his company’s planning for the Spring of 2022 was based on the previous year’s data. However, in the wake of the pandemic all this historical data going back a year or so went out the door. In volatile times, especially when things are changing very rapidly, historical data means very little. Typically, what demand forecasting has relied on has been this historical data, but now more people are very cognisant of the fact that there’s so much uncertainty out there that it’s very difficult to even read regular economic data.”
Even before the pandemic, Krishna points out that many retail customers were becoming very comfortable with the idea of e-commerce. “Then during the pandemic, these customers understandably became even more comfortable booking online. So, companies don’t only have to manage both brick-and-mortar deliveries and direct-to-consumer deliveries, but also need to factor in reverse logistics because some customers have got into the habit of ordering, say, 10 items but only intending to keep 5 of them, or even fewer. So, now there is the additional challenge of managing returns and getting the items back on the shelves or back in the right place in the warehouse or DC to be ready for dispatch to another customer.”
Krishna adds that some companies are still paying attention to historical data but now rely more on data that is only a few months old. “They’re also starting to use more artificial intelligence and trying to triangulate as much of what is happening to try to figure out true demand,” he says. “Just because something happened last year doesn’t mean it’s going to happen this year, so companies want to increase their probability of having a much better sense of accurate data in times of greater uncertainty.”
Also, with climate change, Krishna believes companies need to ask themselves whether it will be a warmer winter because this could impact on greater demand for certain products that historically might not have been so much in demand at that time of the year. “So, things like this are now becoming more top of mind for companies whereby they wouldn’t have thought so much about them in the past when trying to anticipate demand.” In terms of trying to figure out more accurate demand patterns as opposed to relying on historical data, Krishna explains that more companies are now trying to model data better using AI or advanced analytics to start becoming more predictive and prescriptive. “All this can help to introduce more probability into the algorithms,” he says.
The SaaS/on-premise debate
Ball observes that many companies and best-in-class companies are certainly moving, or have already moved, some of their functionality to the SaaS model, both in terms of demand forecasting and planning and ERP. “They may decide primarily to move certain pieces to the cloud, such as decision support,” he says. “They may not decide to move financial planning because they see their financial numbers are their ‘keys to the kingdom’. They might decide to put their planning data in the cloud.
“However, even then they may want to be more secretive about that because their planning data has volume, product, marketing and pricing information. So, they might be guarded about that type of data. Nevertheless, they might decide to take slices of that data and move it off-site. In general, many companies have got past the attitude of keeping everything in-house. That said, there are still many manufacturers who don’t want their secret formula in the cloud and feel safer if it’s on premise. In the case of Covid where people couldn’t continue to work on site, SaaS proved very valuable in ensuring data such as that related to inventory could be accessed wherever the people were who had the authority to see that information.”
Having an edge
Krishna considers that many of the initial concerns related to SaaS have gone away. However, he believes that in certain industries, such as retail, on premise solutions and edge capabilities are equally important in managing the omnichannel model – direct to customer and direct to store. Krishna also makes the point that edge computing can have an advantage over the cloud in terms of reduced latency, something he believes to be increasingly important in a supply chain world where quick response can be critical to keeping up with demand and stock requirements.
“During the pandemic, many people fell sick and quitting also reached high levels,” he says. “Many left their jobs to re-skill or up skill and get into the gig economy. Largely because of this, companies tried to leverage more AI-based automation. So, for example, more inventory management robots were used. These robots are basically edge computing devices on wheels. Meanwhile, RFID scanners and machine vision were deployed to scan items down the aisles to determine what’s in stock and what’s not. So, these types of tasks that might have been considered tedious for human workers can now be effectively done by automation and are able to give you information largely in real time.”
Keeping on top of unexpected trends
Krishna reminds us that when the pandemic hit people started making a beeline for all types of items that under normal circumstances wouldn’t fly off the shelves, such as toilet paper. “In my own local grocery store I had never seen it run out of onions before Covid,” he remarks, adding that some shops then started to ration certain items, allowing two items per customer, for example. “If you have data coming to you in close to real-time you can start monitoring these unexpected trends and institute certain policies that will help you to prevent stock out,” he says.
“However, data sent to the cloud means getting it back will experience some level of latency, and even a small amount of latency can make a big difference to meeting demand and following trends. So, you want to minimise the level of latency. For example, you don’t want your autonomous vehicle talking to the cloud. Instead, you want that vehicle to make autonomous decisions without needing to communicate with the cloud. So, if you have a lot of autonomous vehicles doing last-mile delivery using on-board edge computing capability to make decisions rather than having to go into the cloud and back, this can be much more efficient. Similarly, your inventory management robot in the warehouse using edge computing can let you know in near real time that you’re running short of a certain product and can order more before you experience stock out.”
Factoring in the expense
Krishna adds it is often said that if you throw enough money at the problem the problem will go away. “However, many companies don’t have large amounts of money. Cutting-edge technology can be expensive, so in developing countries where labour is still relatively cheap, many companies will continue to kick the can down the road in terms of investing in cutting-edge technology. Instead, they will just employ more people. If you look at more affluent areas such as North America, Western Europe, South Korea or Japan, you will see more use of warehouse automation and robots, especially in terms of picking robots with active picking arms – although in more complex warehouses where aisles can reach 30 or 40 racks high, picking robots would need to be highly articulated and move at very complex angles meaning a lot more complexity is involved. So, because of this type of complexity and expense, companies need to have a very good economic reason to move to more automation. Many companies don’t think their situation is that dire, and they have enough people available to manage picking in a more manual manner.”
If they have the budget available, Krishna explains that more companies are now using co-bots too. “Nevertheless, as automation becomes more common, I still don’t think the concept of the dark warehouse is going to move forward a great deal over the next two or three years at least,” he says. “The dark warehouse is, of course, a sensitive issue in that machinery can potentially replace much of the human workforce in warehouses and DCs. The counter argument is that in many cases more automated technology can augment and assist the work that the human workforce does.”
The extension factor
Even though SaaS has been around for several years now, many companies are still more comfortable having their servers on premise, maybe for security concern reasons although these are minimal today. However, Murphy explains that if you look at the long-term costs of an on-premise solution, it can be considerably more expensive because of the need to upgrade on site and possibly hire consultants to undertake extension work (extension being the term now commonly used rather than customisation). “Of course, one of the main benefits of a SaaS subscription model where a company pays quarterly or annually is that, at least for most of the upper tier companies, an automatic quarterly update to their software takes place. This means they are always up to date with the software and using the very latest version. I think that’s probably one of the biggest benefits of SaaS.”
Another delineation between on premise and SaaS, according to Murphy, is with on premise if every time you upgrade you decide to add some extensions you will likely need a consultant to come in and manage the extension work. “With the SaaS model, you don’t want to customise the solution for every user so the functionality is normally based on best practices for particular industries. If someone has a particular need for an extension to fit a particular business more tightly what we recommend before you go ahead with this potentially costly plan is that you think carefully about what you want to get out of the software.
It is important to know what the overall benefits will be and whether it makes sense to do it based on the extra cost involved. After careful thought, you may decide that it would it be more beneficial just to rely on the standard software package. So, a cost benefit analysis or change benefit analysis makes sense. If an extension is the preferred option, we can help the software companies design that extension. Doing extensions doesn’t seem to be as complex or difficult a process as it used to be. It’s not now the same as doing some of the heavy-duty customisations that we used to.”
What lies ahead
What might be the next innovations/developments to look out for over the next year or two? Murphy explains that, by building in AI and machine learning into today’s demand forecasting and planning solutions, the technology can continually learn from all the transactions that take place, both at the order and fulfilment end. Something else to think about, says Murphy, is there are so many more sources of data to pull from now to monitor demand trends, including data from social media. “It used to be that you looked at past sales history and economic predictions and what was going on in your marketplace based on different regions and what with the sales trends in those sections of the country were.
“Now, the sources of data are so vast that trying to collect more and better data to put in the system is one of main goals. So, I think if we can find better ways to collect data to use within demand forecasting and planning systems, that’s where the main improvements will lie. I think somebody out there is going to design an even better data collection process to pull in this valuable data from all these vast sources. Then, it’s a question of how this more valuable information is corralled and processed by the best demand forecasting and planning solutions. This will be the next step.”
More automation to mitigate the tight labour market
Continuing the theme of possible future developments, Alex Macpherson, director of solution consulting and account management, Manhattan Associates, points to the continuation of automation to mitigate the tight labour market, especially in the warehouse sector. “This is to provide capacity in the peak periods which are event driven and not just the usual seasonal peaks that business have experienced,” he says. “The format of this automation will vary from conventional ASRS and conveyor-driven automation to cobots and robotics.” Macpherson adds that the use of AI and machine learning will explode within the warehouse environment, driving many tasks that were manually initiated such as running of waves and anticipating labour forecasting. “The sector has been one that has not seen extensive use of AI and that is about to change,” he says.
Macpherson adds that it is going to be interesting to watch how retailers treat returns over next 12 to 18 months. “The extent that returns have on all businesses and the huge costs in managing these has finally been realised and will be tackled,” he says. “Whether this is charging for returns or getting customers to pay annual fees for returning goods, this will be another area that will change quickly and decisively. We have already seen first mover advantage by several high-profile retailers, and this will give the impetus for the rest to act to.”
No lights out
Payne believes we will see a lot more from an AI perspective. “If we look back to pre-Covid times, I heard a lot of end-users saying they wanted lights out planning, no touch planning or autonomous planning. Fortunately, there’s been a realisation by these leading companies that that’s not going to happen. You’re never going to automate all the decision-making in the supply chain. You can automate a lot of it, but you can’t automate all of it. There is still a need for certain types of decisions for human input human judgement, which is what we have always said. Completely autonomous planning was a pipe dream, but you can do a lot more than the very manual way that planning is still done by many companies on spreadsheets.”
According to Payne, generative AI will have an increasing impact. “Currently, many are saying Chat GPT will transform the way we do things. It’s just another AI technique, but the use of large language models could transform the way planners interact with planning systems. So, you could have more of a natural conversation with the planning system. That’s probably where we’re going to see some of the initial use cases in the world of planning.”
Synthetic data
Another area of innovation that will likely gain more traction, in Payne’s view, is the creation of synthetic data. “You could potentially use your digital supply chain twin along with generative AI capabilities to be able to create synthetic data – in other words, data that hasn’t been created by the physical supply chain but has been created digitally. With this data, you could test out all sorts of scenarios and options.”
Structural change
A further development we could see over the next couple of years, according to Payne, is a change in the structure of demand forecasting and planning solutions. “Today, when companies buy a planning technology solution, they might say it’s got to do demand planning, inventory planning, replenishment planning, production planning, sales & operations planning or integrated business planning. Basically, what they’re looking for is a complete end-to-end planning solution. This is where vendors such as Kinaxis, SAP, Oracle, Blue Yonder and all those big platforms play.
However, it might be the case that a company wants extra functionality that isn’t built into the closed platform they currently use and therefore seek third party solutions or build something themselves, maybe using their analytics and data science teams to fill the gap with scheduling or analytics etc. However, a trend is growing for solutions to offer interchangeable building blocks of functionality, whether you use most building blocks from one vendor or a mix. Gartner calls this composability, making a solution much more modular and adaptable.”