VP & Head of APAC
Onebeat
Traditional inventory management systems struggle to keep pace in an industry where consumer demands for personalization, immediate gratification, and constant newness are reshaping business models. Retailers face a perfect storm: expanding product assortments, maintaining higher inventory levels across more locations, and navigating ever-shorter product life cycles—all while watching full-price sell-through rates hover around an alarming 50 percent.
Enter Onebeat, a cloud-based AI solution that’s rewriting the rules of inventory optimization. Roei Raz, VP & Head of APAC at Onebeat, sits down with us to explain how their technology enables retailers to break free from outdated forecasting methods and rigid rule-based systems—allowing them to sell more at full price while keeping less stock. From fashion powerhouses like Aditya Birla’s brands to growing labels like Being Human, Onebeat’s algorithms are helping merchants across the Asia Pacific transform how they allocate, replenish, and transfer inventory in an increasingly dynamic marketplace.
CIO&Leader: How does Onebeat’s AI specifically uncover inventory management patterns that human analysts might miss?
Roei Raz: I suggest starting with a quick background on demand patterns over time and how they behave if you look at the last decade. Then, we can identify three main changes in customer behavior. The first one is the fact that consumers want personalization. They want the products to fit more personalized needs, and that pushes retailers to offer a bigger assortment and more choices.
The second thing is that consumers have an instant need for gratification. Understanding this, retailers must increase their reach to get closer to the customer and keep higher inventories. It puts pressure on inventory as well, not only on assortment.
The third thing is that consumers like freshness. Freshness drives excitement. Knowing this, retailers understand that they must introduce new collections and launches more frequently, reducing and shrinking product life cycles with time. The time they have to sell full-price products is getting shorter.
Hence, with the need to grow, keep more availability, and shorten the life cycle, the time that it has to sell, retailers end up with higher inventory, with a lot of capital locked up in inventory that is not selling. We see margin erosion primarily because they sell less at full price. Today, the benchmark for full-price sales is about 50%. In India, it can go to about 45%. There is also a leftover inventory of about 15% at the end of the season.
Specifically, we see three main things that retailers, or three main habits that retail has traditionally developed over the years, prevent them from adapting to these changes.
The first one is the overuse of forecasting. Forecasting might be good for long-term prediction when deciding what to buy. Still, it becomes meaningless when we must determine how to distribute our inventories with all these dynamics and how to redistribute them between stores. There could be thousands of transactions that we can do between stores to fix our inventories. The second thing is that most of the traditional retail systems are rule-based. To keep it simple, based on the decision-making rule. And the rules are very rigid, and the reality is much more dynamic, okay? Since we don’t adjust these rules frequently, we are often reactive instead of proactive in dealing with that reality.
The last thing is that while retailers know about this behavior, they see it and meet the customer. They strive to gain insights into reality and to get insights about how to use the data, which they use to draw these insights from analytics. Analytics takes time to identify patterns; by the time they recognize this pattern, it’s already too late in many cases.
With these challenges, the idea of using AI is to apply it in places that can help us make decisions more frequently and move toward day-to-day choices. Instead of forecasting to decide how to distribute and redistribute our inventory, we use a short-term prediction algorithm that identifies consumption patterns and clubs similar products together by shortening the time to get the prediction.
Instead of waiting for weeks, we get a reliable prediction in a week or two to know which products will likely sell in which store. The second thing about being dynamic is that the system learns and adjusts the rules; the rules are not fixed. The third thing is that the system is also high-resolution.
Eventually, we need to make decisions. If you take fashion, for example, then the decision is not only about the store, category, product, style, or color but also the size level. We need to resolve the issue of managing all these SKUs across locations based on the specific locality. So, high resolution means that decisions can be made daily and applied to an SKU location. We even go beyond this.
We didn’t have statistics or knowledge about the latest launches when we introduced new launches. Instead of relying on the average performance of products in a category and forecasting methods, we use product attributes. We learn from attributes.
AI can help us identify which attributes, if we break all the products to their attributes, which attributes in a product, in a typical product, really dictate their demand pattern. Accordingly, we can identify similar products from previous collections and learn about their behavior on new launches. With this, we can improve the allocation and the initial allocation decision to stores to be leaner and more aligned with the market demand.
CIO&Leader: With machine learning algorithms powering your adaptive inventory management solution, how do they differ from traditional forecasting methods?
Roei Raz: Yes, so, as I said, we choose AI. AI is our tactic; it’s not the objective. The objective, eventually, is to help retailers improve their sales at full price. That’s the most essential thing in retail. If retailers want to be resilient, they need to build that machine to squeeze more margins from their sales and end up with profitable stores.
To do that, we identified which AI algorithms would specifically help retailers manage their execution better so that they can sell a few more pieces at full price within the product’s life cycle.
It starts with the initial allocation, where we learn about product attributes. It continues with replenishment decisions, which are driven by shorter predictions. Here, we identify the signals—we get the signals on actual demand. We learn and identify patterns of demand between different SKUs so that we can shorten the time to predict and then automate the replenishment decisions.
It continues with moving inventory between stores through store transfers, where we can identify cost-effective transactions that make sense—those that help complete sets of products with missing sizes and put them in the right stores so they can continue selling at full price. That comes with a big optimization challenge because we need to provide the retailer with something doable, something manageable, and not too complex to handle.
We also use machine learning specifically for special events. While we can control allocation, replenishment, and store transfers—the full cycle of the product—special events intervene in between. These can be regional seasons or regional events in India.
It can be Diwali, Akshaya Tritiya, or any other event that drives higher traffic. It can also be store anniversaries, for which retailers usually run promotions. We go back two years into historical data. Usually, that historical data does not exist in full.
So, we take the available data and build it into a complete dataset. Then, the machine learning algorithm learns from both past and future behavior. For example, we want to identify which products are more prone to seasonal demand than others.
We want to learn what kind of special offers impact demand—whether it’s a price discount or bundled deals like 1 plus 1. What is the impact, and how can we quantify it together with price elasticity? We try to give retailers better predictions of how to prepare for these events so they can stock proper inventory and hopefully end the event with fewer mismatches in inventory—both surpluses and shortages.
Our benchmark is that without using AI for exceptional event management, the inventory mismatch after an event is usually 50% higher than the mismatch between shortages and surplus in the off-season. That means shortages increase by about 50%, and surpluses increase by about 50% after events due to miscalculations.
CIO&Leader: Can you explain how your technology processes real-time retail data to make predictive inventory recommendations?
Roei Raz: Yeah, so first of all, we must shift to a cloud mentality to make that happen. Okay, the only way to make these calculations is by using the cloud. So, Onebeat is a multi-tenant cloud solution that integrates with the ERP—the local ERP system—which is fed with relevant data daily; anything from the master of SKUs with all the attributes, the locations, their localities, and the daily transactions that happen.
With this, Onebeat can process big data and provide outcomes. Algorithms like short-term predictions can operate relatively fast, and algorithms for optimization, like store transfers, can run in the background for hours. But since it’s on the cloud, we can balance it across different servers and still provide the output in time for the retailer to decide. Algorithms for special events require machine learning, obviously, and take time to learn.
CIO&Leader: What are the key technological challenges in developing an AI system that accurately predicts retail inventory needs across different market segments?
Roei Raz: Onebeat applies to various types of retailers and different retail segments. It covers different kinds of fashion—from fast fashion to core basic fashion, footwear, accessories, electronics, books, toys, jewelry, premium, and luxury.
Different segments have different challenges. The biggest challenge is not necessarily data integration. The biggest challenge is not applying solutions—because we have to learn and experience each segment and understand the levers we need to control to drive results.
The biggest challenge is adoption, which calls for education. Some industries or segments are more prone to technology and more aware of the need for it. When there is more awareness, it requires less education, and adoption is easier. Some segments are more traditional. Take jewelry, for example, which is very prominent in India.
It is an industry influenced by assortment and can have vast variety. It’s not unusual for a jewelry retail store to offer 10,000 different SKUs to customers—far beyond saturation, right? It’s not that they don’t know; they do—but this is what the customer expects, so they offer this kind of assortment.
They face significant challenges in managing and optimizing their inventory, but there is little awareness about technology and how it can help. In these areas, adoption takes more time—primarily until they are educated about it. Plus, much of the process is influenced by art in these more traditional environments. So, decisions are made partially through process and partially through intuition or experience. For example, the store manager in a jewelry store is not a retailer; he’s a jeweler.
They also use their art to inform decision-making. So, in these cases, we need to use technology not to automate the decision but to make the decision-making process easier and more user-friendly—so they can still apply their art and keep their choices accountable while using advanced technology.
CIO&Leader: Which has been the easiest segment to try these technologies?
Roei Raz: One of our easiest and fastest adoptions is in the fashion segment. The fashion segment usually has access to data. They also deal with complex processes and a lot of decision-making, and the time to make those decisions is very short.
Since they understand that there’s very little they can do during the season and that they need to sell more units at full price when the time comes, there is better realization, understanding, and awareness about the need for technology. There’s better adoption of it.
Usually, we see results coming in faster.
Okay, I can give you an example from one of the recent case studies we did in India with a brand called Being Human. Being Human is a growing fashion brand with an exciting leadership team. They saw the opportunity they were missing and could feel they were losing out on sales during the season—sales they couldn’t quantify.
When we partnered with them, we were able to show them the potential they could unlock by applying AI to their environment. We streamlined replenishment with short-term predictions and faster decisions. We also applied store transfers to move leftover stock before its end-of-life—so it could go to stores with higher potential to sell it.
Over six months, they saw a 10% increase in full-price sell-through, the primary performance measure in that industry—sell-through meaning how much is sold compared to what is bought. So, if the average full-price sell-through in India is about 50%, then this was 10 percentage points higher.
They achieved this with 23% less inventory in the store. That means the store became more linear, had more space for freshness, and could better satisfy the customer’s need for newness.
They were able to meet their assortment needs by providing a higher rate of freshness and better product availability for the consumer—which also resulted in better margins.
CIO&Leader: How does Onebeat’s AI technology integrate with existing retail management systems to provide seamless inventory insights?
Roei Raz: We wanted to make this technology accessible to the masses. Our vision is for any retailer to be able to operate with Onebeat. We did not have the privilege of substituting existing systems.
So, we had to work along with existing systems. Onebeat connects with the existing system, is fed with the data, and then feeds the output back into this existing system. Usually, we take the data from the ERP. If the data we need from the ERP is unavailable, we can connect to the WMS and take the data, or we can connect to the point of sale and take the data.
However, once we have the correct data, we can apply the optimization and allow the user to influence that optimization. If not, we can automate it and let it flow directly into their system. For example, they can get the picking list at the warehouse directly from the ERP system because it is connected to the ERP.
CIO&Leader: What product innovations are you making for enterprise customers?
Roei Raz: So we do. Since our philosophy is SaaS, it’s a multi-tenant cloud. We don’t offer different applications for different clients—whether it’s a top-tier brand or a growing one. But what we do—since we work with top-tier clients—is build for flexibility. For example, we work with clients from the Tata Group, with brands like Titan and Danish, and with Aditya Birla Fashion brands like Pantaloons and Max Fashion. But we also work with growing brands like Being Human, as I mentioned, as well as Caso, Just-In-Time, and many more. So we understand the needs of larger clients.
Therefore, we’ve built a lot of configuration capabilities into the product to suit their needs. These clients may have deeply rooted business rules, and without adhering to them, they cannot adopt new technology. So, instead of going around those rules, we comply with them by integrating them into the optimization model.