Harnessing the power of data analytics to tackle global supply chain challenges took on real urgency when chip shortages began three years ago.
Component access is improving now, but the lessons learned during the peak period of shortages serve electronics distributor Avnet well now, as they take meaningful, practical shape from the board room to the regional sales territories.
In this Q&A article, Avnet’s Rondekka Moore, vice president of the company’s Global Business Analytics Center of Excellence, discusses how machine learning and data analytics helped the company rationalize orders placed during the unprecedented, volatile period of shortages. What was learned during the chip shortage helps Avnet today make more accurate internal and customer recommendations about future orders.
Understanding the strategic uses of data analytics by distributors like Avnet can help OEMs weather this and future component shortages or disruptions.
Can you tell us a little about Avnet’s Global Business Analytics Center of Excellence and your role?
I’ve been with Avent for 29 years in a variety of roles. I’ve been a field account engineering technical manager, and I ran demand creation for the Americas region at Avnet for many years. I left that role in 2014 to take on analytics for the region and focus on machine learning, analytics and reporting. The center of excellence is made up of data scientists and our role is to take business problems and solve them, and we use machine learning quite a bit to do that.
Today I would say that we focus on the practical aspects of machine learning for decision support, and we are moving toward other aspects of artificial intelligence for decision management.
What’s the difference between decision support and decision management?
Decision support says: “It’s going to rain tomorrow.” Then, it lets you decide whether to take an umbrella. Decision management forces you to leave the house tomorrow with an umbrella. What we do is decision support.
How does Avnet use machine learning right now in areas such as forecasting and how is that related to decision support?
We use machine learning in a few different areas and one of those is at the level of forecasting what I call general sales. We use it to understand where we’re going to finish the week, the month or the quarter. We use historical data and place the most weight on recent historical data to make our projections.
We also use it to help us understand opportunities that might exist within customer accounts where they may need other products or technologies.
That brings us to the recommendation piece or decision support, which is most of what we do and we’re really quite good at it. We provide answers to the business, and they then make decisions about what to do with the information. What we do—decision support—is very different from decision management.
Can you tell me how this works, practically speaking, for salespeople?
Based on the analysis, I may want to tell the sales team, “Go sell these particular microcontrollers to customer XYZ.” That recommendation is based on what we know the customer has bought in the past and through understanding what they’re interested in buying. We present the recommendation, and then sales decides whether they’re going to act based on what they know about the customer. That’s decision support, where we’re using machine learning to help the team make a decision.
This came into play during the pandemic when we fine-tuned things a bit. For example, we’d have demand for 1,000 pieces of some widget. We developed a model that told how well you might want to believe that 1,000 pieces because, in reality, what they need might need is 1,500 pieces or 150. It’s difficult to tell without analysis. Our machine learning model was able to help one of the logistic teams based on inputs such as lead times and historical performance, by suggesting the order levels where the team might take action.
We did a lot of that, especially during those early days of the chip shortage. It helped guide decisions. We saw a great deal of savings by being able to avoid costs by double-checking those types of things. What we’re really doing all the time is smoothing product-by-product this mismatch between the customer and a supplier.
How does this apply to supply chain management?
An example would be when a customer says I need 1,000 widgets over a period of time. For the supply chain models we have around the world, we do demand rationalization and we have started to incorporate a little bit more machine learning into that. Basically, it means when you ask for 1,000 widgets, we’re going to use a model to figure out if the amount needed really is 1,000, or do you need 800 within a certain time frame, or will you eventually need 1,200. We need to know these specifics because we need to place the order with the supplier, and we need to be as accurate as possible. So, we get the signal from the customer and then we analyze that signal, rationalize it based on various parameters and act accordingly with the supplier.
How did this type of analysis help Avnet navigate the recent chip shortage?
It is complex, and it’s what makes our business somewhat different from a lot of other types of distribution. A partner might say, “Yes, I can give you that 1,000, but I can’t do it at this moment. I can deliver it in 12 months.” Then, in six months they might say, “OK, I can deliver it two months from now.” It can be a very fluid situation where delivery dates change based on a multitude of factors.
During the chip shortage, the lead times all stretched out to more than a year in some cases. That created a mismatch between what the customers were ordering and when they needed it, versus the delivery time from suppliers.
Machine learning helped guide us through it. It helped us identify outliers, like a customer who, even if you consider a 52-week / a year-long delivery time, has ordered way more than in the past. So, we’d know we need to go investigate, ask and confirm. We did a lot of confirming and reconfirming what the customers ordered, and machine learning helped us do that outlier detection.
Does Avnet do much outlier detection today?
Yes, outlier detection is an area where we are applying machine learning more and more. And when I say outlier, I don’t mean bad orders. I mean that it looks to be outside of what we think is the norm, and so we need to go check.
What other lessons did you take away from this period?
Another discovery from the pandemic is that we likely have way more data than we thought we did. We have to think about data differently, apply it differently and use it more efficiently. In my career, we’ve gone from this trickle of information to a flood of information that we can now distill into something uniquely valuable.
How does this apply to providing recommendations?
We use machine learning to determine what to recommend our customers buy. In a simple example, if most people at the grocery store are buying bread and cereal, but a few are also getting milk, we can then recommend milk to the customers who only bought bread and cereal.
So, what we do is use the data to figure out the pieces of the puzzle that we’re seeing across the market that might help other customers. Then we deliver that information either via Salesforce or directly to the salesperson for follow-up.
What the pandemic and chip shortage really did was change the sales process in a way that’s less happenstance and more scientific.
In the end, everything we do with data analytics and machine learning is about efficiency and growth for our customers, our suppliers and for us.
