Two billion pieces of apparel. Half a million SKUs. Tens of thousands of retailers in more than 40 countries. Being able to analyze millions of production, distribution, sales and inventory data points manually to ensure supply meets demand? Impossible. Doing so in real time? Priceless.
Big data and the diverse team of mathematicians, engineers and business analysts that manage it are hard at work each day at HanesBrands to ensure, among other things, that the right styles, colors and sizes of apparel are on store shelves at the right time for consumers. Ben Martin, chief advanced analytics and global planning officer, explains.
What are a bunch of PhDs, mathematicians and the like doing at an underwear company?
We’re helping HanesBrands be successful! Our advanced analytics department of about two dozen “statistical scientists” are responsible for complex analytics used to crack the toughest problems. The department focuses on four core areas: inventory, demand, consumer and supply chain analytics. On one side, inventory analytics focuses on how to best utilize the company’s working capital investment across the SKUs we offer by setting the inventory target for each SKU. On the other end, demand analytics performs analyses to identify newly forming consumer trends at both the SKU and category level to forecast major market shifts that will cause retailers to respond. Regardless of the focus, our department operates under the “Think. Do.” motto. Our job is to think aggressively about how we can continue to transform HanesBrands’ operations – from supply chain to sales – and act on insights to effect change. We aren’t looking at making small, incremental improvements. We want to leapfrog what we do now, and we rely on data and analytics to inform the decision-making process on how to do so.
Can you give us an example of a project your team has implemented?
We’ve been making T-shirts for 100-plus years, and for decades, have considered ways to make the process more efficient. We took a look at it too, and built a simulated T-shirt sewing line that allowed us to change configurations at the touch of a button. We added and subtracted sewing machines, placed the machines closer together or farther apart and controlled the amount of work done at each station. We staffed the stations 100 percent of time or set rules for “workers” to stop sewing and go to another station after a certain point. Working in a simulated environment, we were able to try any wild and crazy thing we wanted.
The simulation model predicted we could increase throughput in our sewing lines by 30 percent. So we traveled to a sewing facility and implemented the changes recommended by the simulation, and the prediction of machine throughput was within 1 percent of what happened in reality.
What other projects are you working on right now?
Another example of a radical transformation is identifying a future inability to service our retailers, fix the problem before it occurs, and ensure we have product when and where our customer wants it.
Our algorithm is looking upstream and downstream to decide if there is a confluence of factors happening within the supply chain as a whole that’s going to lead to an issue. Information about both our supply chain and our customers’ supply chains is being fed into the system in near real time so we can see what inventory is in retailers’ stores and distribution centers, in transit and in our facilities, along with what’s being sold through the point of sale systems.
Millions of data points are fed into our program and, over time, it learns patterns and behaviors that can identify a supply chain issue before it happens – down to the SKU level. Once the problem is identified, one of our team members analyzes the situation and decides on the appropriate mitigation.
How would that work in the real world?
If we see a particular color taking off at retail, meaning consumers are buying it at a much higher rate than what was planned, the algorithm will recognize that and suggest a shift in production. The system is sifting through millions and millions of data points – something a human could never process in the required time frame – and drawing connections across the chain to identify potential issues.
What’s next on the horizon for the analytics team?
What we just described is a self-learning predictive model, meaning the machine is predicting that we are likely to have a problem. What’s in development now is moving it to prescriptive models. In this scenario, the machine would not only predict a potential issue, it would also recommend and implement the appropriate actions to proactively alleviate the issue.