Monthly Archives: February 2013

Better Supply Chain Decisions through Data Analysis

“Should we cancel the purchase order? We need to know today – and we can’t be wrong on this.” Only two-thirds of stores had reported any sales data from a newly launched product, and my team needed to know whether or not to invest tens of thousands of dollars in additional replenishment inventory. We already had a very large number on order, and based on pre-sales forecasts, we needed nearly double what we had coming. But with just a few data points, was it possible to tell whether we needed the stock or not? It all came down to the data: Could it be trusted? Were my assumptions correct? Was there enough to act on?

Soon after finishing business school and starting my career, I was quickly surprised by the contrast between the discussions that took place in the classroom and in conference rooms at the office. In academic case studies, my classes would look over graphs and charts to find important business lessons that the professor was helping us discover. In the real-world however, emotions, hopes, politics, and persuasion often make decisions much less clear than a business school case study. The small business I work with often relies on its supply chain team to make many decisions that require extensive data analysis. With moderate experience with excel and databases, my team has been able to slowly help our company make better decisions by taking out the myth of emotion and replacing it with the confidence of data.

Data-driven Analysis

Forecasting, inventory, purchasing, logistics, and process improvement are often done by gut-feeling in very small companies. When you don’t have the systems or people to gather and process the information, making your best guess is often all you can do. When operations are small, this works most of the time because you are able to get a feel for most parts of the business since you’re involved with most parts. But as the company grows, staying connected with each part of the supply chain becomes increasingly difficult. If you haven’t already, this is when you must switch from trusting your feeling to trusting your numbers.

There’s an excellent quote that I like:

“In God we trust. All others bring data.”

-W. Edwards Diming [Source]

It highlights that no matter how much we trust our intuition about a decision, numbers are often what really matter.

Which Data to Use?

One of the biggest problems with data in modern systems is the sheer size of the information you collect. If academic case studies were 200 pages instead of 20, schools might focus more on training students how to sift through the noise to find the important data. However, until then, experience and past trends are the best guides for making decisions.

With so many numbers in your system’s database, each department or side of a decision can often build a data-backed case for why their solution is the best. It then becomes important for the head decision maker to be able to judge which information is most relevant and accurate. He or she should start by asking the following questions:

  • Where does this information come from?
  • How was the data collected?
  • What are the assumptions being made?
  • Why do the two (or three or more) data points show different trends?
  • How could these conflicting results actually be pointing to the same conclusion?
  • What numbers can we all agree on?

By agreeing on the same assumptions and the pertinent data sets, you can better reach agreement on how to move forward with the information you have.

When the Numbers Lie

Sometimes, the numbers tell a different story from reality. Even if all analysis points to buying more of product,  only you can tell that the product will soon be discontinued, and that you don’t need more stock. Often data analysis isn’t so much about manipulating the pivot table, but adding the pertinent information that the system doesn’t cover. Adding everything you know that the system doesn’t, and then looking at the data all together often helps you make the wisest decision.

Highest and Lowest Case Scenario

“It’s only about two weeks of data, but it’s enough to make some decisions,” I said.

“I’m just not sure if we can trust it yet. There’s so much we don’t know,” a sales analyst replied. There were still many questions that we couldn’t answer. How many stores had yet to receive the product? Could the product still be in the back room? What if employees had bought the product instead of customers? I needed some way to make the 10 days of sales data point to something – and I needed enough confidence in the numbers that everyone could trust them.

I looked at the data again, and I wrote out what assumptions I could make. To reach my averages, I assumed every possible reason sales could be artificially low was actually taking place. Based on these high assumptions, I determined that average sales per store would likely be less than 3 per week. There were enough stores that had initial sales data that I could confidently say sales would very likely be between a high of 3 and a low of 1 per week. This was significantly lower than the 10 per week forecast for which we were about to place a purchase order.

“The data is convincing – and even if the high boundary number doubled to six per week, we’d still have enough without this order,” the brand manager replied. “Ten per week just isn’t a realistic expectation. Let’s cancel the order.”

By giving a confident range of average sales, I was able to help my team make the right decision and avoid large overstocks of inventory – something we have continually struggled with throughout the company’s history. Sometimes there isn’t quite enough data to make a definitive argument. In cases where you have even some data, using high and low assumptions can often give you enough confidence to move toward the right decision.

We’re not robots – and we shouldn’t focus solely on numbers. However, by incorporating more data into our decisions, we can often find better and more predictable results. The key is to know what data is useful, opposed to noise, and how to use that data correctly. Looking at the same, accurate data, and deciding on high and low assumptions are strategies that have helped move our company in the right direction.