Troubling trends in supply chain design: An uptick in bad models and collapsing teams

Congratulations, you’ve bought the software. Now what? Do you pick your best data analyst to build a model? Or do you focus on building an in-house center of excellence? Both options seem OK on the surface. But do they stand up to the realities of building and sustaining a supply chain design competency?

In my two-and-a-half decades in supply chain network design, I’ve seen two fundamental issues repeatedly:

1. Building solid, trusted supply chain network models isn’t easy
2. Most companies struggle to maintain a team capable of creating confidence with the modeling software in which they’ve invested.

The rise of bad models

The adoption of supply chain modeling tools has increased significantly over the last 20 years, and the tools are better than they used to be. Along with the orders of magnitude improvements in hardware thanks to Moore’s Law, the embedded third-party solvers have made significant strides, putting more power in the modeler’s hands. This additional solving power is complemented by advanced data handling and visualization tools, which contribute to the increase in usability of these applications. So, why aren’t we seeing corresponding dramatic increases in confidence in the outputs of these models?

Unfortunately, as a byproduct of broader adoption, we’re also seeing a corresponding increase in “bad models.” We’ve reviewed many different models, both external and internal. In a few of these cases, management simply lacked the confidence to implement the recommended solution and needed a “gut check.” Most models, however, had deep flaws in the approach, model design, cost or constraint development. Or they made key assumptions that caused errant answers that rightly didn’t build confidence in the results.

Insufficient experience in model design or analysis techniques (for example, transportation costing, capacity modeling, multi-period models or inventory optimization) is the main cause of bad models. To be clear, the user doesn’t typically cause these failures by simply not knowing the software or the tool not working properly. It’s inexperience with applying these tools and techniques to real-world problems.

Keeping your team together

The biggest hurdle in sustaining a team that can consistently deliver high-quality network analyses and answer what-if questions is a critical mass of talent. Given the current demand for these data-driven skillsets, turnover within these teams can be distressingly high. It doesn’t help that people who are good at this type of work tend to be good at lots of things. They often get pulled in many directions. Internally, it could be a promotion. Or they could voluntarily leave for a competitive role at another company. A sustainable team requires five or more people so it can withstand inevitable turnover. Five team members also provide good mentoring — and a meaningful career path — and a second, third or even fourth set of eyes to guarantee a well-developed model.

Large organizations — usually those with more than $15 billion in revenue — can usually justify big enough teams and attract and build the expertise required to support doing excellent work. Companies of this size can generate many complex business questions that quality modeling teams find engaging and rewarding. Organizations like Coca-Cola, Cargill and Starbucks have modeling teams of 10 or more people, and they work hard to maintain them.

In contrast, companies with less than $15 billion in revenue don’t fare as well. Teams at these companies are often one to three people and lack adequate peer review, mentoring, meaningful career paths and competitive compensation. These challenges can stifle a team’s ability to build confidence in their analytics. Small, fragile teams can often struggle to attract talent in the first place. Then there are the inevitable departures or promotions. It’s not unusual for teams of this size to collapse due to attrition or a lack of management support within the first few years.

Without stable peers or mentors to foster sound practices, smaller teams can create poor models and not even know it. This phenomenon has become known as the Dunning-Kruger Effect: “Many people … underperform simply because they don’t know that they could be doing better or [don’t know] what great performance looks like.”

It matters

Twenty-five years ago, we used supply chain models every one to three years to answer basic infrastructure questions like where the next warehouse should go. Today, we use these tools far more often — weekly or monthly in some cases. We also apply the tools to more planning areas, such as plant- and line-level manufacturing decisions, inventory optimization (including build-ahead strategies) and the deployment of private fleets.

Having confidence in your analytical teams and the models they build is one of the key governing constraints on our progress toward the vision of a near-real time “digital supply chain twin.”

— By Steve Ellet