Five Lessons from Walmart's Innovative Approach to Network Design

Walmart won this year’s Edelman Prize.

Their project was about transforming their supply chain (which was already good) with network design, simulation, and load planning. Their overall solution was impressive¹.

I missed it live but just watched the video. The entire presentation and project is impressive. But for now, I’ll focus on the innovative ideas and lessons from the network design.² Here are some lessons for anyone doing network design:

First, the coolest innovative idea was to build a second model to determine how to transition to the new network.

Of course, this idea isn’t completely new. But Walmart set a new standard.

The result of the first model showed the ideal network ten years from now. This became the input to the second model. Then, the objective of the second model was to determine how to get from the current to the final state.

This second model determined when facilities should open, when they should close, and when facilities should be retrofitted. It showed the capital outlays needed by time period (and I assume that this was also a constraint) to help build the business case. And the model made sure that the products flowed smoothly to the stores.

They used this model to build a detailed business case and implementation plan.

Free bonus lesson: This reminds me of something I want to research more: adding constraints to the model to the model that limit the number of changes. That is, you add some additional variables to the model that define a change. A change could be big things like adding a warehouse or something smaller, like making a product at a different location. Then, when you run the model, you limit the number of changes. Or better yet, you run a multi-objective model with changes as one of the variables. The tradeoff curve then shows how much value you get from every change.

The hard parts, in my mind, are defining changes (are some things too trivial to count, do big things matter more than small ones, etc.) and not overwhelming the model (if I’m not careful, the model won’t be solvable).

The second innovative idea was to link the strategic design to load planning (loading and routing the trucks). In parallel to the network design, they built an excellent load planning system (that is for another post). Finally, to link the strategy to the execution, they built a simulation model. The simulation sat in the middle, looked at the tactical capacity issue, and allowed the team to iterate through all three solutions.

Interestingly, they claimed that when they iterated through different scenarios, they could find optimal solutions for the system but not for any one component—a good reminder of all the trade-offs in a supply chain.

The third lesson, and a reminder, is that visualization matters. I would guess that Walmart budgeted billions of dollars for this transition. They had to show the senior leadership hundreds of scenarios to approve this. Standard output reports and slides weren’t going to cut it. So they built a custom visualization tool that allowed drill-downs and had animations showing transition plans.

This is a good reminder that we need to show our work.

The fourth lesson is that network design models are large. This is a good reminder since I’m overly optimistic about solvers getting faster.

Walmart discussed that if they combined the first and second models, it would be 100 million variables. However, they determined they could only solve models with about 1 million variables. So they had to create two models, and within each of those, they had to work hard to reduce the size.

Most supply chains won’t be as big as Walmart’s, but if you model too many details, you can easily get up to 10 million variables, even for relatively small supply chains.

I’ve found that pushing the limits on model size doesn’t get you much value and the longer run times may hurt value. That is, long run times kill the desire to run scenarios. And running lots of scenarios provides value.

The fifth lesson is that these projects take a long time. I’m guilty of forgetting this lesson. Walmart’s project took multiple years. Most projects shouldn’t take that long. But four months slip by quickly if you are doing a large transformational network design project. Give yourself time.

The counterpoint is that you can quickly answer specific questions if you create and maintain supply chain design data and frequently run design projects. And you can drastically reduce the time to do this type of work.

The video is worth watching. They are doing great OR work.

1. I loved seeing their CEO discuss the value of Operations Research (OR) and Analytics (at the end of the video). That is good visibility!

2. The main network design discussion is from about 11:10 to 17:16.