I am currently getting my MBA and Masters of Engineering through MIT’s Leaders for Global Operations (LGO) program. Every year, the program takes the fellows on a Domestic Plant Trek where we get to see manufacturing and operations facilities across the US. This year, we visited companies like Amazon, Boeing, Rivian, Nissan, Caterpillar, and other plants in cities like Phoenix, AZ, South Bend, IL, and Miami, FL. As someone who is focusing on digital technology, my goal on this trip was to see how I could relate physical operations to data operations. This post will detail some of the key similarities and lessons that can be applied across (data) inventory management, building robust processes, and managing people.
A common trait in all of the efficient factories that I visited is that the mechanics didn’t waste time by leaving their station to go look for the tools they needed. Their nuts and bolts were stored as close as possible to their workspace. This was referred to as destination-based storage where inputs are accessible where they are needed and not necessarily where they are created. In the data world, Extract, Transform, and Load (ETL) processes are used to migrate data from where they are created to where they can be used. Imagine how tedious it would be to have to query Stripe’s API every time you want to run a revenue forecast model.
Although implementing ETL is a sign of a mature data organization, don’t overdo it. In one facility, there was so much inventory that workers had a hard time finding what they actually needed. In another, there was so much unused inventory that some of it had a risk of expiring. It’s incredibly difficult and tedious to find the right information you need when it’s surrounded by a lot of noise. Therefore, when planning out an ETL pipeline, start off with only what’s necessary today and expand later if needed.
Boeing has this concept called Foreign Object Debris (FOD). There were signs and employees everywhere reminding workers to keep workspaces clear of FOD. Why is this important? Well, FOD is a tripping hazard. Also, having too much FOD could increase the risk of some of it getting into the final product. Can you think of any FOD your data systems are producing that could be cleaned up?
Perfect the simple, manual process first before introducing advanced tools like artificial intelligence and automation. During the trip, I learned that Amazon operated some of its physical retails stores with absolutely no AI or fancy technologies. They focused first on defining their processes and making sure that the baseline product worked. Then, they automated the micro-process that would yield the highest benefit. I know buzz words like machine learning, neural networks, and minute-by-minute automatic testing sound great, and they might get a few nods in board meetings. However, implementing these tools without first understand the underlying process can lead to unnecessary headaches (and in my case, unnecessary all-nighters).
The top factories focused on where they could provide the most value and didn’t waste resources trying to build everything in-house. Do what you do best and outsource the rest. Think about how you can double down on your organization’s strengths and utilize pre-made or external sources for everything else. If data visualization is your bread and butter, maybe you can use open source or external machine learning model providers. Or perhaps, you can use pre-made data aggregation and ETL tools instead of trying to build everything by hand.
One of the smaller facilities that we visited is currently undergoing major improvements every week. On Fridays, the manager reviews data collected throughout every step of the manufacturing process to see where the major bottlenecks are. Then, during the following week, they solely focus on improving that one bottleneck. Healthy data operations have mechanisms in place to catch, measure, and prioritize data quality issues, application downtime, and process bottlenecks. You can’t fix what you don’t know.
What surprised me is that in a few facilities, all of the workers were cross-trained so that if someone on a different line was out for the day, someone else could step in and do their role. I’m definitely not saying that all of your analysts should be trained data scientists. However, I do believe that a data organization should lean more towards having “T” shaped people and not pure, isolated specialists. Everyone should know exactly how their work affects the entire pipeline and should understand who owns their inputs and who consumes their outputs. They should be empowered to view and learn about (but not interfere with) what other teams are working on and how different jobs are completed.
The biggest piece of advice many managers gave us was to take care of the frontline workers - they should be safe, motivated, and empowered. The main reason is that this is just the right thing to do. The second reason is that hiring is hard, training is tedious, and both can be expensive. In the digital space, safety could mean providing a safe environment for junior people to speak up and maybe even challenge senior management. Motivation could mean making sure everyone understands how their work feeds into the larger organization. Empowerment could mean giving everyone the right tools to do their job on day one so that they don’t need to wait four weeks to get access to one database.
In many of the smaller facilities, the biggest obstacle for scaling wasn’t money. It was having the processes documented well enough so that newcomers can quickly become productive. If you’re a leader in a data organization, imagine that your entire team quit today and that you had a fresh batch of new employees joining tomorrow. How freaked out would you be? If not at all, congratulations. If you’re sweating just thinking about it, take a second to think about what applications, processes, or knowledge for your team are the most obscure and spend the next few days putting together a game plan to improve your knowledge management.
This trip showed me that manufacturing operations and data operations aren’t polar opposites. In both worlds, inventory management, robust processes, and people define a healthy and effective organization. I hope that you were able to connect some of the lessons learned in physical manufacturing that can be applied in the digital space.