Lessons Learned from a Data Analytics Project Manager

A couple of years ago I was given the opportunity to lead a large data analytics project related to using machine learning and natural language processing to gain business insights into our equipment maintenance strategies of over 2 million world-wide equipment.

The ask was simple: figure out whether or not we were performing the right maintenance on the right equipment at the right time. And if we weren’t, use the 10+ years of maintenance, spend and reliability data to improve the maintenance strategies. Sounds pretty straightforward, right?

The prize was significant: potentially tens of millions of dollars of saved money annually by eliminating or extending ineffective maintenance tasks and focusing on more targeted, condition-based maintenance leading to improved equipment reliability.

Unfortunately, the journey was bumpy: I had zero experience in big data analytics and this new world of data science, making the next year more about failing than succeeding. Since we were pioneering the use of data analytics for the first time as a company on this scale and in this part of the business, I learned many lessons during that first year that I believe only failure could have provided. I would like to outline the top 4 lessons I learned in hopes that other project managers can use my failures as jumping off points to faster success in their data analytics projects.

Lesson 1: Get Directly Involved in the Process

I started with a team of 2 data scientists and a heap of disparate and unstructured data. With our goals set, we began down the path which we believed would be short and direct. I was understanding the business value drivers for the company and laying out the tasks for the data scientists to accomplish. I spent my time painstakingly outlining the project roadmap with specific phases and tasks without any regard to understanding the underlying data quality myself. This led to me assigning work to the data scientists that was sometimes impossible given the current state of the data with respect to quality and completeness. Once it became apparent that I was assigning impractical tasks, I spent less time managing the project roadmap and more time face to face with the data scientists actually going through the data together. I even became their go-to data validation engineer since I had the domain engineering expertise, which allowed me to further understand the data, how it could be used and, more importantly, how it couldn’t be used. The result was that my tasks for the data scientists became more targeted and on par with what the data quality would allow, leading to more realistic and achievable goals. I believe a project manager should spend this time in the weeds early on to establish practical goals for the project.

Lesson 2: Implement a Fail Fast, Fail Often Approach

Failures seem to stick in our memory a bit more profoundly. Maybe it’s because of the stigma of failure, the embarrassment of it all. But we all learn better from failures than success and the faster you get to a failure, the faster you will ultimately find the successes. When pioneering something new, you have 2 choices: 1) You can take the time at the beginning to plan out exactly how you will achieve your goals and then attempt to will your plan into submission or 2) you can start with a more arbitrary plan and accept that you will fail your way into the best path forward. I initially began the data analytics project with the former, leading to weeks wasted on attempting to stick to my original plan for the project instead of allowing the data to dictate our path forward. After working with the data scientists directly, it was becoming clear that it is more beneficial to conduct miniature “studies” that test whether or not a particular machine learning algorithm will work as expected. We hired validation engineers whose sole purpose was to validate the machine learning outputs of these “studies” to provide us quick insights into the efficacy of our current plan. When we were not getting the results we wanted, we pivoted onto the next plan. This agility allowed us to react quickly to failure and ultimately find a successful path forward.

Lesson 3: Connect Each Task Directly to a Business Need

I believe team productivity is optimized when each team member understands how their particular role and daily tasks fit into both the projects goals as well as the company’s goals as a whole. This was especially true of the data scientists who were deep into the data every day and therefore benefitted from an understanding from a 10,000-foot view. I laid out each task for the data scientists not only with respect to what we are trying to achieve with the data (such as developing a natural language processing algorithm to determine the failed item from a maintenance description) but also why we need to develop this algorithm from a business point of view. To continue with the previous example, understanding what failed on an equipment gives valuable insights, to the business, into what is causing decreased equipment reliability and cost to the company. I believe a project manager needs to provide clarity to each team member on how their tasks help achieve a greater goal for the company to ensure that produced work is pushing the project forward and not causing rework due to a lack of clear goals.

Lesson 4: Show Business Value Along the Way

One of the biggest challenges I was faced with while managing this project was to balance the overall goals of the project while simultaneously keeping management and other key stakeholders engaged by showing them small examples of how data analytics can add value to the business. Again, this was a brand-new project for the company, therefore there was a level of trust from management that all of this work was going to amount to something of value. Therefore, when management is writing checks to keep the project going, it is best to ensure you prove to them that the value is, in fact, tangible. We took on many ad-hoc requests to provide a glimpse of the value we could add on a larger scale down the road while also solving smaller problems in the interim. We did mini-analyses and give recommendations, we built insight tools and we gave demonstrations. The combination of these activities kept management and stakeholders satisfied with the progress and excited about the future. Looking back, suffering a reduction in project progress in order to showcase small but substantial value-adding wins along the way is a tradeoff I’d gladly take again.

In conclusion, the failures I suffered as a project manager gave me a clearer view of how to succeed in future projects. These failures were difficult to take on in the moment, but an understanding management fueled by the excitement of pioneering something potentially game changing allowed the project team to experiment, learn and grow successful in the long run.

As a data analytics project manager: Understand what is realistically achievable by getting directly involved in the data, fail fast and often to find success quickly, improve team efficiency by providing clarity on how work ties into business goals, and foster stakeholder excitement by showing value along the way.

I hope that these lessons will enlighten you in your journey toward successfully managing your next data analytics project.

References

Image link: data-analytics-trimmed-650.jpg