How Data Scientists Make Impact

• Read original article here

Couple months ago, I left my job as a data scientist at FAANMG. I’m starting my new data science adventure at another company in a few weeks, but I won’t talk about that today. As I transitioned to my new job, I wanted to reflect on my previous one and write down important lessons that my last team and organization taught me. In this article, my hope is to pass on my knowledge to any incoming data scientists or aspiring data scientists to help them be more aware of challenges and pitfalls that await them. To start off this series, I wanted to write about two learnings: (1) the effects of a non-data-driven culture on data scientists (2) how data scientists can lead change to a data-driven culture. One would think one of the top tech firms must be data-driven, but just like people, no company or team is perfect. Also, I only worked for one team at Microsoft, so this is certainly not a testament to all teams across Microsoft. With that said, I believe data scientists are leaders when it comes to driving data-driven culture, so it’s our responsibility to be aware of these challenges and strategically tackle them as they come up.

Note: This is absolutely by no means a criticism of my prior team at Microsoft. These were just some of the learnings from my last job. After all, I do believe creating a data-driven culture is led by not just the leadership, but also by the data scientists.

Microsoft grew up from a ship-first mentality. This means you would first launch a product and then try to analyze the impact of the product’s success afterwards. Data scientists were only afterthoughts of product teams’ innovative idea and engineering’s implementation. However, this method of analyzing impact is very difficult and can lead people to think correlation = causation. This type of process results in making it very difficult for data scientists to deliver insights that drive impact as well as analyze the impact of new product releases. Therefore, it is key for data scientists to plug themselves in the product cycle from the beginning to the end and align the product team’s strategy to data team’s strategy.

Data scientists must not silo themselves with the data team, but try to become part of the product team. Data scientists cannot provide value when the product team is not invested in what they are working on. For all data scientists and leaders, get to know all of your product team members and try to learn about their past and current projects. I like to apply the golden rule here: “do unto others as you would have them do unto you.” As you begin your new data science project, ask what all the product team members (user research, design, engineering, product management, marketing) would be interested in knowing. Proactively update the team on the progress and give them insights they might be interested in knowing that could help you gain a new perspective on the project. Be transparent about the project and convince the product team that data science can help product innovation and positively impact business KPIs.

Learning how to use visualization tools like Power BI is not as always self-explanatory for people outside of the data team. It can be viewed as an obstacle and an additional responsibility. Who would want more responsibilities outside of their own? Data tools are just one part of the story. Often, metrics and methods of analyses are sometimes difficult to grasp. As data becomes confusing and tedious, product teams will begin to ask data scientists the simplest questions. Although these questions are simple, data scientists can spend up to half their day or their day answering ONE question. This throws off data scientists’ schedules and delays their other important data science projects that have the greatest chance of impact. Data scientists need to empower teams to leverage data to answer simple questions on their own by strengthening the product teams’ data muscle.

Data scientists must become teachers and provide clear instructions on how to access and use the data tools and present the team with thorough descriptions of any key metric. Whether it is having a glossary for metrics or scheduling meetings dedicated to learning, data scientists need help product teams become more data literate. From my experience, empowering the teams to use the data on their own can give product teams a sense of success and impact they have made on the team. However, this doesn’t come easy because people are busy and they may be reluctant to come to your brownbags about Power BI or experiments. Maybe instead of just sending a meeting with a general agenda, try to to provide incentives for the meeting. In addition, you can provide teams flexible mediums (i.e. Slack Channels, Office Hours) to come to you for questions.

Making multiple mistakes may lose anyone’s trust, and trust is a must between the data team and product team. Otherwise, a data team may lose the impact they can have on the product team and the organization. With all that said, no data has absolute certainty. There is always a caveat and there is always a bit of uncertainty included in any data pipeline or machine learning model. However, teams outside of data find all of this hard to believe and begin losing trust when they find doubtful numbers in work of the data team. Once the product teams lose trust, they will be reluctant to include the data team as part of the product lifecycle and lean on the data scientists for product recommendations. Therefore, data scientists need to be rigorous in their statistical findings and make sure that uncertainty is measured and well communicated to the stakeholders.

Data scientists must make sure to state any assumptions before any presentation and always check with their team to evaluate their data science work before communicating their results. Just like products, data codes often have bugs and helping product teams realize this early on can help trust to grow overtime. Providing transparency and implementing a rigorous method of checking the work of all data scientists is critical to a more data-driven culture.

There are many articles out there talking about how to create data-driven culture. However, I hope the readers were able to learn a new perspective on data-driven culture and the impact it has on data scientists. This is my personal take on how the lack of a data-driven culture impacted me as a data scientist and what I learned from it the most.