Data Science for Business: Lessons Learned While Building a Data Team

By Gpaulin F

I joined the RappiPay team back in 2020 after leaving a very different organization as Director of Artificial Intelligence, and coming into RappiPay was a massive challenge when it came to data. Like all new organizations who are putting together a data team (yes, RappiPay Mexico was brand new at the time), struggled with role misconceptions, and Data is no different. Even though I was originally hired as Head of Data Science, I was promptly given Business Intelligence and Analytics responsibilities because of my close relationship to data and because there was no one else doing this for the business. After a couple of months of providing value to the business through Data, I faced another challenge: there was a huge disconnection between the data that was being produced for the business and the needs of the business itself. So, I took care of the Data Engineering team and started managing the backlog for data acquisition and processing for the business.

The following lines gather many lessons I learned the hard way while putting together a data team for the organization:

Become a Data Evangelizer

Not all organizations are used to data or insights. Therefore, if they are not using the data you (or your team) put together for the business, it is also your duty to show them how they can exploit it to make their jobs easier and better. Sometimes it gets as easy as a simple one-liner explaining the purpose of the data, and sometimes you’ll need to explain basic probability or statistical concepts. Roll your sleeves and get ready to explain as simple as possible as if you were a 1st-grade math teacher.

The Business Units Will Need Data to Operate

At RappiPay there are multiple business units (risk, fraud, growth, customer service, etc…), and the business units team members are the ones who know better which data they might need, which concepts will they use, and which key performance indicators they will be measuring when becoming operational. So, take your time to map their needs and concepts, and facilitate data to them the way they need it, not the way you think it is better. After all, they will be the ones responsible for such indicators.

Automate as much as possible

There’s nothing more soul-breaking and time-consuming than recurring tasks. These tasks take the creativity of your team and throw it into the trash can. Therefore, invest time in analyzing which tasks are recurrent and talk to your team and their stakeholders about how such tasks can be automated. For example, reports could be sent automatically through DataViz tools like PowerBi, Tableau, Spotfire, etc… If larger data is needed, such tasks could be automated using data flow orchestrators, such as apache airflow, Luigi, etc. This investment will save you and your team a lot of time.

Maintain your work or kill it

If you release a report, table, view, or any data, the business leaders could go on and consult periodically, maintain it, or kill it. Otherwise, it could be catastrophic if the data sources of the dashboards are not maintained and they use the indicators for a report. Avoid giving the business units non-updated or non-maintained data. It is helpful to spend some time every sprint doing this, it will avoid headaches.

How to put together a Data team without dying in the attempt

Putting together a Data team is hard when there’s very little built and so many things you could potentially start with. Therefore, the first thing that should be done is to map the business needs and project how the business will scale in the time to come. If your business is scaling up as Rappi, then plan for the next 6 months as if you were planning for a year. You might even fall short.

Make sure your future data team will be capable of tackling the problems you face regularly by developing custom-made challenges and questions for the different positions you’re recruiting for.

Then, start with the basics by hiring Data Engineers who will be responsible for collecting the already mapped data, processing, and ingesting it onto your data infrastructure, whether it’s a data lake or a data warehouse. Keep in mind that data quality is extremely important, especially when it comes to customer data. Therefore invest some time in data quality before it’s too late.

In parallel, the data analytics team could start growing so, as soon as the data comes out of the oven, they can take it and make a great impact on the business through DataBiz tools. Don’t forget to evangelize the rest of the business team on your creations, and soon you will see they will start taking your dashboards and indicators to drive their businesses. You will only take your dashboards to the next level if you don’t forget to keep the business always in the loop. The analytics and B.I. team should be the closest to the business, therefore never stop iterating with the business leaders on the indicators, definitions, and dashboards you’ve built.

Finally, don’t jump right away into hiring data scientists or machine learning engineers. This is very tempting, but if your data is not yet ready to be consumed, hiring them could be a waste of talent. Therefore closely monitor your data health status and once you’re close to being ready, start bringing in the data science talent, as they will need some time to get used to the data.

No team will ever create an impact if the business needs are ill-mapped. Thus, if you’re the one leading a data team, pay close attention to the business needs, map them to the data, and try to visualize how, by automating the decision-making process, your business could save either time or money. Discuss your findings with your data science team, and build a minimum viable product (MVP). Once this is created, carefully put it in production to be consumed. Make sure it’s performing as expected and monitor your model’s health by creating dashboards to visualize your data input (e.g., how many of your input variables are empty) and the data drift, which is how much your data has changed since your model is in production. The former will help you to monitor if there’s anything wrong with your data input and the latter will be an indicator of when you should retrain your model.

Final thoughts

There’s no universally accepted rule when it comes to working with data, nevertheless, a recommendation I can spare is that, regardless of your knowledge or experience with advanced algorithms, if you can’t create something useful to the business then you won’t create an impact through data. Therefore (and I can’t stress this enough), keep the business leaders in the loop. Something that has worked for my data team is to have one session with all the business leaders before starting a sprint, this helps us to map the business leaders’ needs, refine the need, and avoid working twice on the same requirement. These sessions also help us to discuss the business’ priorities with the rest of the business leaders and only work with the top ones.

Finally, no one gets to a company knowing everything. I had great luck finding amazing mentors at Rappi that I consulted very often on different topics I needed to solve. I highly recommend you to find yours in your company.

Good luck driving your data team and your business to the next level!