Menu Close Log in Get started

How we used OKRs for the data science team to improve our data stack

Data Culture

“Can you give me data for X?”

A question Data Scientists might receive quite often - as so did our team.

Even though, in most of the cases, the data people are asking for is not what people actually need, it was a sign that something was odd.

We realized that one of the reasons we received those questions was that we failed to implement the basis for self-service analytics. We needed to face the truth and acknowledge the problems at hand:

  • We did a bad job at educating people on how to use and analyze data;
  • Combining data for analysis was not possible for non-engineers or people without database and SQL knowledge because of bad integrations between different data sources;
  • Data Science became the bottleneck for data driven decisions and were even preventing them when we were too slow;

Enabling self-service analytics with Data Science OKRs

A strategy shift towards more data accessibility and enabling self-serving analytics needed to happen and we needed a framework to properly align the team and be able to measure our progress. During that time our company started looking into OKRs (objectives and key results).

Our Data Science team wanted to become pioneers and give the framework a try. The key idea is that you define ambitious goals which can be measured by 3-5 key results each.

We set together and analyzed the current situation to derive objectives and key results. Our goal was to lay the foundation for bigger steps in the future and start off setting up a data warehouse structure with pre-joined and processed tables which are easy to analyze for our business users. Thus, an example objective we set ourselves was “Data is accessible to everyone in the company”.

Using Metabase dashboards to track OKR progress for data science

This included the following key results we used to measure our success:

  • All new questions in Metabase only use the data warehouse;
  • Questions in Metabase are stored in a clean structure;
  • All warehouse tables have a description;
  • All columns of tables in the warehouse layer for business users have a description;

Those sound like pretty basic key results at first sight. However, they turned out to play a key role in measuring basic accessibility.

We created a Dashboard in Metabase to measure our progress and look at the data during our weekly OKR progress check-ins.

a screenshot from Metabase where OKR are used

Despite having still a long way to go, we saw a great impact on our work already. One example you can see in the dashboard is that the newly created questions almost exclusively use the new warehouse instead of the old database now. Especially the team alignment through the OKR framework helped us improve our data stack and to take huge steps towards enabling self-serving analytics in our company.

Contributed by
Thomas Schmidt

Thomas is the Lead Data Scientist at Agrando, a company that makes regional agricultural structures fit for the future with the power of digitization. You can find Thomas on LinkedIn and on Twitter at @somtom91.

photo of Thomas Schmidt