Highly effective data teams
A list of seven habits
This post captures the essence of my experience building teams at various companies. The title and outline are from the book by Stephen Covey.
Effective data teams are proactive about infrastructure decisions, about customer data protection, and about the design of their organization.
Begin with the end in mind.
Effective data teams identify, firstly, the business impact of of their work.
When doing data analysis, they deliver insights not tables, action items not dashboards. The analysis of data is not complete until a conclusion is drawn from all the generated tables and graphs. A table belongs in a database. Graphs sometimes reveal only noise. Presenting tables and graphs puts the burden of inference on the stakeholder. Inference by looking at visuals is having to run stat-sig tests in your head. The data team is in the best position to make sense of the data, to tell its story, and to determine if a change is worth acting upon.
When embarking on an ML project, effective data teams identify the goal of the model and how it is meant to impact the user experience. They use this understanding to determine shipping criteria in the context of an experiment.
Put first things first.
Having built confidence that the underlying data is correct, effective data teams prioritize thoughtful metric design.
Metrics heavily summarize and simplify reality — disposing of its many details in order to handle complexity. This highly simplified version of reality is used to inform decisions. Business processes and ML models are built such that the value of a metric is maximized or minimized. It’s an important responsibility.
Metric design is an opportunity to build relationships with stakeholders, review data quality, and understand the business. The success of a data team relies upon the relationships, trust, and intuition the team gains, and the quality control processes they implement in this part of the work.
Effective data teams are the voice of the customer. A successful business needs happy customers. Low quality releases, short-term wins, and bugs the team can get away with, decay the product and consequently the brand.
Seek first to understand, then to be understood.
The data team does not operate independently — just like every other team. The work is useful if it’s rooted in what the company can accomplish. Effective data teams understand the capabilities of their cross-functional collaborators in acting upon the results and recommendations.
Effective data teams are in sync with all cross-functional collaborators. The team culture is such that collaboration is prized over competition. Teams achieve more than individuals working independently.
Sharpen the saw.
Effective data teams hire for experience to help establish best practices, provide training, and to solve complex problems.
What practices have you found useful in building effective data science teams? Let me know!