Data Science Staffing Models — a summary
This is adapted from ‘Models for integrating data science teams within organizations — a comparative analysis‘ by Pardis Noorzad. I wanted my own summary for quick reference. I hope you get value from it, too.
The Think Tank (Center of Excellence) Model
What it looks like: Data Science is its own department, analyzing the organization’s data and making independent suggestions for how to deliver more value. When it works: Data people have the time and space to think outside the box and create real innovations, without being distracted by the day-to-day concerns of running the business. When it doesn’t work: The think tank produces smart ideas and original analyses, only to be treated like a ghost by the rest of the org, who are busy “solving real problems”.The Accounting (Business Intelligence) Model
What it looks like: Data people are relentlessly focused on measuring the org’s True North KPIs. They make scorecards and canned reports, pushing them out to all who will listen. When it works: “What gets measured gets managed.” Having the measurement people focused on True North KPIs can’t help but produce an org aligned toward hitting those KPIs. When it doesn’t work: The most interesting insights hide below the surface. Focusing on top-line numbers misses the real stories happening closer to the ground, and misses the biggest improvement opportunities.The Customer Support (Consultant) Model
What it looks like: Data Science team is assigned “tickets”, there is a prioritization process that assigns data people to the hottest requests “right now”. When it works: The team can start small (as small as one person). There is no risk of building something that’s far removed from the real problems of the org because all requests come from real problems. When it doesn’t work: A data scientist with no subject matter expertise may need to be trained, or may just produce low-quality output. If you’re a data scientist who knows they won’t be working with this team/project long term, you feel no buy-in or ownership of the problem.The DIY (Democratic) Model
What it looks like: Data science team maintains a flexible tool, like Tableau, and enables everyone to create their own data dashboards, reports, etc., using whatever people and skills they have. When it works: Having the org’s data be largely transparent frees people up to find trends and new ideas that aren’t technically “their problem”, but are still worth pursuing. When it doesn’t work: Nobody follows data best practices, one team’s numbers can’t be compared to anyone else’s because there is limited attention/knowhow for data governance. Also, if data is everyone’s job, it often becomes no one’s job.The Embedded Model
What it looks like: For each business unit or function, there is an embedded data scientist responsible for supporting that business unit or function’s specific activities. When it works: Data scientists are close to the action and feel great buy-in and ownership. Each function performs better because they do their work without getting distracted by the “part time job” of measurement. When it doesn’t work: An embedded data scientist can have no career path, leading to lower retention. Having no peers to bounce ideas off can lead to lower quality output. Small teams that can’t afford a data scientist don’t get the benefit of support they would get under a customer service model.The Hybrid (Embedded + Independent) Model
What it looks like: Data scientists are embedded in teams AND belong to a data science team, with management and a career ladder. When it works: Data scientists focus on moving the needle for their particular team/job function, leading to long-term buy-in and ownership. Work product improves because data scientists also have peers to bounce ideas off. When it doesn’t work: It’s costly to maintain embedded data scientists on every team AND independent data science management. With two management reporting structures, communication overhead is heavy.Get more content like this!
More for you…
What great managers do to deepen their bench
I think most managers would agree: When a new project comes in, there are only a select few people you'd trust to run with it: your top performers. If you're like many managers, that's where your thought process stops. Once you've figured out who you trust to deliver...