Data Platforms, Data Culture

Organizing Talent: Return of the Data Center of Excellence

data center of excellence

Shane Murray

Shane is Field CTO of Monte Carlo. Previously, he served as the SVP of Data & Insights at The New York Times.

Will Larson (writer of An Elegant Puzzle – recommended read) may have said it best when he wrote that one of the best kinds of reorganization is the one you don’t do. 

However, data leaders inevitably reach a point where, due to team growth or evolving business demands, things just don’t work. Faced with these challenges, data organizations may swing back-and-forth between centralized vs. decentralized organizational structures until they achieve the right balance. 

Overly decentralized teams are plagued by redundancies and uneven standards, while centralized teams are often looking to become more agile and aligned with the business. It  can feel like a devil’s bargain, trading one set of problems for another.

So if the best type of reorganization is the one you don’t do, the second best reorganization is the one that leverages a data center of excellence and central data platform to mitigate the risks inherent with both extremes. Data teams at Freshly, Sanne Group, and others are deploying data centers of excellence.

In this post, we will review the most common data team organization structures and then dive into why data centers of excellence are becoming cool again, and best practices for deployment.

What are the common data team organization options?

Central data squad car
  1. Central data squad

The compact car– This is usually the best starting point, especially for small teams, as it connects the work of the data platform engineers to their most important partners, the data analysts and data scientists.  Data leadership can prioritize the most critical structural work, such as migrating to the cloud or enriching core product data, while unleashing teams of analysts and data scientists on the most impactful projects across the business. 

Center of excellence, often a hub and spoke
  1. Center of Excellence (often a “hub & spoke”)  

The best mid-size vehicle – Once you have critical mass, data leaders often deploy individuals or teams into an embedded structure within business units or product teams (“spokes”), while managing the work standards, skill and career development via a center of excellence model. 

Fully decentralized data team organization
  1. Fully decentralized

The SUV – This is common in large enterprises, where business units have historically hired and deployed specialized analytical talent. A data platform initiative will commonly require a significant shift away from siloed operations in order to unlock new business opportunities from interoperable data, commonly framed as a data mesh or data fabric

Data platforms and the rise of the data center of excellence

The driving force behind the data center of excellence is the power and extensibility of the modern data platform; not just the data warehouse, but the foundation on which your data teams can build analytical data products, provide self-service access to data, run experiments, and tailor product or marketing experiences. 

While historically data platform centralization was often promised but rarely achieved, in the last five years the investment in cloud data platforms has made centralization a reality and a competitive necessity. 

Delivering insights or models on business problems commonly requires analysts and data scientists to traverse multiple domains, such as making trade-offs between competing objectives or revenue streams. Point solutions for different branches of analytics are now only considered if they integrate neatly with a central data platform.

In parallel to the emergence of the modern data platform, many organizations have moved towards a Spotify-esque operating model, deploying data experts within objective-driven squads alongside PMs, designers and software engineers. 

Spotify’s organizational model via Functionly.
Spotify’s organizational model via Functionly.

While this structure delivers speed, focus and domain expertise to each squad, an analyst running an experiment that optimizes the number of ads on a page will likely want to navigate data from multiple sources (e.g. ad tech, web/app engagement, CMS) and collaborate with experts across teams to deliver a complete analysis. 

Talent optimization is also a consideration. Even with layoff news reverberating through Silicon Valley, overall unemployment levels are at historic lows and hiring is still a priority for data teams. Too frequently, there just aren’t enough amazing data analysts, scientists or data engineers to go around and be embedded in every business team. Data centers of excellence can operate like SWAT teams this way to deploy top talent where there is the largest need and most urgent priorities.

These trends have shifted the momentum behind the data center of excellence model, making it the smart choice for most growing data organizations. But this model is not without its own risks or pitfalls. Here’s how to not screw it up.

Making the data center of excellence work

Embed teams not individuals

The logical outcome of the data center of excellence model is to embed individuals across product and business teams, increasing the velocity of the collective. But how do you avoid the pitfalls that come with an analyst getting swept into a team that has two pizzas worth of engineers, plus designers and product managers, who may not see the value in data projects that go beyond the scope of their team. 

The first answer is to deploy teams, not individuals. Those analytics teams can flex towards important problems; they can work together on data products that serve more than one of their partners; and they can benefit from shared ownership, collaboration and code review. 

The second answer is to embed only once expectations, tools and standards are sufficiently mature. Embed too early and your stakeholders might be happy, but when the embedded talent eventually leaves most of the work product goes with them. 

While I was at the New York Times, we re-centralized the data science team (under the data platform umbrella) in order to focus on the most valuable product problems, empower data scientists to own the problem space and build teams, and identify where solutions (e.g. feature stores) might be shared across projects. 

Centralize critical data products in the platform, but allow some flexibility

Data products can be both centralized and decentralized

Often data teams are considering a binary choice, either they continue to manage all data assets centrally or they recognize their capacity limits and opt to subdivide the business into domains who will own their respective data. 

The pragmatic middle ground is where most data organizations land, with critical data products used across multiple domains living within the platform hub, while domain-specific data products may be wholly owned in the spokes. The data platform group is thus a mix of platform and product – sorry purists.

Use your data center of excellence to build momentum for federated governance

Data steward programs have typically been used to support data governance initiatives across a distributed organization, but do they work? Stewardship has sometimes felt like a part-time commitment to critical priorities like data privacy and data quality

Within a platform hub, data teams can define standards and policies, proving their effectiveness on the data assets that are managed centrally. For example, if you intend to enforce data SLAs, then the initiative should start with the shared data managed by your platform teams. 

Beyond the platform hub, data center of excellence leaders can then make use of the embedded teams within “spokes” to export standards and policies across the organization, making it a strategic priority for the enterprise.  

Clawback your teams on the regular

One strategy I’ve heard from larger, more decentralized data teams that prefer to maintain smaller central teams is to “clawback” their teams on a regular basis. 

For example, one data lead discussed how every member of the data team is expected to attend a weekly session to help pay down data debt, contribute to centralized initiatives, or work on developing the central platform capabilities. This also helps maintain the larger data team identity and culture.

Excellence at the core

Will Larson also wrote that the next best type of reorganization was the one that solved a structural problem. 

Whether you are a smaller team experiencing hypergrowth and looking to get closer to the business, or a Fortune 500 organization that has operated in siloes since the advent of the database, a data center of excellence can help you maintain the competitive edge your organization needs.

-Shane Murray


Curious about how data centers of excellence and data observability initiatives can improve your team’s efficiency and data quality? Fill out the form below to schedule time with us.

Our promise: we will show you the product.