Your Data Team Structure is Probably Wrong
By jason thompson
There are two dominant philosophies for how organizations structure their data teams: centralized and distributed. i've spent twenty years watching companies commit to one or the other, and i've come to a conclusion that might frustrate the people who designed your current org chart.
Both approaches are wrong.
Not partially wrong. Not "it depends on your company size" wrong. Fundamentally, structurally wrong. Each model solves one set of problems while creating another set that's arguably just as damaging. The answer is a hybrid model that includes centralized expertise and leadership with embedded liaisons in the business units you serve. Let me walk through why.
The centralized model: expertise without context
A centralized data team has real strengths. You get strong leadership with a unified vision for what data should do across the organization. You get deep expertise, people who understand data pipelines, capture, analysis, reporting, and optimization at a level that only comes from doing it full-time. You get knowledge of the platforms your brand runs on. These aren't small advantages.
But centralized teams have two problems that undermine everything they're good at.
First, they're often at odds with engineering and product. Sometimes this is just misaligned priorities and roadmaps, the data team needs something built, the dev team has a different backlog. In healthier organizations this is manageable friction. In unhealthy ones, it becomes outright conflict where IT, product, and engineering want nothing to do with supporting analytics. The data team becomes an island.
Second, and this is the more insidious problem, centralized teams lack context. They're disconnected from what the business units they support actually need. Marketing needs to understand customer acquisition patterns. Product needs to understand feature adoption. The revenue team needs to understand conversion behavior. A centralized team sitting three org chart layers away from those teams often doesn't know what questions to ask, let alone how to prioritize the answers. They have the skill to analyze but not the proximity to know what analysis matters.
The distributed model: context without leadership
The distributed model swings the other direction. Push data people into each business unit. Get them close to the work. Sounds logical.
In practice, it fails worse than centralization.
Distributed data teams tend to lack leadership. There is no grand vision for what the organization is doing with data. Everything is compartmentalized and focused on individual business units, which makes it nearly impossible to make decisions at a broader brand level.
They also tend to lack expertise. Most business units can't afford full-time data specialists, so data work gets assigned to someone who has some experience, some interest, or was simply available. It becomes a task alongside other non-data-related tasks. You end up with people who lack deep knowledge of data platforms and analytical methods trying to make infrastructure and governance decisions they aren't equipped to make.
The result is siloed data, misaligned definitions, inconsistent formats, and conflicting metrics as you traverse the user experience. One business unit defines "conversion" differently than the next. The data captured by marketing doesn't connect to the data captured by product. Customer records fragment across systems that were never designed to talk to each other.
And here's where it gets worse, this internal dysfunction gets exposed to the customer.
We've all experienced this as consumers. You call a brand for help and they tell you "that's in another database" or "i need to transfer you to another team." We don't care how a company has structured things internally. We want a cohesive brand experience, not a compartmentalized one. When organizations distribute their data teams and push them into individual functions, it doesn't just result in bad data and poor governance. It results in a fragmented customer experience that erodes trust and loyalty.
i would argue the purely distributed model is actually worse than the purely centralized one. At least a centralized team produces consistent data, even if it lacks business context. A distributed team often produces inconsistent data and lacks the leadership to recognize the problem.
The hybrid model: centralized leadership with embedded liaisons
The model that works, and i've seen this consistently across organizations that have healthy data cultures, is a hybrid.
You maintain a centralized data team with centralized leadership. This team has deep expertise in data capture, pipeline architecture, analysis, reporting, and optimization. They are the experts. They drive the program. They set the standards, own the governance, and keep the organization looking forward rather than being reactive.
But you don't stop there.
You create formal ties into each business unit you support. In most healthy organizations, this takes the form of what's often called a dotted-line relationship. You establish named liaisons, specific people in each business unit who serve as the bridge between the central data team and the work happening in their part of the organization.
These liaisons aren't just communication channels. They're advocates. They help the central team understand what the business unit needs, what questions they're trying to answer, what decisions they're trying to make. They help coordinate work and manage priorities so the central team isn't guessing about what matters. And they help the business unit understand what the data team can deliver, what the constraints are, and why governance and standards exist.
It isn't perfect. No organizational structure is. But of the three models, the hybrid approach delivers the best outcomes on every dimension that matters:
A cohesive customer experience. Because data standards are centralized, the customer doesn't feel the seams between business units. Their experience is consistent regardless of which part of the organization they're interacting with.
Cleaner, more trustworthy data. Central governance means consistent definitions, consistent formats, and a single source of truth. The liaison model means that governance reflects actual business needs rather than abstract ideals.
Strong data leadership. Someone owns the vision. Someone is thinking about where data strategy needs to go in two years, not just what report is due this week. This forward-looking orientation is almost impossible to maintain in a distributed model where everyone is heads-down in their own function.
Deep data intelligence. You're not just distributing data to stakeholders and hoping they make sense of it. You have experts who mine the data, do complex analysis, and truly understand your customers. The difference between a team that can pull a report and a team that can tell you what the report means and what to do about it is the difference between data distribution and data intelligence.
The ranking
If i had to rank them:
Hybrid: Centralized expertise and leadership with embedded liaisons in each business unit. Best data quality, best customer experience, best strategic positioning.
Pure centralized: Strong expertise and governance, but disconnected from the business. The data is consistent and trustworthy, but the team struggles to deliver what the business actually needs without formal hooks into other parts of the organization.
Pure distributed: The worst outcome. Poor leadership, inconsistent data, fragmented customer experience, and no one with the authority or perspective to fix it.
If your organization is running a purely distributed model, that's the most urgent thing to fix. If you're running a purely centralized model, it's working better than you might think but you're leaving significant value on the table by not creating those formal liaisons into the business.
The structure of your data team is a strategic one that shapes the quality of every decision your organization makes with data. Get it right, and data becomes a competitive advantage. Get it wrong, and your customers will feel it before your leadership team ever notices.
