There have been endless debates for years on how the operating model for data, analytics and now AI teams should be structured. Should they be centralised or decentralised?

These debates are extremely predictable in their outcomes. Some will vouch for the centralised model because it promises control, standardisation, and efficiency, but they often slow decision-making and create bottlenecks. On the other hand, decentralised models empower business units with speed and flexibility, but they lead to chaos, duplication, and inconsistency.

Neither model works. And if you’re still stuck in this debate, you’re asking the wrong question.

The Symptoms of a Broken Operating Model

If your organisation is wrestling with how to structure data analytics and AI (DAAI) teams, you have probably seen these symptoms:

• Current teams are either overwhelmed or ignored: through my experience with several customers we have worked with, in many centralised models, the DAAI teams become gatekeepers, slowing down the business. In decentralised models, they get bypassed altogether, leading to poor-quality decisions.
• Business units end up building their own shadow data teams: when the central function is too slow, business teams start hiring their own DAAI people, creating multiple versions of the truth, standards fall apart, and the wild west takes over. Howdy partner!
• DAAI initiatives projects stall: it’s generally down to no data and AI strategy, not pinning it to the business outcomes, too much to bureaucracy (in centralised models) or because they are poorly executed and lack governance (in decentralised models). Or because there isn’t enough focus on change management from the outset. Or maybe it’s just because that old statistic keeps being bounded around and used as the failure scope!
• Executives are frustrated: They’ve invested millions nay, in fact some have invested billions depends on the time horizon and size of the company in their DAAI initiatives and just can’t see the impact. So, when they get rid or pair down the DAAI teams and push them into a small corner to work on “reports”, don’t cry wolf!

Sound familiar? That’s because the problem isn’t where DAAI sits in your organisation, it’s how they operate within the business.

The reason centralised vs. decentralised keeps failing is simple:

1. DAAI are treated as technical functions: Instead of embedding them into business value creation, organisations set them up as technology teams focused on outputs not outcomes. Data platforms, data catalogues, MDM etc. That’s fine but for what? To focus on such lovely pristine tech that goes nowhere fast?
2. Ownership is unclear: the business functions expect DAAI teams to deliver insights, while data teams expect the business to act on them. No one is accountable for outcomes. It’s a classic us and them, them and us approach.
3. Metrics don’t align with business success: centralised teams measure success by how much data they collect. Decentralised teams measure it by how quickly they can spin up a dashboard. Neither is tied to actual business impact. I’ve just seen this in a client we are working with rolling out data migration and quality for a new platform. No success measures, no understanding what needs to be tracked and no data migration strategy!
4. Governance becomes an afterthought: centralised models impose rigid, one-size-fits-all governance. Decentralised models ignore it altogether. Neither approach works.

It’s time to stop treating this as a binary choice.

The best organisations don’t pick centralised or decentralised. They use a federated model that embeds data and AI into the business while maintaining control where it matters most.

This means:

• Business focused, but centrally enabled: DAAI teams don’t dictate how the business uses insights. Instead, they provide the capabilities, frameworks, and governance that enable business teams to act independently within guardrails.
• Use-case driven, not technology-driven: instead of designing an organisational structure and hoping it works, companies start with business problems and align data and AI to solve them. Yes, I know I am a broken record, but it’s only for your own good!
• Clear accountability for results: DAAI teams aren’t just responsible for models and dashboards, they’re tied to measurable business outcomes. They need to show how they enabled these business outcomes. It’s not good enough just to say well I don’t know how to measure this. Figure it out! No slopey shoulders please!
• Governance that scales: instead of rigid top-down rules or a free-for-all, federated models create structured autonomy as business units can move fast, but within defined constraints. Governance doesn’t need to be endless bodies and a nightmare of an approval process!

This is what actually works.

When we work with clients, we don’t waste time debating centralised vs. decentralised. Instead, we design an operating model that aligns DAAI initiatives to value creation.

Here’s how:

1. Start with Business Outcomes: define the key value drivers (revenue, efficiency, risk reduction) and work backwards to data and AI.
2. Embed Data & AI Teams into the Business: each major business function has embedded data and AI specialists, but they follow common standards, tooling, and governance set by a central team.
3. Create a ‘Thin’ Central Function: instead of controlling everything, the central team focuses on enabling the business, owning key data assets, defining governance, standards, managing assets through the pipeline to production and ensuring interoperability.
4. Assign Clear Accountability : we make sure every AI and data initiative has an owner tied to a measurable business result, not just technical KPIs.
5. Ensure Governance is Flexible but Firm: guardrails are set centrally but adapted locally. This means governance is built-in, not bolted on.

This model moves beyond the false choice of centralised vs. decentralised. It builds an ecosystem where DAAI initiatives drive business results without never ending chaos.

One global supply chain company we worked with had a highly centralised data team. They controlled everything from data access, to reports, to analytics, and then AI initiatives.

The result? Business units couldn’t get the insights they needed quickly. Operations teams were waiting weeks for supply chain forecasts, procurement couldn’t access supplier performance data when needed and frustrated teams started creating their own spreadsheets and reports, leading to conflicting numbers and inefficiencies.

After moving to a federated model, here’s what changed:

• Warehouse and logistics teams could access inventory insights, helping them reduce stock shortages without waiting for approvals.
• The central data team shifted to enabling, not controlling, focusing on governance, standardisation, and maintaining core datasets, while business teams worked more autonomously within a clear framework.

This wasn’t about choosing centralisation or decentralisation. It was about creating a model that allowed business teams to move faster while ensuring data quality and control where it mattered most.

Since adopting this model, companies have seen:

• Faster AI and analytics adoption: business teams don’t have to wait for central approval to act.
• Clearer accountability: every data and AI initiative has a business owner tied to results.
• Stronger governance without bureaucracy: guardrails keep everything aligned, but business teams retain agility.
• Higher ROI on DAAI investments: because initiatives are directly tied to business impact, not technical milestones.

So, forget centralised vs. decentralised. The real question isn’t where data and AI should sit. It’s how they should work and drive outcomes.

If you’re still stuck in this debate, you’re already behind, stuck in traffic and late to the party!