Where should analytics actually sit?

Where should analytics actually sit?

Analytics should sit close enough to a decision to understand its context and far enough across the enterprise to reuse data, methods and controls. The organisational choice is therefore not simply centralised or decentralised. It is a design of decision rights.

Centralise what benefits from scale and consistency: core data platforms, identity, security, enterprise definitions, shared tooling, advanced specialist pools and professional standards. Federate what depends on local knowledge: problem framing, prioritisation, interpretation, workflow redesign and ownership of realised value. A hub-and-spoke model can combine both, but only when authority is explicit.

Two failure modes reveal a poor design. In the first, a central team becomes a ticket queue, delivering reports without accountability for adoption. In the second, embedded analysts rebuild pipelines and definitions inside each function, producing speed locally and contradiction globally. Moving boxes on an organisation chart does not solve either problem.

Organise durable teams around decision domains or data products, with named business and technical owners. Give them service expectations for quality, freshness and access; common engineering paths; and forums to resolve cross-domain semantics. Rotate talent and maintain communities of practice so scarce expertise travels without severing business proximity.

Judge the model by decision cycle time, reuse, data reliability, adoption, duplicated effort and attributable outcome�not by the size of the central function. Analytics belongs wherever it can combine context with craft. Architecture and governance create the connective tissue; accountable business teams turn insight into changed performance.