One question: who owns the how
The debate about centralised and decentralised marketing is often muddled because it mixes two separate questions. One is about organisation design: where marketers sit, who they report to and how teams are structured. That question belongs to management, and the team models and organisation design guide covers it. The other is about operations: who decides the way marketing work is done, which systems it lives in, which rules it follows and how its data is kept. That is the question this page answers.
The two can be answered differently. An organisation can have marketers sitting in every region, reporting locally, while one central operations function owns the tools, the taxonomy and the approval matrix that all of them use. It can also have a single central marketing team in which nobody owns operations at all, so each specialist works their own way. Separating the questions makes the choice clearer.
Under centralised operations, one function sets the standards, chooses and administers the tools, owns the data and the taxonomy, and runs governance for everyone. Under decentralised operations, each team chooses its own tools, processes, naming and reviews, within whatever minimal rules the organisation imposes. Neither is wrong in itself. Each fits some kinds of work and some kinds of organisation better than others, and most real organisations mix them.
Compare table
Centralised operations vs Decentralised operations
| Row label | Centralised operations | Decentralised operations |
|---|---|---|
| Consistency | High: one process, one taxonomy, one set of standards across teams | Varies by team; customers may see inconsistent brand and messages |
| Speed of execution | Slower for local work if the centre also controls execution; fast for shared work once systems exist | Fast locally; teams decide without waiting for a central queue |
| Standards and procedures | Written once, maintained by one owner, applied everywhere | Each team writes its own, or none; good practice spreads slowly |
| Tooling | One stack, negotiated and administered centrally; fewer overlapping tools | Teams choose tools that fit them; duplication and overlapping licences are common |
| Data and reporting | Shared taxonomy and naming; results combine across teams without manual matching | Each team names and tracks differently; combining reports is manual and error-prone |
| Governance and compliance | One approval matrix and audit trail; risk managed consistently | Risk managed unevenly; a lesson learned in one team is not shared with another |
| Local relevance | Risk of generic work that ignores local markets, languages and timing | High: teams closest to the market decide what fits it |
| Cost | Lower duplication; a central team to fund | No central team to fund; higher hidden cost of duplication and rework |
| Ability to change | One decision changes everyone's way of working; hard to experiment locally | Teams experiment freely; spreading a good change across teams is hard |
| Typical failure | The centre becomes a bottleneck and local teams work around it | Fragmentation: nobody can see the whole picture or trust combined numbers |
The table describes the pure forms. In practice, the trade-offs depend on what is centralised. Centralising data and standards has few of the speed costs listed; centralising execution has most of them.
Where centralised wins
Centralisation wins wherever variation destroys value. The clearest case is data. A campaign taxonomy is only useful if every team uses it; a UTMURL tags identifying traffic source. convention applied by most teams still leaves the combined report wrong. Contact data, suppression lists and audience definitions held in a customer data platform (CDPUnified first-party customer profiles.) or a CRM only stay clean if one function owns the rules for what enters them. When each team runs its own, the organisation pays later in manual reconciliation and, where suppression is concerned, in compliance risk.
Tooling is the second case. Several teams buying separate tools for the same job pay more, integrate less and spread knowledge thinly. A single stack, administered centrally, lets the organisation negotiate once, integrate once and train once. The technical layer has the same logic: the tag container and the tracking it controls should have one set of change rules, and the tag container governance guide explains why uncontrolled changes by several teams are a common cause of broken measurement.
Governance is the third. One approval matrix, one claims library and one audit trail mean that a risk identified anywhere is managed everywhere. The approvals and compliance guide builds the matrix; it works far better as a shared asset than as a set of local versions. Centralisation also favours organisations in regulated sectors, those with a single brand across markets, and those where leadership reports a common KPIA metric tied directly to an objective. set and needs the numbers to mean the same thing everywhere.
Where decentralised wins
Decentralisation wins wherever local knowledge creates value. The team that knows a market, a language, a customer segment or a product's buyers will make better decisions about message, timing, channel mix and creative than a central team working from a distance. When every local campaign must pass through a central queue, the organisation loses that knowledge and gains a delay.
Speed is the second advantage. A team that can launch a campaign without waiting for central capacity can respond to a competitor, a local event or a seasonal moment in days rather than weeks. For organisations whose markets move quickly, that speed can matter more than consistency.
Experimentation is the third. Decentralised teams try new channels, tools and ways of working, and some of those experiments become the organisation's future standard. A fully centralised operation can stifle that, because every change needs central approval and central capacity.
Decentralisation also suits organisations whose business units are genuinely different: separate brands, separate customers, separate sales models. Forcing a single process on businesses that share little may cost more in fit than it saves in consistency. Even then, a thin shared layer, at least for data protection, suppression and brand protection, is usually worth keeping.
Hub-and-spoke in practice
Most organisations of any size end up with a hub-and-spoke model: a central operations function, the hub, owns the things that lose value when they vary, and the teams it serves, the spokes, own the things that lose value when they are standardised. The model works when the split is written down; it fails when it is left to negotiation, because then every disagreement becomes a turf argument. The tutorial on writing a marketing operations charter puts the hub's side of that split into a signed document, with what it does not own named as carefully as what it does.
| Area | Hub owns | Spokes own |
|---|---|---|
| Process | Workflow design, gates, SOP format and core procedures | Local procedures within the format; running work through the workflow |
| Systems | Tool selection and administration; work system configuration | Day-to-day use; requests for new work types or fields |
| Data | Taxonomy, naming rules, hygiene routines, suppression | Applying names correctly; local list sources declared at intake |
| Governance | Approval matrix, claims library, access policy, audits | Applying the matrix; local regulatory knowledge fed back to the hub |
| Brand | Guidelines, core assets and templates | Local adaptation within the guidelines |
| Execution | Shared campaigns and central channels where agreed | Local campaigns, messages, timing and channel mix |
| Reporting | Shared definitions and combined reporting | Local analysis and decisions |
Three practices make the model hold. The hub treats the spokes as customers, with a request route, a service level and a regular forum where spokes can propose changes to standards. Spokes have named operations contacts who know the standards and act as the hub's eyes in each team. And the split is reviewed periodically, because what should be central changes as the organisation grows, enters markets or adds products.
A consumer electronics brand moves to hub-and-spoke
Arclight Audio, a consumer electronics brand selling in eight European markets, ran fully decentralised marketing: each market chose its own tools, named campaigns its own way and ran its own approvals. The group marketing director could not produce a combined view of spend and results without two weeks of spreadsheet work, and the company was paying for five different email platforms. A central operations team of four took ownership of the stack, the taxonomy, the approval matrix and the asset library, while market teams kept control of local campaigns and channel mix. Within a year, the email platforms were consolidated to one, the combined monthly report took two days, and two markets that had been running promotions without legal review were brought into the shared matrix.
Common mistakes
- Centralising execution along with standards. The centre becomes a queue for every local post, and spokes work around it. Centralise the how; leave the doing close to the market.
- Decentralising data. Each team's taxonomy makes sense locally and makes combined reporting impossible. Data is the first thing to centralise.
- Leaving the split unwritten. Every disagreement about ownership becomes a negotiation. Write a one-page split and review it periodically.
- A hub with no service level. Spokes cannot plan around a central team whose turnaround is unknown, so they buy their own tools. The tutorial on setting service levels for marketing requests builds a catalogue the spokes agree to, with their own side of the bargain written in.
- No route for spokes to change standards. Local knowledge never reaches the standards, and spokes stop believing the standards apply to them.
- Confusing the operating model with the org chart. Where people report and who owns the how are separate decisions; the governance hub and the marketing operations hub explain how the operating model fits alongside process and systems.
