Self-service vs centralised BI

What it does

The operating model choice: who owns and manages content. Microsoft describes three strategies — business-led self-service, managed self-service, and enterprise — and the choice drives governance, support, training and the shape of the Centre of Excellence.

Key facts

Business-led self-service means creators and subject matter experts within a business unit own everything, data and reports alike. Microsoft also calls this decentralised or bottom-up.
Managed self-service means a central team owns the data and business users own the reports and dashboards. Microsoft's phrase for it is "discipline at the core and flexibility at the edge".
Enterprise means a central team — IT, enterprise BI, or the CoE — owns everything.
Microsoft is explicit that organisations rarely operate exclusively in one mode. The strategy varies by solution and by team, and a single team can be both a consumer of enterprise content and a producer of its own.
Governance intensity tracks the strategy: business-led gets the lightest oversight, managed self-service moderate, enterprise the most rigorous.
Microsoft separates four roles that clients routinely conflate: data steward (quality and master data), subject matter expert (what the data means), technical owner (builds, maintains, publishes, secures) and domain owner (higher-level policy decision-maker).
The strategy chosen depends on solution requirements, user skills, ongoing commitment to training, flexibility needed, complexity, and leadership commitment.
Governance level also depends on the scope of content delivery, the sensitivity of the subject area, and whether the data drives critical decisions.
Microsoft maps the strategies onto its usage scenarios — personal and team BI for business-led, team and departmental BI for managed self-service, enterprise BI for the centralised end.

When to use / skip

Managed self-service is the answer for most mid-size and larger clients, and it's the one to advocate for. Central ownership of the semantic model gives you a single set of numbers, controlled refresh and real security; local ownership of reports means the business doesn't queue for six weeks to change a chart title. It's also the model the platform is built around — shared semantic models, Build permission, workspace separation — so you're working with the grain.

Full enterprise ownership is right where the data is regulated, the numbers are externally reported, or the consequences of someone getting it wrong are legal rather than embarrassing. It is not right as a general default, and the clients who choose it as a default end up with a two-year backlog and a shadow estate of Excel files nobody can see.

Full business-led self-service is right for genuinely local, low-stakes analysis and nothing else. Left as the organisation-wide model it produces four versions of revenue and an argument in every leadership meeting.

The realistic answer is nearly always a hybrid: enterprise ownership for the certified core, managed self-service for the departmental layer, and a permissive personal layer that nobody pretends is governed. Say that out loud rather than letting it happen by accident.

Configuration decisions

Which content sits in which tier, and what the promotion path is between them. A report that becomes important should have a route into central ownership that isn't "rebuild it".
Who owns the semantic models, and whether that team has the capacity to serve every report team that will depend on them.
What certification and endorsement actually mean in this organisation, and who is allowed to award them.
Which governance controls are mandatory and which are advisory. Making everything mandatory is how governance gets ignored entirely.
What the personal layer looks like — My Workspace, local files — and whether the organisation is prepared to accept that it exists.
Who staffs the CoE, and whether that's a real funded role or someone's evening job.

Gotchas

Governance that requires a ticket for every change produces a shadow estate rather than compliance. The Excel files don't go away; they just stop being visible.
Central model ownership becomes a bottleneck the moment three report teams depend on it and the modelling team has two people. Name the resourcing implication when you propose the model.
The four roles get collapsed into "the BI person", and then nobody owns data quality. Separating steward from technical owner is often the single most useful thing you do on an adoption engagement.
Business-led content gets business-critical without anyone noticing. The trigger is usually a director quoting it in a board pack, and by then it has no owner, no documentation and no refresh monitoring.
Different teams legitimately sit in different modes. Enforcing one operating model tenant-wide is what makes governance feel arbitrary to the teams it doesn't fit.

Consultant notes

This is an organisational design conversation wearing a technology hat. If the sponsor isn't senior enough to change how teams work, you're documenting a preference, not implementing an operating model.
Ask what happens today when someone wants a new report. The current lead time tells you more about the real operating model than any policy document.
Propose the promotion path explicitly — how does a good departmental report become a certified one. Clients almost never have an answer and it's the mechanism that makes hybrid work.
Adoption dies faster from friction than from lack of features. When you recommend a control, be able to say what it prevents and what it costs the people it applies to.
Endorsement and certification are cheap to turn on and worthless without an owner who applies them. Don't put them in a governance plan without naming that person.

Revisit after any significant change to the client's org structure — the operating model follows the org chart, not the platform