Endorsement and certification

What it does

Puts a badge on content so users can tell what's trustworthy. Three badges: Promoted, Certified and Master data. Endorsed items are labelled in the UI and get precedence in some lists.

Key facts

Promoted means the creators think it's ready for sharing and reuse. Any user with write permission on an item can promote it. No approval, no gatekeeper.
Certified means an organisation-authorised reviewer has confirmed it meets the organisation's quality standards. Anyone can request certification; only users named by a Fabric admin can actually certify.
Master data marks an item as the authoritative source for a kind of organisational data — product codes, customer lists. It applies only to items containing data, such as semantic models and lakehouses, and only users named by a Fabric admin can apply it.
All Fabric and Power BI items except Power BI dashboards can be promoted or certified.
Certification and master data endorsement are off until a Fabric admin enables them for the organisation.
Certification enablement can be delegated to domain administrators, so each domain can have its own reviewer set.

When to use / skip

Turn certification on as soon as there's more than one semantic model doing roughly the same job, which happens earlier than anyone plans for. Promotion is self-service and costs nothing, so let people use it — treat it as "the author says this is fit to reuse" and nothing more. The thing that makes or breaks this is whether certification means something. A badge that gets applied on request, without a review, is worse than no badge, because it tells users to trust content nobody checked. If the client won't resource reviewers, don't enable certification; enable promotion and be honest about what it means.

Configuration decisions

Who the certifiers are — a central BI team, domain-by-domain reviewers, or both — and whether certification enablement is delegated to domain admins.
What the certification criteria actually are: documentation, RLS tested, refresh monitored, owner named, naming standard met. Write them down before the first review.
Whether master data endorsement is used at all. It's useful in a mature tenant and noise in a young one.
Where the certification request queue lives and what the turnaround commitment is.
What triggers recertification — a model change, a schedule, or nothing.

Gotchas

Promotion is self-service, so it means precisely as much as the person who applied it. Users conflate it with certification constantly.
Power BI dashboards can't be endorsed at all, which surprises clients whose main artefact is a dashboard.
There's no expiry. A model certified two years ago and rewritten since keeps its badge until somebody removes it, and nobody removes it.
Badges don't change permissions. A certified model is still only as accessible as its sharing settings, and an uncertified one isn't blocked.
If the reviewer list is one busy person, the queue becomes the reason people stop asking, and the badge stops appearing on anything new.

Consultant notes

The badge is the easy part. Sell the client on the review checklist and the reviewer rota — that's the deliverable, and it's the bit that decays if nobody owns it.
Certify a small number of genuinely central models first. A tenant with three certified models everyone uses is in better shape than one with forty.
Delegate certification to domain admins where the client is big enough to have real domains. Central certification of everything is how the queue dies.
Put recertification on a calendar. Twelve months is a defensible default, and it forces the conversation about whether the model still matches the business.

Revisit if endorsement gains expiry or automated recertification, or once the client's domain structure changes who should be reviewing.