One number,
one owner

Certified semantic models, row-level security defined where the data lives, and refresh health visible to the people relying on it, so reporting is something your leadership acts on rather than argues about.

Most Power BI estates end the same way: two reports disagree and every meeting opens with whose figure is right. That is a governance failure, not a design one.

Get in touch See the delivery method

ModelsCertified per subject area

One semantic model per subject area with a named owner, and ad hoc reports built on it rather than beside it.

NumbersOne owner each

Every reported figure traced to a single owning model, so a disagreement has somewhere to be settled.

SecurityEnforced in the model

Row-level security defined once where the data lives, not reimplemented in each report by whoever built it.

RefreshMonitored and dated

Failures alert to a named person, and every report states when its data was last good.

What we deliver

Reporting built on one agreed source

Power BI is easy to start and hard to keep honest. What decides whether anyone trusts the output is the modelling and the governance underneath it rather than the choice of visual.

Semantic models

Star schemas, conformed dimensions, an agreed grain, and measures written once in the model instead of repeated in every report that needs them.

Each model covers one subject area, carries a named owner and a written definition per measure, so a figure can be explained without opening the file it came from.

One place to change a definition, and one place to defend it.

Certification and promotion

Certified and promoted datasets, a documented path from development to production, and workspace structure that keeps authored content separate from published content.

Certification is a review with a named reviewer rather than a badge somebody clicks. The record states who signed the model off and against which definitions.

A dataset list where the trustworthy ones are obvious.

Row-level security

Security roles designed against the organisation as it actually reports: entity, region, cost centre, and the exceptions senior people always turn out to need.

Roles live in the model and are tested against real user contexts, because a filter applied inside a report leaks the moment someone duplicates the visual.

A report that cannot show what its author forgot to hide.

Sources and preparation

Connections to Dynamics 365, Oracle, SQL Server, files and the third parties in between, with dataflows or a warehouse layer where the transformation is too heavy to sit in a model.

We will say when a warehouse is not needed yet, and we will say when modelling is being used to paper over a problem in the source system.

Transformation in the right layer rather than the convenient one.

Refresh health

Scheduled and incremental refresh, gateway management, failure alerting to a named recipient, and a freshness stamp on the face of the report.

A silently stale dashboard is more dangerous than a missing one, because people keep making decisions on it without noticing.

Staleness that announces itself before someone acts on it.

Governance and adoption

Tenant and workspace settings, sharing rules, licence position, and a periodic review that retires the reports nobody opens.

Adoption is measured as use of the certified models, not as the number of reports published. A growing report count is usually a symptom.

An estate that stops growing sideways.

How a Power BI engagement runs

StageOneAudit

What exists today: workspaces, datasets, the reports actually being opened, the figures that disagree, and the person each number is supposed to belong to.

Nearly every estate has two datasets computing the same measure differently. Naming them is the first useful output, and it is usually uncomfortable.

You receive:

A findings report

Duplicate and conflicting measures listed, orphaned datasets identified, refresh failures and stale content flagged, and each issue ranked by the decision it distorts.

This is diagnostic work and the report is yours whatever happens next.

StageTwoModel

The model before the visuals: subject areas, grain, conformed dimensions, measure definitions agreed with the people who own the numbers, and the security design.

This is the step teams compress under deadline pressure, and it is the one that decides whether the reporting is ever trusted.

You receive:

An agreed model design

Entity and grain map, one certified source per reported figure, measure definitions written in business language, and the row-level security design.

Signed off by the finance and operations owners of each number, not only by IT.

StageThreeCertify

Models built and tested against real data, security roles exercised with real user contexts, reports built on the certified model, and the promotion path used rather than described.

Testing includes a deliberate refresh failure, because how an estate behaves when data is late is the part nobody rehearses.

You receive:

Certified datasets in production

Certified models with named owners, tested security roles, refresh schedules with alerting, and a measure dictionary published where report authors will find it.

Report authors are trained on the model before they are given access to build on it.

StageFourOperate

Steady state: refresh monitored, model changes managed through the promotion path, new measures reviewed before certification, and the estate reviewed for content nobody uses.

A governed estate decays quietly if nobody owns it. The cadence is what keeps certification meaning something a year later.

You receive:

A monthly service review

Refresh outcomes and staleness, model changes released, new certification requests and their disposition, and the retirement list for unused content.

Licence position reviewed quarterly, including the seats nobody has signed in with.

Coverage

What we run, and what it produces

Delivered as a project, a managed service, or a project that converts into a managed service once the first models are certified.

Microsoft Power BI coverage
Area	Scope	Cadence	Evidence produced
Semantic models	Star schemas, grain, measures, written definitions	Continuous, reviewed monthly	Measure dictionary with an owner per figure
Certification	Reviewer, promotion path, development to production	Per release	Certification record naming the reviewer and the definitions
Security	Row-level roles, entity and region scoping, exceptions	Reviewed quarterly	Role test evidence per user context
Sources	Gateways, dataflows, warehouse layer where justified	Continuous	Source and lineage map per certified model
Refresh	Schedules, incremental policy, failure alerting	Daily verification	Refresh outcome log and a freshness stamp per report
Estate	Workspaces, sharing, licence position, unused content	Quarterly review	Retirement list and licence position statement

Trust

Two reports that disagree cost more than no report

A missing report is a gap someone works around. Two reports that disagree are worse, because the meeting stops being about the decision and becomes about the data, and after that happens twice nobody brings a number to the room at all.

The fix is governance rather than visuals: one certified model per subject area, one named owner per figure, measures defined in the model instead of in each report, security enforced where the data lives, and refresh health visible to the people relying on it.

Ask for a reporting review

Certified modelsOne semantic model per subject area, marked certified, with a named owner. Ad hoc reports build on it rather than beside it.
Security in the modelRow-level security defined once where the data lives, so a report cannot leak what its author forgot to filter.
Refresh monitoredFailures and staleness alert to a person, and each report shows when its data was last good. A silently stale dashboard is the expensive failure mode.

Applies to models we build, and it is the review standard we assess an existing estate against.

Questions we get asked

Before a Power BI review

Do we need a data warehouse before Power BI is useful?

Not to start. A well-built model over one or two well-understood sources delivers value early and teaches you where the real data problems sit. A warehouse earns its cost once reporting spans several systems that disagree with each other, or once volume makes direct queries unworkable, and that point becomes visible in the first months of running live rather than at business case stage.

We already have hundreds of reports. Where does this start?

With an audit of what is actually opened. Most estates carry a long tail of content nobody has viewed in months, a handful of reports that run the business, and two or three measures defined differently in different places. The certified models are built for the reports that matter, and the tail is retired rather than migrated.

Who owns the workspaces and the certification decision?

You do, and we set the governance up so that is practical: workspace structure, a written certification process, a promotion path from development to production, and a named owner per certified model. Ungoverned self-service reporting is the state most estates arrive in, and it is the reason nobody trusts the figures.

Can self-service reporting continue?

Yes, and it should. The distinction is between authoring reports and authoring measures. Anyone can build on a certified model. Defining a new measure goes through review before it is certified, because that is the step which decides whether two teams end up reporting revenue differently.

Can you work on an estate somebody else built?

That is the normal case. The first work is documenting the models, the measures, the security roles and the refresh paths, because that documentation almost never exists. Once it does, the disagreements between reports usually resolve into a small number of definition differences rather than a data quality crisis.

Numbers peopleact on.

Start with a reporting review. It needs read access rather than a contract, and it produces a list of conflicting measures and a certification plan you keep either way.

Get in touch Book a scoping call

AI Solutions

Oracle Services

Salesforce

Microsoft Enterprise Services

Application Management

Managed IT Services

Managed Business Services

One number,
one owner

Reporting built on one agreed source

Semantic models

Certification and promotion

Row-level security

Sources and preparation

Refresh health

Governance and adoption

StageOneAudit

A findings report

StageTwoModel

An agreed model design

StageThreeCertify

Certified datasets in production

StageFourOperate

A monthly service review

What we run, and what it produces

Two reports that disagree cost more than no report

Before a Power BI review

Numbers peopleact on.

One number,one owner

Reporting built on one agreed source

Semantic models

Certification and promotion

Row-level security

Sources and preparation

Refresh health

Governance and adoption

StageOneAudit

A findings report

StageTwoModel

An agreed model design

StageThreeCertify

Certified datasets in production

StageFourOperate

A monthly service review

What we run, and what it produces

Two reports that disagree cost more than no report

Before a Power BI review

Numbers peopleact on.

One number,
one owner