Data Modelling
Part of Self-service analytics
Providing trusted datasets for non-technical teams
Make reusable datasets understandable with approved measures, visible limits, ownership and a practical first-user check.
A dataset is useful for self-service reporting when people can find it, understand what its measures mean and see where its use stops. Give it an accountable owner, plain-language definitions, an update point and documented limits. An endorsement badge alone cannot supply that context.
Start with the questions it can answer
Describe the decisions the dataset supports and the population it covers. For an illustrative sales dataset, say whether cancelled orders, refunds and incomplete days are included. Identify the reporting date and the grain of each exposed table or measure. A creator needs those facts before choosing a chart.
Keep the creator-facing fields understandable. Approved measures can prevent several similar raw columns from becoming competing versions of the same figure. Document fields that need specialist interpretation or leave them out of the creator-facing view. The right measures depend on the organisation's agreed definitions.
Give the dataset a compact data sheet
Question / What the creator needs
- What is it for?
- Intended business questions and audience
- What is being counted?
- Reporting grain, key and approved measure definitions
- Who is included?
- Population, exclusions and important filters
- How current is it?
- Refresh expectation and last successful update
- Who owns it?
- Business and technical contacts
- Where does it stop?
- Known gaps, restricted uses and change route
Put the answers where people find the dataset. Update them when a definition changes or a refresh fails. Otherwise, a clear-looking catalogue entry may lead people to an outdated or misunderstood figure.
Key steps to make datasets trusted for non-technical teams
- Define the dataset's purpose and target audienceSupports decisions on sales performance for regional managers
- Clarify what is being counted and the reporting grainMeasures are at daily transaction level; includes cancelled orders but excludes refunds
- Specify who is included and any exclusionsCovers all Australian retail outlets; excludes international sales and incomplete days
- Identify business and technical ownersBusiness owner: Sales Operations Lead ([email protected]); Technical owner: Data Engineering Team
- Document known gaps and restricted usesDoes not include GST adjustments; not suitable for financial audit reports
Separate discovery from suitability and access
In Power BI, an owner or someone with workspace write rights can promote content they consider useful. Certification requires authorised reviewers and an administrator-enabled process. Endorsed content is easier to find, but discoverability does not mean the model suits every question.
The same principle works without a certification feature: maintain a list of supported datasets, the questions each answers and the person to contact for exceptions. Remove misleading names and superseded copies from that list.
Pros and cons of endorsement vs. manual curation for dataset trust
- Pros of Power BI endorsementAutomated discoverability; visible badge; integrated with workspace permissions
- Cons of Power BI endorsementDoes not guarantee suitability for all use cases; may mislead if context is missing
- Pros of manual curationGreater control over content quality; clear alignment with business needs
- Cons of manual curationRequires ongoing maintenance; risk of outdated or duplicated entries
Check usability with a new creator
As a proposed acceptance check, give someone outside the modelling team a realistic question. Ask them to find the dataset, choose the intended measure, filter to an agreed period and explain the result and its limits. Compare their answer with an approved reference. If they choose the wrong field, check whether its name or description led them there.
Record unclear terms, missing fields and requests outside scope. The owner can use those observations to improve the dataset and tell dependent report owners when a change affects their work.
How to validate a dataset with a new creator
- Ask the user to find the correct dataset and measureEnsure they select the approved 'Daily Sales (AUD)' measure, not raw 'Sales Amount' columns.
- Compare with approved reference answerCheck consistency in measure selection, filtering and interpretation of limits.
