In short
A business data strategy sets out which decisions need better information, where that information comes from, what the numbers mean and who keeps them reliable. The result should be answers people can find, understand and act on, with the right access and a visible refresh time.
In this guide
A business view of how the pieces connect.
01 Where the information starts
CRMCustomers and sales
Finance / ERPInvoices and costs
OperationsOrders and delivery
Other sourcesPeople, files and events
02 Your governed data environment
EDMPEnterprise data management platform
A / Lake + warehouse
- Microsoft Fabric
- OneLake provides the shared data lake.
Other designs use cloud lake storage connected to the chosen platform.
- Microsoft Fabric
- Fabric Warehouse
- Snowflake
- Managed tables and SQL analytics
B / Lakehouse
Lake storage + managed analytical tables
- Databricks
- Lakehouse with managed analytical tables
- Microsoft Fabric
- Fabric Lakehouse on OneLake
Your business agrees the shared structure and identifiers. The delivery team implements those mappings in the chosen platform as data is prepared.
Your business defines revenue, margin and active customers. These product features can hold and reuse those definitions.
- Microsoft Fabric
- Power BI semantic models
- Databricks
- Unity Catalog metric views
- Snowflake
- Semantic views
03 Information people can use
- DashboardsSee what changed
- AlertsKnow when to act
- Approved AIAsk in business terms
"Which customers are growing, at what margin, and where should we act?"
Named products illustrate where capabilities fit, not a requirement to buy all three. Features overlap and patterns can be combined; a warehouse can also receive data directly. Refresh speed follows the decision, from scheduled updates to streaming where it is useful.
Start with the question that keeps coming back.
Which customers are growing but becoming less profitable? Where is delivery falling behind? What cash is likely to arrive next week? Too often, answering a straightforward question means exporting files, chasing colleagues and explaining why two reports disagree.
A data strategy turns that repeated effort into a dependable business capability. Choose a decision, name the person responsible, agree the information they need and decide how quickly it must arrive. Then work backwards to the systems and data.
"Every Monday, our operations lead needs to see overdue work by customer, its financial impact and who owns the next action."
That gives the team a user, a question, a cadence and an action to design around.The outcome may be fewer hours reconciling reports, earlier intervention in delayed work or a clearer view of customer value. Establish the starting position and measure the change. A platform purchase alone does not establish any of those benefits.
What is an EDMP, and where does a data lake fit?
Here, EDMP means enterprise data management platform: the connected capabilities used to bring information together, organise it, control access and make it usable across a business. The acronym is not a universal product category, so confirm what a supplier includes. It might describe a combination of services rather than one product.
Think of the platform as the supporting infrastructure. Your data strategy also covers the people, definitions, priorities and operating responsibilities that make it useful. A data lake is one possible part of that infrastructure.
- 01 / KeepData lake
- Stores many kinds of information, including detailed source records and files.
- Useful when you need flexibility and history.
- 02 / OrganiseData warehouse
- Structures data for repeatable reporting, comparison and analysis.
- Useful when teams need consistent business reporting.
- 03 / CombineLakehouse
- Adds managed tables, reliability and analytical capabilities to lake storage.
- Useful when reporting and engineering share a foundation.
These are architecture patterns, not three products every business needs to buy.
A folder of exports is not automatically a trusted data lake. Someone still needs to maintain the connections, record where data came from and catch errors. In a common lakehouse pattern, data moves from raw to checked to ready for a business purpose. You may hear those stages called bronze, silver and gold. Databricks describes this as the medallion architecture.
A smaller business may start with its existing reporting tools, a few controlled connections and shared definitions. Add a broader platform when fragmented sources, volume, reuse or control requirements justify it. Avoid buying a complex architecture to answer a question an existing system can already answer reliably.
Canonical connects the records. Semantic connects the meaning.
These two concepts solve different problems that often get mixed together.
A canonical data model is an agreed common representation of a business thing, such as a customer, product or order. It helps different systems exchange and combine records consistently. It does not automatically identify duplicates or decide which source is correct. Matching rules, source priorities and an owner for uncertain cases are still needed.
Match verified identifiers. Review uncertain matches.
Source references retained. Legal entity and trading name kept distinct.
Fictional records. Similar names alone are not enough to merge customers.
Microsoft's Common Data Model is one example of a shared model intended to support consistency between applications. A business can also define its own model for the data it needs. Start with the entities in the first use case rather than attempting to standardise the entire company at once.
A semantic layer gives that data business meaning. It defines measures, relationships and usable names so a report, an analyst and an approved AI assistant can work from the same definitions. "Active customer" could mean a current contract, a recent purchase or a service interaction. The business must choose the appropriate definition and label alternatives clearly.
- Basis
- Excluding GST, by posting date
- Owner
- Finance approves the definition
- Boundary
- Not cash received or recognised revenue
The same definition in the dashboard, the spreadsheet and an approved question interface.
The example is a reporting definition, not an accounting policy. Currency, entities, adjustments and period rules need to be agreed for the actual business. Power BI semantic models and Snowflake semantic views are examples of tools for expressing shared business definitions. Permissions must be enforced in the underlying data and reporting configuration; a friendly label is not an access control.
Where do Databricks, Snowflake and Microsoft Fabric fit?
All three can support substantial analytics work. Their capabilities overlap, and the right choice depends on your existing systems, skills, data and operating needs. The comparison below gives starting questions for an evaluation, not a universal ranking.
| Platform | In business terms | Question to test |
|---|---|---|
| Microsoft Fabric | An integrated analytics environment spanning data movement, engineering, warehousing, real-time workloads and Power BI, with OneLake as its shared data foundation. | Can the existing Microsoft and Power BI team operate the full workflow, with suitable capacity, access and running costs? |
| Databricks | A lakehouse-based platform for data engineering, SQL analytics, data science and AI over a governed data foundation. | Do the workloads benefit from that engineering flexibility, and who will build, monitor and support them? |
| Snowflake | A managed data platform for analytical workloads and data sharing, with semantic views that express business entities, relationships and measures. | How well does it support the reporting and sharing needs, and how will the team manage workloads and cost? |
Ask each shortlisted option to answer the same real business question using a representative data sample. Test reconciliation, access restrictions, refresh failures and ongoing effort as well as the attractive dashboard. Include licensing, compute, storage, connectors, migration, support and staff capability in the comparison. A business does not need all three platforms to have a sound data strategy.
"Real-time" should match the decision.
An equipment alert and a monthly profitability review do not need the same refresh speed. Live streaming can help when a delay changes the response. A dependable daily refresh can be entirely appropriate for a planning decision.
- SecondsRespond nowAn operational event needs immediate attention.
- MinutesKeep today's work movingWatch queues, orders or availability during the day.
- DailyPlan the next cycleReview receivables, capacity or customer performance.
- Period closeMake a considered comparisonUse reconciled results for the management review.
Illustrative decision windows, not platform performance promises.
The full delay includes the source system, extraction, processing, quality checks and the report itself. A dashboard that refreshes every minute can still display yesterday's accounting extract. Show data current to as well as report refreshed at, and flag stale or missing sources.
Fabric Real-Time Intelligence supports event-driven analytics. Snowflake dynamic tables use a target lag to guide refreshes, but actual lag can exceed that target. The practical requirement is an end-to-end freshness expectation, monitoring and an owner who acts when it is missed.
What does data at your fingertips actually look like?
It means the right person can answer an everyday question without rebuilding the numbers. The answer shows its scope, definition, freshness and source records. It also makes the next action clear.
CEO: Which customers deserve our attention?
Bring together: customer identity, invoiced sales, agreed cost measures and service history.
Make visible: the reporting period, margin basis, missing costs and the account owner.
Next action: review the relationship, service model or pricing with the responsible team.
COO: Where is work getting stuck?
Bring together: orders, tasks, handoff times, capacity and open exceptions.
Make visible: the oldest unresolved items, source freshness and who owns the next step.
Next action: resolve a blocked handoff or rebalance work before the delay spreads.
Finance: Which expected receipts need follow-up?
Bring together: open invoices, payment status, disputes and collection notes.
Make visible: the last reconciled position, assumptions and any bank-feed delay.
Next action: assign a follow-up and update the cash expectation with its owner.
Those answers might appear in Power BI, a governed spreadsheet, an alert or an approved conversational interface. An AI assistant is useful when it can query defined measures, respect permissions and point back to supporting records. Test ambiguous questions and incorrect assumptions. Fluent wording should never be mistaken for a verified answer.
Self-service also needs boundaries. A branch manager may see only their branch; a finance analyst may need a consolidated view. Keep sensitive fields restricted, retain an audit trail and give users a clear way to report a number they cannot reconcile.
Build one answer people will use.
The first deliverable should be small enough to validate and important enough to use. Here is a suggested first-month sequence for a limited use case with accessible data. Timing will depend on source access, quality and approvals.
- Week 1Agree the decision
Name the user, question, measures, owner and current effort.
- Week 2Connect and reconcile
Match the records, agree definitions and expose missing data.
- Week 3Put it in people's hands
Build the view, set permissions and test freshness and failures.
- Week 4Check the business result
Observe real use, compare the effort and decide what earns expansion.
Accept the result when a user can trace a measure to its source, explain its definition, recognise stale data and take the intended action. Track reporting effort, time to answer, reconciliation exceptions and usage before claiming a return on investment.
Start with your business
Which decision takes too long because the information is scattered?
Bring that question, the systems involved and the reports your team uses today. Advery can help define the data, ownership and first useful delivery step.
Discuss your data prioritiesSources and guidance
- Microsoft: What is Microsoft Fabric?
- Microsoft: Lakehouse in Fabric
- Microsoft: Fabric Data Warehouse
- Databricks: What is a data lakehouse?
- Databricks: From raw data to trusted data products
- Microsoft: Common Data Model
- Databricks: Unity Catalog metric views
- Microsoft: Semantic models in Power BI
- Snowflake: Semantic views
- Snowflake: Dynamic tables and target lag
- Microsoft: Real-Time Intelligence
About Advery
Advery draws on 15+ years of enterprise experience across operations, customer growth and technology, connecting leadership decisions with hands-on delivery.
Explore the experience behind Advery