Case study · A receivables management business
Copilot ready models, and a customer graph behind skip tracing
Snowflake stays the source of truth. The Microsoft AI stack gets the data anyway.
Raised a receivables business's Power BI semantic models to a standard Microsoft Copilot can safely answer from, proved a no migration path onto Microsoft Fabric, and is now building the customer graph behind manual skip tracing.
- 20%of contact centre effort in manual skip tracing, what the build targets
- Zero migrationSnowflake stays the source of truth, mirrored into Microsoft Fabric
- Copilot readypriority Power BI semantic models raised to a governed standard
- 4 engagementsconsecutive, from proof of value to the build running now
About the client
The client acquires and services purchased debt portfolios for Australian banks, financial institutions and utility providers. Its analytics run on Snowflake and Power BI, and it wanted to reach Microsoft Copilot without moving either.
The challenge
Two problems arrived together, and they turned out to be the same problem: nothing had a shared, described view of the data underneath it.
- The Power BI semantic models were not structured, described or governed to a standard that could safely carry Microsoft Copilot or self service question answering.
- Around 20% of contact centre effort was going on manual skip tracing, across disconnected internal and bureau sources.
- There was no shared view of a debtor's people, employers, addresses and contact points, so every search started again.
- The client wanted the pattern captured as something its own team could repeat, not a one off outside build.
What we delivered
We started with a Microsoft Fabric assessment and proof of value, then sequenced three further engagements so each one built on the model, reporting and AI readiness pattern proven before it. The client directed priorities throughout.
Assess, then prove
- A Microsoft Fabric environment assessment and an AI readiness evaluation set the current state baseline.
- A Snowflake to Microsoft Fabric mirroring proof of value validated a no migration path: Snowflake stays the source of truth and the data reaches the Microsoft AI stack anyway.
Mature and enable
- Priority Power BI semantic models raised to a Copilot ready standard: structure, DAX, field and measure descriptions, verified answers and endorsement.
- Microsoft 365 Copilot technical readiness, and a Microsoft Agent 365 orientation.
Extend to a new use case
- A foundational customer graph in Microsoft Fabric, with a read only data agent and next best action scoring, aimed at the effort spent on manual skip tracing.
Outcomes
No migration
- Snowflake stays the source of truth. Mirroring brings the data into Microsoft Fabric, so the AI stack works on it without a platform move.
Models Copilot can answer from
- Priority semantic models structured, described, verified and endorsed, so self service question answering has something governed underneath it.
A repeatable pattern
- Captured as something the client's own team applies to the next model, rather than a build only we could repeat.
A target worth aiming at
- The customer graph being built now is aimed at the 20% of contact centre effort spent on manual skip tracing.
What is next
The current phase is standing up the customer graph and the data agent for skip tracing, and testing next best action scoring against real contact outcomes. A go or no go on scale up follows, with live application programming interface sources and agentic actioning the likely next candidates.
Your route
Build a result like this.
Tell us what is relevant to your organisation. The case study comes with your enquiry.
Discuss this work →