The Decision
Growth across 80+ clinics depends on numbers leaders can trust.
The client is a fast-growing US urgent care network that runs 80+ walk-in clinics. In this business, speed of service, patient satisfaction and operational efficiency drive loyalty and growth.
As the network expanded, leadership needed consistent, timely visibility into patient experience, clinic operations, collections and workforce productivity. The data to answer those questions already sat across clinical, operational and financial systems. The task was to turn it into decisions the business could act on with confidence.
The Problem, Reframed
The data was there. A governed way to combine it was not.
The network drew on five sources: Experity, eCW, a Snowflake data store, multiple APIs and spreadsheets, along with a set of legacy reports. Each feed arrived in a different structure, and business rules changed often. With no standardized layer applying data governance across the feeds, the same metric could be calculated three different ways. The effects showed up in three places. Leaders could not reliably track patient satisfaction, door-to-door time, collections or labor hours, and some insights stayed locked in legacy reports. Data engineering, modelling and logic validation were done by hand, so every rule change stretched delivery cycles. When data jobs failed, the failures surfaced late and root cause analysis took significant time, which delayed business reporting.
The Work
Apexon unified and governed the data first, then put AI to work on the engineering.
Apexon followed a unify, govern, automate approach drawn from its Harness Engineering practice. The team built a governed Microsoft Fabric and OneLake foundation on a Bronze, Silver and Gold medallion architecture and consolidated all five sources into a single trusted layer. Reusable data models cover clinics, providers, patient flow, financials and performance. Fabric notebooks handle the complex business logic, so rules are standardized and KPIs stay consistent.
On this foundation, Apexon delivered six Power BI modules spanning patient experience, operations, finance and workforce. Semantic models, AD group-based access control and a centralized Power BI landing page give users one secure entry point.
AI is built into the engineering lifecycle. AI data agents automate repetitive data engineering and modelling tasks, which speeds up build cycles. Copilot, guided by the client’s own business instructions, validates notebooks and transformation logic before each release, applying AI assurance to the engineering lifecycle rather than to the reports it produces. Data jobs run on a schedule, and when one fails, an email alert goes out with the root cause analysis attached.
What Changed
Leaders now track patient, financial and workforce performance from one place.
Six modules are live. NPS, Door-to-Door Time and Patient Volume give leaders insight into patient experience and growth. Collection Time of Services adds visibility into collections and financial performance. Labor Hour Work supports workforce productivity, and Unlocking Charts brings insights out of legacy reports. Every module draws on the same governed layer and standardized logic, so KPIs are consistent across all 80+ clinics.
The data team works differently too. Engineering and modelling work that used to be manual now runs through data agents. Copilot checks business logic before every release. Failure alerts with RCA have cut troubleshooting time and replaced firefighting with automated monitoring.
Now in its second year, the partnership continues as an ongoing managed data services engagement. The platform was built to scale and is ready to extend with new modules and AI capabilities. The outcome: one trusted view of the business across 80+ clinics, powered by a data operation that is faster, AI-assisted and self-monitoring.
Outcomes
Data sources: 5 to 1
Five disparate sources unified into a single governed OneLake layer.
Dev and validation effort: 30%
Data agents build the pipelines; Copilot semi-automates logic validation
Troubleshooting time: ~ 50%
Automated failure alerts with RCA replace late discovery and manual investigation.
1+ year
Ongoing partnership, with the platform ready for new modules and AI capabilities.
ARCHITECTURE DECISIONS
Governed OneLake layer over siloed feeds. Bronze, Silver and Gold layers bring five sources together and apply one set of business rules.
Data agents over manual engineering. Repetitive engineering and modelling tasks are automated, which accelerates build cycles.
Copilot validation over manual logic checks. Copilot, trained on client-specific business instructions, checks notebooks and transformation logic before release.
Why These Choices, Not the Obvious Ones
SILOED FEEDS
Why build a governed layer before building reports?
Data from five sources arrived in different structures, and business rules changed often. Without a governed layer, reporting was delayed and inconsistent. A Bronze, Silver and Gold architecture on OneLake brings every source into one place, and Fabric notebooks apply standardized logic once, so all six modules report the same KPIs.
MANUAL ENGINEERING
Why automate data engineering with AI agents?
Engineering and modelling were done by hand, and every change to business rules stretched delivery cycles. Data agents now take on the repetitive engineering and modelling tasks, which accelerates build cycles.
MANUAL VALIDATION
Why train Copilot on the client’s business rules?
Logic validation was done by hand. Copilot, guided by client-specific business instructions, validates notebooks and transformation logic before each release. Business logic is checked against the client’s own rules every time.
The Broader Implication
Dashboards are what leaders see. The lasting value is a data foundation that keeps up when business rules change.
A growing care network rarely lacks reports. The harder problem is keeping numbers consistent when data comes from many systems and the rules behind each KPI keep shifting. Without a governed layer, every rule change means manual rework, and every report risks telling a slightly different story.
This engagement points to an order of work, and to a data strategy decision that comes before any tooling choice. Govern the data first, so business logic is defined once and every module draws on it. Then put AI where the effort repeats: data agents for engineering and modelling, and Copilot to check logic against the business’s own rules before release. The outcome is a reporting capability that can add new modules and absorb rule changes without starting over, so decisions stay grounded in data leaders trust as the network grows.
FAQ
Apexon followed a unify, govern, automate approach. It built a governed Microsoft Fabric and OneLake foundation on a Bronze, Silver and Gold medallion architecture, consolidated five sources into a single trusted layer, and delivered six Power BI modules spanning patient experience, operations, finance and workforce.
A medallion architecture organises data into Bronze, Silver and Gold layers so raw feeds are progressively cleaned, conformed and made ready for reporting. In this engagement it brought five differently structured sources into one place, with Fabric notebooks applying standardised business logic once so every module reports the same KPIs.
Data from five sources arrived in different structures and business rules changed often. Without a governed layer, reporting was delayed and inconsistent, and every rule change meant manual rework. Defining business logic once, in one place, is what lets six modules report consistent KPIs across all 80+ clinics.
Data agents automate repetitive data engineering and modelling tasks, which accelerates build cycles. Copilot, guided by the client’s own business instructions, validates notebooks and transformation logic before each release. Data jobs run on a schedule, and a failed job triggers an email alert with root cause analysis attached.
The case study reports five data sources unified into one governed OneLake layer, a 30% reduction in development and validation effort, roughly 50% less troubleshooting time through automated failure alerts with root cause analysis, and six live Power BI modules in a partnership now in its second year.