For a global fintech connecting lenders with borrowers, the challenge was not simply moving 28 feeds or processing datasets of up to 70 million rows. It was knowing how the feeds actually worked. Years of lender-specific rules, exceptions, validation steps and operating decisions had accumulated across SQL Server workflows, SSIS packages, dense documentation and tacit specialist knowledge. That operating logic was business-critical but difficult to locate, interpret, reuse or scale.
As volumes, partner requirements and regulatory expectations ballooned, that hidden logic became the real constraint. Every change required teams to reconstruct intent from code, documents and tribal knowledge before they could safely modify the process. Persistent looked beyond platform modernization and focused on extracting the client’s operating logic, translating it into reusable and machine-readable components and creating a governed execution model that could absorb greater scale and complexity without multiplying human effort.
Logic the Business Could Not Readily Reuse
Seventy million rows made the scale visible. Tribal knowledge made the risk harder to see. Large structures were processed through bespoke workflows whose rules had been encoded incrementally over time. The rationale behind transformations, file formats, lender exceptions, reconciliation thresholds and sequencing often lived in individual packages, heavy documentation or specialist routines. Teams could run the process, but changing it safely demanded manual interpretation and the continued availability of people who understood its history. As a result, the operating model did not just process complexity; it preserved and compounded it.
This created a business continuity and growth problem. New lender requirements could not be handled by simply adding compute; teams first had to rediscover the relevant operating logic, verify dependencies and rebuild confidence in the output. That friction put strategic KPIs, such as SLA compliance, feed turnaround and time to onboard a new partner, at risk. McKinsey’s 2026 Global Tech Agenda notes that leading companies are moving beyond technology modernization to rewire operating models for growth. For this client, that meant making institutional knowledge executable, not leaving it buried in artifacts and individual memory.
- Legacy workflow migration: Transitioning feed generation from SQL Server and SSIS to Databricks to achieve long-term scalability and maintainability.
- Complex data structures: Refactoring large, intricate tables such as Marketing_Extended and managing high-volume datasets of up to 70 million rows.
- Functional parity and performance: Guaranteeing a zero-defect migration, with accurate record counts, exact output formatting and efficient execution at scale, so that lenders experienced no disruption.
Converting Institutional Knowledge into Executable Feed Model
Persistent began by reconstructing the client’s feed operating model: tracing rules, exceptions, dependencies, control points and decision paths across code, documentation and expert knowledge. The team then translated that logic into a modular feed factory on Databricks. A single parameterized engine replaced 28 separately maintained execution paths. Configurable lender rules made variations explicit and machine-readable. Reusable transformation components captured logic once instead of rebuilding it feed by feed. Automated reconciliation against golden outputs converted expert validation into an embedded control, while Unity Catalog and CI/CD made governance and release practices repeatable.
The result was more than a modern architecture: it was a durable context layer for feed operations, one that preserved business intent while allowing execution to scale. This matters because Bain’s 2026 report warns that modernization can create the “legacy tech of tomorrow” when architecture investment is not accompanied by operating-model change.
Reusable Logic Turned Growing Complexity into Scalable Capacity
Once the operating logic was explicit, reusable and executable, scale no longer required a corresponding increase in interpretation and manual effort. All 28 feeds moved with full parity and zero-defect validation.
The standardized framework cut annual platform cost by 60%, reduced manual effort by up to 80% and halved feed generation time.
More importantly, the business gained a model that could accommodate new volumes, lender rules and reporting requirements by extending governed components rather than rediscovering and recoding institutional knowledge. The outcome aligns with BCG’s 2026 Applied AI Index, which finds that companies combining strategic clarity with applied technology generate five times as much value as those lacking both.
- 60% annual cost reduction: Moving to Databricks significantly lowered the yearly cost of running the client’s data platform.
- 100% of feeds migrated with full parity: All 28 feeds were successfully migrated and validated against golden outputs, ensuring a zero-defect transition for lending partners.
- Up to 80% reduction in manual effort: Automated workflows replaced SSIS-based processes, cutting operational overhead and execution time by up to 80%.
- 50% faster feed generation: Halving feed generation time enables faster lender reporting and stronger SLA compliance.
The transformation proved that technology can scale only when the business logic it runs on is no longer trapped in code, documents or people’s heads. By extracting that logic, making it explicit and encoding it as governed, reusable and machine-readable components, Persistent created an asset that compounds the impact of every subsequent change. New feeds, partner requirements and process variations can now build on a common operating foundation rather than restart discovery each time.
This also creates a runway for evolving integration, allowing new data platforms, automation tools, AI agents and emerging technologies to consume trusted business context without recreating it—so the client can modernize continuously while preserving the operating intelligence that makes each process work.
