Corios helped an $82B bank replace outdated software and rethink reporting processes without disrupting a single deadline

Corios specializes in one of the hardest parts of analytics modernization: moving critical Statistical Analysis System (SAS) workloads to the cloud without losing the logic, controls, and institutional knowledge already built into them. Its approach brings assessment, translation, validation, and team enablement together so enterprises can modernize without sacrificing the rigor their most critical analytics depend on.

That expertise became essential for an $82 billion U.S. regional bank when the technology behind its Current Expected Credit Losses (CECL) reporting approached the end of support. The bank needed to replace it on a fixed timeline without interrupting the quarterly reserve calculations required for financial reporting.

Corios helped migrate models across 10 loan portfolios to Dataiku and Databricks, keep quarterly reporting on schedule, and turn a heavily manual process into one that runs dramatically faster, produces reproducible results, and removes a known control weakness.

Modernizing without missing a reporting cycle

CECL determines how much the bank sets aside for expected credit losses, with direct implications for financial reporting and capital adequacy. Those reserve numbers are relied on by regulators, auditors, management, and the board.Producing them still required analysts to manually sequence programs, coordinate runs through email and meetings, and publish results through copy-and-paste steps.

Corios had to rebuild the models, data pipelines, controls, and operating process without disrupting quarterly reporting. The migration also created an opportunity to leave the bank with a faster, more reproducible way to run CECL.

Rebuilding the process around the people who run it

Corios migrated the bank’s core credit-risk models from SAS into Python and PySpark, with model versions managed in Databricks MLflow, then rebuilt the surrounding workflow on Dataiku and Databricks. 

Forecast dates, economic scenarios, and portfolio inputs can now be updated centrally and carried across all 10 model groups instead of being changed program by program. Dataiku coordinates model execution, reporting, and production approvals on Databricks, replacing a manually sequenced series of jobs with one workflow the credit analytics team can manage directly.

The Dataiku LLM Mesh serves the underlying models into that workflow without tying it to a single modeling framework so model versions and techniques can evolve without rebuilding the surrounding CECL process.

Multi-day model runs now finish in under 30 minutes

Individual credit-loss model runs that once took multiple days now complete in under 30 minutes. Results move through analysis and management reporting, publish to Power BI, and trigger stakeholder notifications without the manual coordination previously required each cycle.

Each production run also preserves the underlying data, model versions, code, assumptions, and management adjustments behind that final number. The bank gets to its results faster and can reproduce each reserve calculation.

Validating the new results before go-live

A migration this critical could not be judged by technical completion alone. The bank needed confidence that the new results matched the old ones before relying on them in live production.

Corios helped complete two full dress rehearsals using live data, comparing outputs from the old and new environments with its Validator tooling. Checks covered row counts, column totals, data types, and record-level differences, creating a documented view of migration accuracy.

The bank’s model risk team signed off, and the new financial-reporting controls passed Internal Audit and KPMG walkthroughs. They approved production results on March 30, 2025, meeting the migration deadline. The bank also eliminated recurring SAS licensing costs and removed the manual publishing step previously identified as a control weakness.

A migration deadline became a broader business win

Corios trained the bank’s modeling, analytics, and credit-risk teams in Python, PySpark, Dataiku, and Databricks so they could operate and extend the new environment themselves. The same architecture can support future work in stress testing, asset-liability management, credit risk rating, and other analytics use cases.

What started as a need to replace SAS evolved, giving  the bank a stronger way to run CECL with model runs that finish in minutes instead of days, reproducible reserve calculations, lower recurring software costs, stronger controls, and teams equipped to build on the new environment.