Data Governance and Modeling

Three regulatory reports, partially produced and reconciled by machine instead of by hand

The result

Regulatory reporting ran 40% faster across a live reporting cycle for three different reports with different frequencies.
Reports were assembled from a governed model with lineage behind every figure. Five months to the shipped solution, three parallel reporting cycles before cutover.

The challenge

Reports were assembled by hand, from source material arriving as documents, spreadsheets and connections into FP&A tools. The reports had different frequencies and data models.

A daily position-level report is filed every business day, two days after the as-of date. There is no human in the loop, so a control that is not automated does not exist.

The quarterly systemic indicators report had poorly defined owners. Parts of the report arrived as different teams emailed their own data.

The consolidated report was governed by the close calendar, and late manual adjustments (outside the pipeline) were not captured in the lineage. Historical reports and line items needed to be preserved and explained.

The solution

Decomposition worked backward, from the end reports to the source systems that fed them.

From source systems to report, with verification after each hop Source material arrives as documents, spreadsheets and connections into FP and A tools. It feeds a governed model that holds classification as dated mappings, so a figure filed two years ago still explains under the rules then in force. The governed model feeds the three regulatory reports, which have different frequencies and data models. Verification runs after each hop. Finance can override an individual figure at the report stage, with a mandatory reason and a named approver. Source systems Documents, spreadsheets and connections into FP&A tools Verification Governed model Dated classification mappings: a figure filed two years ago still explains under the rules then in force Verification Three regulatory reports Different frequencies and data models Finance override Reason and named approver
Data flows top to bottom, from source to report, with verification after each hop, so every figure can be traced back along the same path.

Scope was initially set to prove coverage. 180 items on the daily report: 60 each from the inflow, outflow and supplemental tables, selected so all thirteen product categories were sampled. 25 records per selected schedule on the consolidated report. 45 high-risk fields on the systemic indicators report.

A large share of those items had no formal system of record and no authorized distributor. No distributor means nobody is accountable for the quality of the figure. Naming report and data owners became a workstream of its own and had to be escalated to the project sponsor.

Three design decisions followed.

  1. Classification is held as dated mappings rather than hard-coded, so one figure can classify differently per report, and a figure filed two years ago can still be explained under the definitions then in force.
  2. Verification runs after each hop. It confirms the data arrived intact, but not why a figure is what it is.
  3. Stricter lineage to prove where the figure came from, and every calculation applied along the way. The question that actually arrives is why a figure moved since last quarter when the business did not change, and the answer is almost always a rule change or a manual adjustment.

One requirement came from Finance rather than from us. They needed the ability to override an individual figure, built on day one with a mandatory reason and a named approver.

The outcome

Report creation itself did not yield the biggest time saving, because of legacy system constraints. End to end, from creation through reconciliation and publication, the cycle ran roughly 40% quicker, thanks to a reduction in manual effort. That saving recurs every cycle rather than once at build time.

In addition to the operational efficiencies, risks related to manual adjustments and gaps in lineage decreased.

Reporting that cannot explain itself?

Governance and modeling engagements begin with one figure nobody can trace.

Get in touch About this practice