Audit Evidence Surface
A semantic layer over six core banking systems turned the audit pack from a weekend assembly job into a live query, roughly halving the average audit cycle.
The challenge
A regional lender's internal audit team was fighting its own tooling every cycle:
- Every audit pack was assembled by hand from six source systems, usually over a weekend before Monday review
- Two to three staff were pulled off other audits each cycle just to build the evidence pack
- Data was often a week or more stale by the time the pack reached reviewers
- Auditors couldn't ask a follow-up question without requesting a fresh extract and waiting days
- The average audit cycle ran 9 weeks from kickoff to closing memo, well above the team's target
How it works
Turning the evidence pack into a live query, not a weekend build
Instead of automating the extract process, the underlying data was made directly queryable:
- 01
Built a semantic layer over the six core systems — loan origination, servicing, GL, collections, collateral, deposits — with consistent field definitions
- 02
Modelled the standard audit evidence requests as reusable queries instead of one-off extracts
- 03
Gave auditors direct, read-only query access to current data instead of static packs
- 04
Automated the recurring evidence pulls — sampling, reconciliations, exception listings — that previously took a weekend
- 05
Ran the new surface in parallel with the old process for two audit cycles to validate against manually built packs
- 06
Retired the manual pack-assembly process once the parallel run matched on both audits
What we built
Key capabilities
One evidence layer, six systems
Loan origination, servicing, GL, collections, collateral, and deposits sit behind one consistent query surface.
Reusable, not rebuilt
Standard evidence requests are modelled once as queries and reused every audit, not rebuilt from scratch.
Follow-ups in minutes
Auditors query current data directly instead of filing an extract request and waiting on it.
Current, not stale
Evidence reflects the current state of the source systems, not a weekend-old snapshot.
Before vs after
What changed in the audit cycle
- Audit cycle time
- 9 wks → 4.5 wks
- Pack assembly
- Manual weekend build → Live query
- Staff diverted per audit
- 2–3 → 0
- Data freshness
- Up to a week stale → Current
- Follow-up turnaround
- Days → Minutes
Business impact
What it changed
Audit cycle roughly halved
Average cycle time fell from 9 weeks to 4.5 across all 30 annual audits, with no change to audit scope or rigor.
No more diverted staff
The two to three staff previously pulled into weekend pack assembly stayed on their assigned audits instead.
Follow-up questions answered live
Auditors query current data directly, replacing a multi-day extract-request cycle with a live lookup.
Technology stack
“The pack didn't need to be built faster — it needed to stop being built at all. Making the data queryable did that.”
Keep reading