Banking and financial services · Credit risk
Compressing credit file preparation in a commercial lending team
Analysts assembled credit files by reading statements and agreements and re-keying figures into a spreading template. Covenant terms were transcribed by hand and errors surfaced at committee.
- Company shape
- Mid-market commercial lender, 500 to 1,500 employees
- Timebox
- 11 weeks
- Systems
- Loan origination system, Document repository, Spreading templates
IllustrativeThis write-up is a composite drawn from work of this shape. It is labelled because presenting a composite as a single engagement would be a fabrication.
Before
The state we started from
Analysts assembled credit files by reading statements and agreements and re-keying figures into a spreading template. Covenant terms were transcribed by hand and errors surfaced at committee.
The decision
What had to be decided first
The value map named analyst hours per file, the Head of Credit as owner, and dual-key comparison against a held-out sample as the validation method. Governance required attribution to the source page for every extracted figure.
Built
What was actually built
An extraction and assembly interface with page-level source attribution, a confidence flag per field, and mandatory analyst confirmation before the file can be submitted. Permissions inherit from the origination system.
What changed
Preparation time per file fell from 6.5 hours to 2.75 hours, at equal committee rework rate
- Measurement window
- Measured over 120 files across eight weeks
- Methodology
- Held-out comparison sample prepared under the previous process during the same period. Rework tracked at committee. Analyst confirmation time is included in the after figure.
What did not work
Automated covenant interpretation was cut after the evaluation set showed the failure mode was confident and wrong on non-standard drafting. Extraction with attribution stayed; interpretation went back to the analyst.
Systems involved
- Loan origination system
- Document repository
- Spreading templates
More from the library
Other workflows with a measurable baseline
Manufacturing · Finance
Reducing month-end reconciliation review time in a mid-market manufacturer
Review time per close cycle fell from 38 hours to 11 hours
9 weeks
GCC in India · Shared services and IT
Clearing a three-quarter internal tool backlog in a GCC
Nine tools published in one quarter against a prior run rate of two, with median request age falling from 9 months to 6 weeks
One quarter, first wave
Healthcare · Clinical operations
Reducing clinical documentation time in an outpatient service
Documentation time fell from a median of 11.4 minutes to 6.2 minutes per encounter
10 weeks
Next step
Have a workflow of this shape?
Bring the workflow and the number you would want to move. We will tell you whether it is a good first case, including when the answer is no.