Work · Credit · Decisioning
4 min Median time to a loan decision
A working capital lender had files sitting two days in a review queue. We built the decision engine that clears the median file in four minutes, inside the same credit policy.
My worry was drift, so every Friday I pull a sample and replay the week's decisions against policy. That ritual only works because the engine logs everything. My analysts spend their day on the borderline files now, the ones actually worth an hour of thinking, and an applicant doesn't get punished for arriving on a busy Tuesday.
4 min
Median time to decision
2 days
What it replaced
100%
Inside the existing policy
01The challenge
Applications queued behind a small credit team. The median file waited two days for a first decision, and applicants did not wait with it. The policy was sound. The bottleneck was throughput.
02The system
The policy, encoded
The same rules the credit team ran on paper, expressed as configuration they own. A threshold change is a configuration edit, not an engineering request. Every version is kept, so the policy in force on any date is a fact.
Data pulled, not chased
Bank data, application fields, and history assembled the moment a file lands. What used to be a day of email is ninety seconds of pipeline, and the engine starts from a complete picture.
Three outcomes
Approve, decline with reasons, or escalate to a human with the analysis done. Clear files leave with their reasons stated. Borderlines reach an analyst with the full workup already attached.
Every decision logged
Inputs, rule path, outcome, captured for every file. Any decision can be replayed months later, declines included. When someone asks why, the answer is a record, not a recollection.
Nothing waits in limbo
An incomplete file bounces back immediately with exactly what is missing. Anything the rules cannot settle goes to an analyst instead of stalling.
Reviewed every week
The client's credit officer pulls a sample of the week's decisions and replays them against policy. The logging makes that ritual cheap, so drift gets caught early.
03The impact
4 min
A complete file now gets its answer in four minutes instead of two days. The fastest answer usually wins the deal, so that gap is revenue.
Same policy
Every decision runs the credit policy the team had already approved, nothing relaxed for speed. The risk position is unchanged, so the speed is safe to scale.
Every file
Each application ends in a decision or a handoff to an analyst, with a record that can be replayed. Nothing leaves the system unexplained, which keeps oversight cheap.
The four minutes, step by step Expand
Where the time goes
- 0:00File landsThe application enters complete or is bounced back immediately with exactly what is missing, so nothing waits in limbo.
- 0:00 to 1:30AssemblyBank data, application fields, and history are pulled and normalized. The step that used to be a day of email is ninety seconds of pipeline.
- 1:30 to 2:00ScoringThe file runs the policy: capacity, stability, exposure, exceptions. Seconds, because the policy is code, not a queue.
- by 4:00Decision or escalationApprove and decline leave with stated reasons. Borderlines route to an analyst with the analysis already attached.
Keeping it honest
Speed without drift is the whole design brief. Policy configuration is versioned, every decision is replayable, and the client's credit officer samples each week's decisions against policy, a ritual the logging makes cheap.
What the queue became
Analysts see only borderline files, delivered with the full workup. The two-day queue did not get faster. It stopped existing for the files that never needed it.
I used to spend my day confirming clean files were clean. Now a file only reaches me when there's an actual argument to have, and the workup is already on my screen when it does.
Next step
Have a queue standing between you and the deal?
Sixty seconds to scope, then a 30-minute call. Bring the workflow, we bring the business case and the controls plan.
More closed work