Context
A regional bank with about $6 billion in assets had been losing small-business loan applicants to fintech competitors. The bank's internal SLA for loan origination — from application submitted to funds disbursed — was 10 business days. The actual median was 11. Competitors were advertising "approval in 48 hours." The board had greenlit a digital transformation program; they needed a disciplined analyst to run the diagnosis and vendor selection before writing any cheques.
Challenge
This was not a simple automation project. The loan-origination flow crossed seven teams (branch, underwriting, risk, compliance, legal, operations, treasury), used three systems that didn't talk to each other, and was governed by regulatory requirements that couldn't be skipped. The risk of transformation theatre was high — buying shiny software and changing nothing operationally.
Three stakeholder groups had different definitions of success:
- Retail banking wanted speed for customer experience.
- Risk and compliance wanted no loosening of controls.
- Finance wanted a payback inside 18 months.
Any solution had to satisfy all three. That shaped the approach.
Approach
I structured the program into three clearly sequenced phases so decisions weren't made under time pressure.
1. Diagnose (weeks 1–4)
I mapped the as-is flow across all seven teams in BPMN — 42 activities, 18 hand-offs, 6 approval gates. I pulled 12 months of application data and timestamped each hand-off. The baseline findings:
- Median cycle time: 11.2 days; P90: 19 days.
- Underwriting was only 2.4 days of that. The rest was queue time, document chasing, and re-keying between systems.
- 36% of applications were re-keyed at least twice across the three systems.
- The compliance check — which the team believed was the long pole — was actually only 0.9 days on average. A myth the whole program nearly optimized for.
2. Design & select (weeks 5–8)
I drafted the to-be process (30 activities, 9 hand-offs, 3 approval gates) and used it to build a requirements matrix of 62 line items. The matrix became the backbone of the RFP. I ran a structured vendor evaluation: seven candidates on paper, four invited to demo day, scored on a weighted rubric co-owned with the steering committee. The winning vendor scored highest on integration depth and on-site change-management support, not on the prettiest UI.
3. Build & pilot (weeks 9–16)
I was the business-side product owner through build. I wrote the user stories, ran the three-times-weekly standups with the vendor and IT, maintained the RAID log, and held a weekly demo for the steering committee so no nasty surprises piled up. Pilot launched at one branch in week 14 and ran for two weeks before rollout.
4. Hypercare
For the first eight weeks post-launch I built a daily dashboard: applications started, applications completed, median cycle time, number of exceptions, user feedback tickets. Daily standup at 9:00 with operations until metrics stabilized.
Solution
The technology was a new loan-origination platform from the selected vendor, integrated to the core banking system and the credit bureau APIs. But the program's result came as much from the process and policy changes layered on top:
- Parallel, not sequential, reviews. Compliance and credit risk reviews now run in parallel on the same application, not in series. Saved ~2 days alone.
- Risk-tiered underwriting. Low-risk applications (clearly within policy) auto-approve; the underwriter spends attention on the 20% that actually need human judgment.
- Single source of customer data. Eliminating the re-keying between the three systems cut 1.6 days out of the average cycle.
- Steering committee KPI dashboard. Adoption, cycle time, and exception volume tracked weekly. When one branch fell behind at week 6, we saw it and fixed it in week 7, not week 12.
"We'd bought loan-origination software twice before and never moved the needle. This time we didn't just buy software — we re-designed the process around it. That's why it worked." — Head of Retail Banking
Impact
Finance confirmed payback on the program (software + implementation + internal cost) inside 14 months — four months ahead of the board's target. Risk and compliance signed off that no controls had been weakened; two had been strengthened.
Lessons
- Map the process before you buy the software. The requirements matrix was only possible because we had a to-be model. Without it, RFPs turn into feature beauty contests.
- Measure the whole chain, not the step everyone suspects. Compliance took 0.9 days, not the mythical 5. Transformation projects get derailed by folklore; data punctures folklore.
- Change management is a product, not a phase. The hypercare dashboard and the weekly branch stand-ups did as much for the result as the software did.
- Ship steering-committee updates weekly, not quarterly. Decisions made in small weekly doses don't pile up into a crisis.