How I Built ComeBk

ComeBk began as a question: what if commercial lending teams had the same analytical rigor as institutional investors — without the headcount?

The starting point

Commercial lending is judgment-heavy and data-fragmented. Teams spend more time assembling information than interpreting it. The opportunity was not to replace judgment — it was to remove friction between data and decision.

First principles

1. Map the workflow before automating it — credit decisions have steps that must remain human
2. Build for incomplete data — production environments are never as clean as test datasets
3. Explainability is non-negotiable — every output must survive a credit committee conversation
4. Architecture over models — the product should outlast any single AI vendor

What shipped first

The earliest version focused on financial statement analysis — the highest-friction step in most commercial credit workflows. Automated extraction, normalization, and ratio analysis with human-readable outputs.

What came next

Workflow integration — embedding intelligence into the tools teams already use rather than asking them to adopt another dashboard.

Lessons

• BFSI customers adopt depth, not breadth
• Trust is earned through reliability on edge cases, not demos on perfect data
• The best product decisions were the ones that removed steps, not added features

For the broader philosophy behind this approach, read Building for the Long Term.