Investor Loan Platform

A bank was running its loan operations in Excel. Deals worth millions of dollars, no audit trail, no validation, and no way to tell who changed what.

Three teams had already tried to replace it and failed. I spent a while figuring out why before I designed anything, and the answer was the same every time: they all tried to improve parts of the job the loan officers never asked to have improved.

I started by mapping the whole loan lifecycle on a whiteboard, from application through to audit, so I could see where the spreadsheet was actually holding things together and where it was quietly failing.

Then I sketched the order builder rough and fast, because the thing I needed to test was the sequence, not the styling. Once the steps held up, I wrote the flow down as four moves: pick a deal, choose Cashless Roll or Book-builder, add an order or attest to one, update the grid. That is the version loan officers recognized as their own job, and it is the one that got adopted.

Validation runs inline as you type, so a malformed entry gets caught where it happens instead of at submit. You never see it when the entry is fine.

I replaced filter panels with predictive search. Loan officers do not think in filters, they think in name fragments and deal codes, and they want the deal on screen in two keystrokes. That is a Bloomberg habit and it was the right one to design for here.

The order builder walks you forward: pick the lender first, then the terms open up. It looks restrictive in a screenshot. In practice it stops people building an order against nothing.

The audit trail sits right next to the record instead of in some admin tool nobody opens. That single decision is why compliance ended up arguing for the platform rather than against it.

My first version copied too much of Excel's structure. I was trying to keep the change small, and I ended up with something that was worse than the spreadsheet at being a spreadsheet and worse than the platform at being a platform.

I also built dashboards too early. Nobody starts their day on a dashboard. They open a specific deal, because a specific deal is what someone is asking them about.

I don't hold the source for this one, so the sheet below is measured off the delivered screens rather than exported from a token file. It is still the real thing: one institutional blue that only ever means "you can act here", and four status hues that appear twice each, once as a pill fill and once at full strength on the progress bar.

Nothing else gets colour. In a product where the numbers are eight figures, decoration reads as a bug.

They adopted it. Three teams tried before me and none of them got there.