Harbor

Harbor's merchant dashboard had grown one feature at a time for six years. Every team that needed a surface had added one, and the result was a product where the most common task — reconciling a settlement against a batch of payouts — took eleven clicks and two exports to a spreadsheet.

Starting with the spreadsheet

We spent the first two weeks watching finance teams work rather than talking about the dashboard. Almost everyone had the same workaround: export both views to CSV, join them by hand, and do the actual thinking in Excel. That spreadsheet was the real product. The dashboard was just where you went to get the data out of it.

That reframed the brief entirely. We were not redesigning a dashboard; we were trying to bring the spreadsheet inside.

One table, not eleven screens

The new reconciliation view puts settlements and payouts in a single table with the join already made. You can group, filter, and annotate in place, and the annotations persist for whoever picks the account up next month — which turned out to matter more than any of the visual work.

The first customer who saw the prototype asked when they could stop using their spreadsheet. That was week four.

The system underneath

The table needed a data-density model the existing components could not express, so we rebuilt the foundation: a proper type scale for tabular data, three density modes, and a set of table primitives that four other teams have since adopted for their own surfaces.

Where it landed

Reconciliation went from eleven clicks to two. Support tickets about onboarding dropped by roughly two thirds in the first quarter after launch. The spreadsheet workaround has not disappeared entirely, and I do not think it should — but it is now the exception rather than the method.