Comparables, rebuilt for the multimodal era

The problem

Picket Homes is a real-estate analytics platform used by single-family-rental investors. Comparables — the picks of similar homes used to price an asset — were the most-used surface on the product and the most-complained-about. Inputs had grown from a handful of fields (beds, baths, square footage) into a multimodal mess: structural features, condition signals, photos, geo proximity, school quality. The old UI couldn't keep up.

My role

End-to-end. I owned the redesign as Product Lead: scoping the problem with the analyst customers, redesigning the filtering surface, working with the data team on the new feature set, and shipping the engineering changes alongside the team.

Approach

• Cut the choice space. The previous UI exposed every available filter. We cut to the five filters that explained 80% of analyst overrides, and made the rest accessible behind an "advanced" reveal.
• Make AI legible. Each suggested comparable showed why it was included — structural match, location score, recent transaction, photo similarity — instead of a single opaque score.
• Tighten the loop. Analyst overrides flowed back into the ranking model so the next session learned what the user trusted.

Outcome

The redesigned comparables module contributed to a 15% engagement lift on the landing and detail page workflows. Override-to-acceptance time dropped meaningfully, and the surface stopped being the top complaint in customer interviews.

What I'd bring to your build

This is the shape of a lot of early-stage product work: a feature that grew organically until it was the bottleneck, and the rebuild needs product, data, and engineering moving in lockstep. That's the engagement I take.