Addressing data anomalies with quick-win improvements.
Role
Led UX within cross-functional squad
Team
Squad (PM, Developers), Data analyst, Sales, UI Designer, UX Researcher
Challenge
Property investment platform Crowdhouse prompted users to submit buying preferences to power property recommendations. The location preference question, however, made broad location selection too easy.
Users were defaulting to language-based Swiss regions, but their actual preferences, shared directly with sales, were more specific. The data was systematically less accurate than it needed to be, skewing property recommendations.
Impact
75% of users made smaller Grossregionen selections post-release
Compared to the large proportion previously defaulting to the whole German-speaking region, this was a significant shift toward more considered, accurate location selection.
More accurate location data fed better property recommendations
Sharper buyer preferences meant the recommendation engine had more faithful signals to work with.
A refined location model at Grossregionen level with a map-led selector so buyers could orient spatially.
The importance of ongoing evaluation.
I'd been leading regular sessions where we evaluated users' saved investment preference data, alongside feedback from sales.
We discovered that the location data feeding our property recommendations didn't always reflect what people actually wanted.
We had learned that many property investors wanted to focus their search on certain language areas, so made this option prominent and easy to select online. But in sales conversations, the same buyers described more specific cantonal preferences.
The power of diverse knowledge to drive quick decisions.
With many other team priorities to juggle, we worked together to explore a quick improvement. I led a kick off where we pulled apart the issue to consider how we could guide users to more accurate location selections without costing them too much effort.
Input from our data specialist led us to consider 'Grossregionen' as a better, balanced level of granularity.
Shaping direction with our UI Designer.
Working with our UI Designer, I shaped the direction of each option and pushed for the team to review all three together, rather than immediately defaulting to one.
Selector chips
We had an autocomplete component available but feared it would require more mental overhead to recall all possible regions.
Nested checkboxes
Potentially simple to build, but space-heavy and less visually coherent with the existing UI.
Checkboxes and tabs (Our choice)
The pragmatic call for v1. This made use of an existing pattern. We also wanted to test whether using tabs rather than a toggle to switch between regions and cantons would improve visibility and selection.
The importance of robust evaluation metrics.
Post-release analysis showed a significant increase in location data granularity vs the older design — a success for our main hypothesis.
However, from carefully monitoring all aspects of question completion, we also caught an unintentional consequence.
The data showed:
Although we knew the majority of users focussed on canton level selections, we still wanted to resolve this to avoid missing out on more specific preferences.
As a result, we planned a future task to continue iterating on the design to see if we could get the best of both worlds.
Lean experiment, clear result
Within a tight timeframe and without significant development requirements, the experiment worked. Location data became more specific, better reflecting real buyer preferences.
More accurate location choices enabled better property recommendations
Our property recommendations were fed by preference data for each potential buyer. The more granular location choices we received meant we stood a better chance of highlighting properties that would appeal.
No drop in completion rate
Increasing data accuracy without adding friction or drop-off was the core design challenge. Completion of the location question held steady, so the redesign didn't put users off answering.
Success metrics: behaviour vs intent
When we introduced language region selection, our success metric was the proportion of users who chose by region. We'd just never questioned whether that choice reflected what they actually wanted. The data looked right. It wasn't. It took comparison with sales conversations to see the issue.
Evaluation as a habit
Actively monitoring our work's impact meant we could spot and act on this issue. That meant not just measuring data, but also bringing in sales feedback.
Collaboration as a shortcut
This improvement only happened at speed because we brought everyone into the same conversation: data insight, sales context, and a practical eye on what we could actually build. This is what turned it into a 'quick win' rather than a lengthy project.
More featured work
Rapid exploration & a pragmatic staged release.
I led work to uncover complex energy information requirements, quickly designing platform adjustments for sellers, buyers and staff, allowing our first German sellers to list properties without delay.
Research-led redesign of a bloated staff portal.
I transformed an outdated and confusing funding application platform, significantly reducing time for staff to reach priority features and eliminating unnecessary tool switching.
