Manually doing processes yourself in the early stages of building a marketplace is crucial to understanding what needs to be built, rather than automating everything too soon. When we started onboarding cleaning professionals, we asked them numerous questions over the phone and in person, followed by a test clean. This process led to an acceptance rate of only three to five percent, meaning many candidates never made it onto the platform.

We realized we needed to learn how to clean properly before bringing professional cleaners onto our platform, so I decided to get a job at a cleaning company to gain firsthand experience.

After a few weeks, I quit the cleaning job because I wasn't learning anything new, but it gave us confidence that we were on the right track with our platform.

Working in the cleaning job was informative; I learned how cleaning companies operate and identified inefficiencies in their booking and scheduling processes.

Our interview process includes giving candidates messy data sets to analyze, assessing their ability to identify and address underlying assumptions in the data.

When we opened our platform to a wider audience, we found that self-selection by users led to better engagement and feedback than our initial closed beta approach.

We conducted surveys with our customers to understand why they bought our product, which helped us tailor our marketing messaging.

Our early debate about the quality of data on Facebook led us to prioritize clean and accurate information, which helped establish a reliable platform that users could trust.

Our platform not only evaluates models but also helps partners select the best-performing models based on user feedback.