Darwin Homes

Building a recommendation tool for home repair vendors

Speeding up service requests by 30%

My Role:
Product Design Lead
Darwin provides maintenance services to residents living on their properties. I helped Darwin select the right vendor for the job. I worked with product, engineering, and data to speed up service request speed by 30% and reduce vendor decline rates by 15%

Residents rely on their property management company to provide maintenance as quickly as possible.

Darwin’s role as a property management company involves fulfilling maintenance requests for its residents. Residents file a maintenance request, and Darwin’s experts triage the request and dispatch a vendor to the property. A vendor can then accept or decline a job or work order.

Selecting the wrong vendor for a job is costly for Darwin, its clients, and the residents.

Darwin creates an average of 4000 work orders every month. Out of 1400 declined work orders a month, 33% (402) were declined due to skillset or service areas mismatch. We sent jobs to vendors that were outside of their service area or expertise. Assignment mistakes increase the time it takes to provide maintenance. It also affects Darwin’s reputation while costing the owner more. This compounds when the resident experiences an issue that requires more than just a quick fix.

There are many factors when picking the right vendor.

The right vendor must have the right service area, the right trade category, be available, provide quality work, etc. To complicate things, vendors often change the services they provide based on their staffing. Your local one-stop-shop may stop providing electrical services if they’re down an electrician or two.

Good vendor selection requires industry knowledge, but expertise is expensive. How can we provide great vendor selection at scale?

Dispatching vendors requires knowledgeable dispatchers to triage the problem. After speaking with the resident, our dispatchers must know what the issue may be. Attracting this kind of talent is difficult and does not scale well. We want to use our software, Origin, to make the decision intelligently.

We designed a powerful Vendor selection algorithm to pick vendors based on non-negotiable attributes, overrides, and past performance.

We separated vendor data into 3 categories.

  • Non-Negotiable Attributes- the minimum requirements a vendor must meet

    • Service Geography - the area the vendor does work in.
    • Trade categories - the type of work the vendor does.
  • Overrides - these attributes have the largest amount of weight in a vendor score

    • Home warranty - vendors that may serve a home warranty at no additional cost.
    • Owner preferred - vendors that owners indicate preference for.
    • New vendor - vendors that are newly onboarded are given priority to ensure they receive work before they are given a performance score.
  • Past performance - these attributes indicate the quality of our vendors

    • Capacity - the number of work orders a vendor can take at once.
    • Response time - the time it takes for a vendor to respond to a job.
    • Acceptance rate - the percentage of work orders a vendor accepts

We simplified our UI to make vendor selection easy- aiming for complete automation.

Our early designs provided lots of data to help a human operator make the best decision. However, if our goal is complete automation, we want to remove human intervention as much as possible. Our final designs were simple, which allowed us to spend more time refining our algorithm and less time crafting an experience that do not meet future goals.

Collecting data when we deviate from our top selection will build a better algorithm.

As our recommendation engine evolves, we will face edge cases that are not accounted for. Users have the option to select other vendors in the event that our recommended vendor does not meet their criteria. We’ll monitor these exceptions to build a better system that can eventually function solely based on data.