Project leadership & Design management
Spreadsheets, Talkdesk, Figjam, Figma
Honor, 2023
Managed and supported solo Product Designer in manual pilot & initial product launches, filling in gaps in PM, Ops, and Design to successfully test and iterate a new staffing model.
Org background
Honor is revolutionizing the home care industry by building a platform that connects professional caregivers with families looking for quality home care.
Project context
When signing up for home care, customers are really looking for one thing: high quality caregivers who show up reliably. This is clear in the data as well. The two highest indicators of client churn are negative in-home experiences and caregiver swaps in the first two weeks of care.
Staffing at scale is incredibly challenging:
Matchmaking against shifts is an O(n^m) problem, where humans become increasingly inefficient when they no longer know the bios and latest updates of every client and caregiver.
Clients have high expectations on the quality of the service that span up to 24/7 presence in their home, it is more comparable to finding a romantic partner than getting an Uber or Doordash.
Client A
Problem Alignment and Opportunity
The problem: Honor's solution up to this point was a human staffer with the aid of advanced matchmaking and communication tools. The staffer still largely chose which caregivers to contact and to put in the home, and the technology just made their job easier. As we scaled, match quality and operational efficiency worsened.
The opportunity: while it's an unsolved problem for an algorithm to rival the home care matchmaking capabilities of a full-time staffer, this is offset by it's ability to optimize matches at scale, meaning that more clients more often get the caregiver that best aligns with their schedule and skillset needs.
Vision & Solution definition
System-driven staffing
As the most tenured person on the staffing problem and working with a relatively new team, I took lead in identifying which elements in the solution space had been previously attempted and why they were or weren't successful to keep the team focused on more novel approaches.
From there, the Design Lead proposed a product vision that moved a staffer over to a fully optimized system in phases. To start, the team would build an advanced ML algorithm that would provide staffing recommendations across an entire market, and the staffer would maintain oversight and override control. But over time, Ops and Product leadership would collaborate on reducing the manual overrides.
To bring together this vision, Design played two key roles:
Design was heavily involved in the match-making algorithm to identify data elements to prioritize in our experimenting with in our models, as well as help the data team understand the algorithm's shortcomings during it's pre-launch development.
Design acted as a staffer for several weeks to figure out the workflow for interacting with the model and start designing the tools that would best support the future staffing processes— adding inputs, creating a draft plan, and facilitating outreach and assignment.
HMW build ops tools that center around algorithmic recommendations
What we built
There were a number of pages that needed to be designed to provide a market-level view of staffing with entry points to easily update staffing inputs to feed the algorithm. This was somewhat straightforward design work. When it came to reviewing draft plans and starting outreach, design became more challenging.
For example, one of the existential challenges of moving to an algorithm solution is that there still had to be a human conversation with a caregiver. Without one, the caregiver, who largely acts like a freelancer between agencies, would rarely accept a commitment to take on a new client ongoing.
Design was challenged to figure out how a largely automated system could introduce tasks to staffers, whose outcomes would become a variable into the algorithm.
Our Design Lead presented a dynamic work-queue for outreach that prioritized and differentiated clients by the difficulty of restaffing, ensuring conversations were occurring in the most logical order. Additionally if the plan received enough declines, re-planning was automatically suggested.
Market view of all caregivers (aka Care Pros) used to determine supply health
Market view of all clients used to get a sense of regional demand
Select shifts to include in a market plan. Some are included by default like unstaffed shifts, or suboptimal matches etc.
Generate and review a draft plan
Outreach tracking
Outreach tracking early iterations. We ended up splitting it into a queue and detail view instead of a single page due to volume of content and actions.
In Fall 2023, the team launched a manual pilot with just the algorithm running side-by-side to normal ops processes and proved higher quality schedules across the market.
In spring 2024, the team launched Market Planner to incorporate the algorithm into ops daily process and reinvent our staffing practice from the ground up.
In summer 2024 Market Planner was rolled out to all markets.
Decrease in time to staff open shifts. Staffers used to spend majority of their time staffing, now they spend that time with higher touch outreach to top caregiver matches.
Higher job acceptance rate for shifts recommended by algorithm and delivered via new outreach process.
< Previous