How I helped a ML focused product team go from underperforming to over-achieving
In 2018 I took on a consulting project at Meta HQ in Menlo Park. I embedded within a team that builds internal recruiting tools. As Product Lead, my job was to uncover the reason for low adoption of their first machine learning feature. I would help revise product strategy and work with the team to build and execute on a new roadmap.
The challenge: Engineering-driven product team had launched a ML feature that was met with low enthusiasm and adoption.
Step 1: Understand
- The team’s priorities and strategy: the team was measured on their ability to increase the number of candidates identified using ML—as a proxy for recruiter efficiency.
- The product and how it worked: I worked with the cross functional team to deep dive on the existing workflow, features and user experience.
- The users: I partnered with user research on studies to model user workflow and pain points.
User research was key to understanding the problem. We had good data on how users engaged with the web application. But we lacked a comprehensive picture of the sourcing workflow—especially tasks that were happening offline and in other channels. This generative research provided new insights on recruiter pain points and priorities.
Step 2: Identify the gaps
I did 24 user interviews across roles, regions, and seniority. Two clear patterns emerged:
- Fragmented workflow: Recruiters only spent about one-third of their time in our tool. They relied on a patchwork of spreadsheets, Notes, and Asana to manage status, priorities, and collaboration with hiring managers.
- Lack of trust: Recruiters had high-stakes, incentive-driven goals and leaned heavily on familiar, personal strategies. They avoided ML features they didn’t understand.
Step 3: Revise product strategy
We revised our product strategy to address these unmet needs.
- I led multiple workshops to synthesize our findings and brainstorm solutions.
- Through roadmapping and creative briefs I focused the team’s efforts in promising directions.
- I used prototypes and concept testing to validate new approaches.
- I shared updates with executives and cross functional teams to get buy-in on our revised strategy.
Results: We launched a sourcing dashboard, “co-pilot mode” and “explainable AI” features that drove dramatic results.
- 130% increase in claims within 30 days
- By year’s end, we had a 10X increase in claims from ML features
- The team exceeded it’s yearly goal by 9%, despite a rocky start
