01
The Opportunity
Employers don't need another AI tool—they need an AI partner that works where they already work and understands their unique hiring needs. I worked closely with 2 UX Designers to launch Indeed Talent Scout, an agentic AI assistant that brings 595M job seeker profiles into employers' existing ATS workflows.
The Challenge
Create an AI agent that enhances recruiters' expertise rather than replaces them, works seamlessly within existing ATS rather than forcing workflow changes, and moves beyond text-only chat to provide rich, visual UI that makes AI recommendations quickly evaluable and actionable. With LLM capabilities emerging, Indeed had an opportunity to solve the match quality problem at scale by deeply understanding what employers actually need—not just what they write in job descriptions.
My Role
02
Scope & Strategy
For the MVP launching in September 2025, we prioritized deeply understanding employer intent and delivering quality matches.
Prioritization Criteria
High ROI at scale
Unique competitive advantage
Complex problems requiring AI reasoning
Leverage existing capabilities
03
Design challenge #1
Solving the Blank Canvas Problem
Unlike traditional UI with buttons and menus, conversational AI presents a blank canvas: employers stare at an empty chat box with no clear sense of what's possible.
Context selector
Two Critical Barriers
"What Can Scout Do?" — Employers arriving at an empty chat faced decision paralysis. Without visible affordances, they asked basic questions, hesitated, or abandoned before discovering Scout's value.
"Which Job Are You Talking About?" — Scout operates as an overlay, reading job context from the page behind. But employers couldn't tell what Scout understood, especially when managing multiple jobs with similar titles.
The Solutions
Progressive Discovery
Static starter prompts surfaced high-value actions ("Show top matches")
Dynamic prompts adapted to context—suggesting "Generate match criteria" for new jobs or "Show candidates" when criteria existed
GIF-based onboarding demonstrated workflows in 3-5 seconds with clickable example prompts
Explicit Context Controls
"@" referencing let employers tag specific jobs: "Show me matches for @Software Engineer Austin"
Clear indicator showed active context: "Referencing: @Software Engineer Austin"
Smart error recovery turned ambiguity into teaching moments: "I see you have multiple Software Engineer roles. Please reference a specific job by typing @..."
60%
of users who opened Scout engaged in meaningful conversation
25%
click-through rate on starter prompts
24%
week-over-week retention rate
04
Design challenge #2
Why Upfront Calibration Failed
Employers complained about bad matches, so I designed a 3-step calibration flow: add requirements, rank by importance, set thresholds. This failed spectacularly—6 of 8 employers abandoned within 90 seconds.
The more work you make me do upfront, the more perfect I expect the results to be."
Every setup step raised expectations, creating an impossible bar to clear. In a critical design review, I convinced Product and ML leadership to flip the strategy: accept 70% initial accuracy for zero-friction start. 75% abandonment was objectively worse than 70% accuracy with sustained engagement.
Four Pathways to Understand Hiring Intent
Auto-generated: Scout generates initial criteria from the job description by default. No setup required.
Conversational refinement: Scout asks focused questions based on what's missing in the job description. As employers respond, Scout updates criteria in real-time and shows what was added.
Document upload: Scout extracts requirements from recruiters' detailed intake docs or reference resumes and asks for confirmation, capturing nuanced intent that wouldn't surface through chat.
Direct editing: An Edit button opens a criteria editor with a match meter showing whether criteria are too narrow or broad.
Transparency features built trust: update cards showing exactly what was added, explainability snippets on candidates ("Matches 7/10 criteria"), and real-time pool size visualization.
It’s so easy and user friendly. It only takes me 5 seconds or so to type one sentence prompt and then Talent Scout generates 30-plus candidates almost immediately. It’s fast, and it's easy.”
05
Design challenge #3
Earning the Right to Recommend
The Problem
When employers complained about low-quality candidates, we initially focused on the matching algorithm. Research revealed something deeper: employers couldn't diagnose why their jobs underperformed—they were flying blind without competitive market context.
The core tension: employers needed diagnostic intelligence but were skeptical of AI recommendations they couldn't verify. "I think it's really loose, the basis for the data...I don't know how legitimate it is."
Critical constraints shaped our approach:
Recruiters work from their ATS as the master platform, not Indeed
Company policies and fixed budgets limit what they can change
They need suggestions as inspiration for advocating with hiring managers, not mandates
The Solution: Progressive Diagnosis
We designed a flow mirroring how doctors build patient confidence: observe symptoms, explain diagnosis, recommend treatment.
WHAT — Surface familiar performance metrics without jargon. Numbers alone don't tell the full story.
WHY — Provide competitive market context employers can't access independently. This transparency directly addresses skepticism.
HOW — Deliver ranked, actionable recommendations with estimated impact. Employers know where to focus first.
This staged approach prevented overwhelm while building confidence. We transformed suggestions from arbitrary to insightful—advocacy tools recruiters could use to persuade hiring managers.
06
Impact & Outcomes
Early Adoption Signals Strong Product-Market Fit
Talent Scout launched at FutureWorks 2025. Within three months of limited release, Talent Scout achieved 57% engagement among employers who opened it. Over 6,900 employer accounts were exposed to the product, with 1,619 actively trying it and 915 engaging meaningfully.
User Response
The FutureWorks 2025 demo revealed immediate product-market fit. Demo observers noted "mouths agape" when seeing Deep Sourcing run for the first time—one employer called it "riveting." Executives were particularly drawn to Job Optimization insights, with one planning to use the recommendations in union negotiations to improve job title searchability. Workday users immediately grasped the integration value: "That's one less browser tab for my people—that will simplify so much!"
Early customers reported measurable hiring gains. A recruitment specialist doubled hiring efficiency to 22.3% conversion (from under 10% before). An HR generalist doubled their candidate pool from 5-7 to 10-15 applicants for hard-to-fill roles. A clinical recruiter made 10 hires while saving 8 hours weekly, noting it took just "5 seconds to type one prompt" to get 30+ quality candidates.
As an HR Generalist, I don’t have time to dedicate solely to recruiting. If I was a full time recruiter I could maybe do some of this stuff, but even then I still would want something like this because it lets me reach candidates I can't otherwise get to. As far as the amount of time saved, gosh, I can't quantify it, but it's a lot.”
What Made It Work
These outcomes trace directly to four strategic design decisions:
Transparent AI reasoning built trust. Showing why candidates matched and how criteria affected pool size gave recruiters confidence to act on recommendations rather than second-guessing them, turning skepticism into adoption.
Weekly research and rapid iteration ensured real-world fit. Running user research every week allowed us to validate solutions against actual pain points and iterate quickly, keeping the product grounded in recruiter needs rather than assumptions.
LLM reasoning unlocked better matches. By leveraging deep AI reasoning to surface candidates beyond restrictive job descriptions, we achieved match quality significantly better than any Indeed system before it, broadening search while improving ranking accuracy.
Relentless simplicity drove daily usage. Recruiters consistently praised how streamlined and intuitive Scout felt, with intelligent guidance at every step. This simplicity reduced cognitive load and made AI feel like a natural part of their workflow rather than another tool to learn.
What I Learned
Building Indeed's first employer AI agent taught me that novel interaction patterns require extra attention to clarity. The context referencing challenge forced me to think beyond traditional UI and create education moments that felt helpful, not intrusive.
This fundamentally changed my approach: I now budget 30% of design time for trust-building mechanisms—not as nice-to-haves, but as core to the product experience. The transparent reasoning patterns I established for Talent Scout are now adopted across Indeed's AI products, validating that investing in clarity pays dividends at scale.
OTHER CASE STUDIES
Automated Hiring Workflows
Coming soon
Interview Evaluation
Coming soon
