AI Work & Freddie HR — Product & UX Teardown
Turning a strong AI engine into a usable, trustworthy product through onboarding, IA, and UX fixes.
AI Work is a platform of role-specific AI workers — Freddie (HR), Luca (finance), Saras (reports), Orion and Hermes (markets) — each scoped to one domain and built to produce structured, professional output. The AI engine is genuinely strong: Freddie generates accurate job descriptions and multi-dimensional, evidence-grounded candidate scoring that compares well with dedicated ATS tools. This teardown diagnosed why almost nothing in the experience guides users to those moments — and laid out the roadmap to fix it.
The defining problem was the gap between AI capability and product experience. A new user signs up and is greeted by '$0 Credits' before any value is delivered; the strongest output in the product — candidate scoring — is buried in a 14-column horizontally-scrolling table with the Overall Fit Score as the last visible column; and a hidden Email Identity dependency creates a hard blocker mid-workflow when scheduling interviews. Unverified landing-page statistics (1.2M MAU, 98% time saved) further undermined B2B trust. Most users would abandon before reaching the product's best moment.
How it came together
Webapp Review
Assessed first impression, app positioning, navigation, and conversion across the platform — finding a six-step path from landing page to first AI output and no product proof above the fold.
Freddie HR Deep-Dive
Walked the full recruiter journey (create job → set criteria → upload CVs → review scores → schedule) as an HR manager screening 50 applicants, rating AI output quality feature by feature.
Prioritized Roadmap
Translated findings into a P0/P1/P2 roadmap weighted by impact and effort, from a one-line auth-state CTA fix to a Workspaces model and criterion weighting.
Wireframes & Copy
Produced structural wireframes for the landing page, platform hub, and Freddie dashboard, plus a ten-row copy-improvement table replacing internal jargon with value-led language.
What we shipped
- Redesigned candidate results as a ranked card view sorted by Overall Fit Score, with the 14-column table retained as a secondary detailed view.
- Rebuilt the first-login experience: gifted-credit framing in human terms ('49 credits ≈ 120 candidates') plus a three-step onboarding checklist.
- Made Email Identity setup a required onboarding step to eliminate the hidden scheduling blocker.
- Introduced a Workspaces model so agencies and multi-entity businesses can manage multiple company contexts under one account.
- Added criterion weighting (Critical/High/Medium/Low) to scoring, plus benefit-led landing copy, real product screenshots, and attributed social proof.
Results that mattered
concept differentiation rating
AI output quality rating
UX & onboarding rating (pre-fix)
release-critical P0 fixes identified
prioritized roadmap items (P0–P2)
The single highest-ROI investment wasn't more AI — it was getting users to the existing AI output faster and presenting it clearly. When the strongest feature in a product is also its most under-marketed and least usable asset, the work is product and UX, not model quality.