Huzzle - Rebuilding Recruitment Around AI

Overview

What Huzzle is, and what I do here

Huzzle is an agentic, AI-native talent marketplace that sources, vets, and places the top 2% of global talent across 60+ roles in sales, operations, marketing, product, and engineering - handling interviewing, hiring, and cross-border payroll end to end. An AI recruiter runs the whole lifecycle autonomously, so clients make a perfect hire in around 11 days. Candidates apply once, complete a single AI-led interview, and get matched to roles with companies across the UK, US, and EU.

I joined the Talent Marketplace team in April 2024 as a Senior AI Product Manager, working directly with the CTO, Head of Engineering, and UX designer on the core product decisions that shaped how candidates are screened, matched, and monetized. I owned where product, AI, and the business model meet - building AI features that also improved per-placement profitability.

Industry

Recruitment / HR Tech

Model

Agentic, AI-native talent marketplace

Talent pool

300,000+ pre-vetted

Speed

Perfect hire in ~11 days

Stack touchpoints

Workable, HubSpot, Anthropic & OpenAI APIs

Focus

AI interviewing, evaluation, matching, pricing

The Problem

Vetting quality was high. The economics weren't.

When I joined, vetting depended on human-run interviews. Quality was high, but the funnel was expensive and slow: every candidate consumed recruiter hours, cost-per-acquisition stayed high, and manual interviewing capped how fast the marketplace could grow.

Two structural problems stood out. First, the human bottleneck made scale a hiring problem rather than a product one. Second, early churn - placed candidates leaving within 2–3 months - eroded client trust and the economics of each placement. I framed my work around three bets: automate the funnel without losing quality, sharpen match relevance, and rebuild pricing to reward retention.

My Role

What I owned

I led product for the Talent Marketplace's AI initiatives - owning problem framing, prioritization, and the build alongside engineering leadership. I was deeply embedded in technical product decisions, in constant contact with the CTO and Head of Engineering, translating business constraints into an AI architecture the team could ship.

• AI interview strategy & rollout
• End-to-end candidate pipeline design
• AI profile-matching logic
• LLM-based question generation
• GTM & Pricing Strategy
• Stakeholder alignment (Eng, GTM, Exec)

Initiative 01 - Interviewing

"Luna," the AI Interviewer

I proposed replacing the human-interview bottleneck with an AI-led interview, and partnered directly with the CTO and Head of Engineering to build it. The outcome was Luna, the AI interviewer that conducts structured interviews at scale - the evaluation engine inside Huzzle's AI recruiter, which runs sourcing, screening, and matching end to end.

The judgment call that mattered: I kept a human-interview path any candidate could request if they weren't comfortable with AI. Automating the default while preserving a human option protected fairness and candidate trust - and removed bias from first-pass screening without making the experience feel impersonal.

The most direct measure of Luna's impact is raw interview throughput. The human-run process topped out at roughly 800 to 1,000 full interviews a month; Luna now conducts around 12,000 a month, each a full 20 to 30 minute structured interview, a step change no amount of recruiter hiring could have matched.

Initiative 02 - Pipeline & Matching

An AI-native candidate pipeline

I redesigned the end-to-end pipeline so each step that didn't require human judgment was handled by AI - and the steps that did, became sharper.

1. 01 ApplyCandidate submits resume via Workable
2. 02 AI screeningAI scans and screens the resume automatically
3. 03 AI interviewLuna conducts a structured, role-specific interview and AI evaluates the candidate
4. 04 Talent poolSuitable candidates added to the pool
5. 05 AI matchingAI matches against the client ICP

Resume-tailored questions via a two-model pipeline. Instead of a fixed script, I introduced interviews where the questions are generated for each individual candidate. The design chains two LLMs: Anthropic's Claude builds the prompting layer, which then drives OpenAI's models to produce role- and resume-specific questions. I defined the approach and partnered with engineering to ship it, so every interview adapts to the person in front of it.

AI matching. On the sourcing side, manual candidate review was the bottleneck: every client ICP sent someone combing the talent pool by hand. I proposed an AI matching step instead. It reads each client's requirements and job description, scores them against our 300K+ talent pool, and returns a shortlist of the two or three strongest candidates.

I set one constraint on purpose: keep that shortlist to two or three. A marketplace runs on confidence, not volume, so a handful of high-fit profiles a client can act on beats a long list they have to sift through. From there the client picks who to meet, the interview is scheduled once the candidate agrees, and a placement follows acceptance.

The impact showed up on the numbers a marketplace lives on:

• Active placements up 159%
• Monthly placements up 57%
• Time-to-placement down from 24 to 16 days
• Placement rate up from 60% to 85%, the cleanest proxy for cost-per-hire
• Average placement tenure up from 2.75 to 4.5 months

Initiative 03 - Monetization

Pricing that rewards retention

The problem: churn was high - candidates often left within 2–3 months of placement, which hurt both client trust and per-placement economics. The existing model, Talent on Demand, was a recurring monthly subscription priced as a share of salary.

I introduced Direct Hire, a placement-fee model in two flavors. On Huzzle Standard, the client pays a one-time fee and the talent joins their payroll from day one. On Huzzle Pro, the client pays the talent's agreed rate while Huzzle runs the payment, backs it with three replacements inside six months, and charges a management fee that keeps revenue flowing even after the hire is made. Alongside the recurring Talent on Demand model, these are the pricing options Huzzle runs today.

Result: the new pricing lifted revenue per placement by 50%.