Recruiters were sourcing blind — and the market had moved on
Oorwin was a full-stack recruiting platform — ATS, CRM, and analytics — serving staffing agencies and enterprise talent teams. When I joined as PM, the platform had strong candidate management capabilities but a critical gap: real-time sourcing.
Recruiters were leaving the platform to manually hunt for open jobs and candidate resumes on external job boards — Indeed, LinkedIn, Dice, Monster — copying data by hand into Oorwin. The platform was a system of record. It wasn't a system of intelligence.
Every day, recruiters spent 2–3 hours outside Oorwin manually harvesting job postings and resumes. Data entered the system stale. Personalisation was impossible. And the platform's value proposition — having everything in one place — was being undermined by the sourcing workflow that sat entirely outside it.
The opportunity was clear: if Oorwin could bring real-time job and resume intelligence inside the platform — personalised, AI-ranked, and continuously refreshed — recruiters would have no reason to leave. And competitors weren't doing this yet at the mid-market price point.
Validating the idea before writing a single spec
Before pitching the build to engineering, I ran a two-week discovery sprint to confirm the opportunity was real, sizeable, and technically feasible at a price point Oorwin could sustain.
The decisions that determined whether this would scale or stall
Several critical decisions shaped whether the Harvester would be a niche tool or a platform-defining feature. Each one involved tradeoffs I had to own explicitly.
| Decision | Options considered | Choice | Rationale |
|---|---|---|---|
| Data sourcing architecture | Build crawler vs third-party APIs | Third-party APIs | 12-week faster to market; better data quality; lets us focus engineering on the AI layer |
| Personalisation model | Rule-based filters vs AI ranking | AI ranking | Filters require recruiter configuration; AI learns from behaviour and improves without effort |
| UX entry point | Standalone module vs embedded in workflow | Embedded in workflow | Adoption requires zero friction; harvested jobs/resumes appear inline with active requirements |
| Resume harvesting scope | V1 include vs defer to V2 | Deferred to V2 | Job harvesting alone was a complete, valuable V1; resume harvesting added 6 weeks of scope |
| Pricing model | Add-on vs bundled in platform | Bundled | Maximises adoption and platform stickiness; add-on creates a paywall that slows discovery |
"The AI layer was the bet. Rule-based filters would have been a commodity. Personalised ranking was the moat."
— Vamshi, reflecting on the core product decisionReal-time AI sourcing, embedded where recruiters already work
The Job/Resume Harvester surfaced personalised, AI-ranked job postings and candidate resumes from across the web — directly inside Oorwin, in real time, matched to the recruiter's active requirements.
Real-time job harvesting
Continuously pulled live job postings from 15+ sources — Indeed, LinkedIn, Dice, Monster, Broadbean — matched to employer requirements open in Oorwin. Refreshed every 4 hours.
AI-powered resume harvesting
Sourced candidate resumes from integrated job boards and resume databases, ranked by AI match score against open job requirements. Reduced cold-start sourcing time from 18 min to under 3 min.
Personalised AI ranking
Ranking model trained on recruiter behaviour — clicks, submits, ignores — to progressively sharpen relevance per recruiter over time. Not just keyword match: semantic understanding of role fit.
One-click import to ATS
Harvested jobs and resumes imported directly into Oorwin with a single click — candidate record created, job tagged, no copy-paste. Eliminated the entire manual data entry step.
Employer-level personalisation
Employers could define sourcing preferences — preferred job boards, geographic filters, seniority bands — giving them control without requiring daily configuration effort.
Smart alerts & digest
Daily email digest of top-matched harvested jobs and resumes per recruiter. Kept the feature top-of-mind even on days recruiters didn't log in — driving re-engagement.
How a single feature became the platform's growth engine
Reaching 200+ customers wasn't an accident — it required a deliberate scaling strategy that spanned product iteration, go-to-market motion, and customer success alignment. Here's how the journey unfolded across three phases.
0–10 customers
10–50 customers
50–200+ customers
What made it scale — the four levers
Platform's best-selling feature — and a new standard for recruiter productivity
Within one product cycle at Oorwin, Job/Resume Harvester went from zero to the single most commercially successful feature on the platform — cited in sales, customer success, and renewal conversations alike.
The success of the Harvester directly informed the product strategy at Workllama — where I subsequently applied the same AI-first, embedded-distribution approach to the Communication Hub and Best Match AI features.
On measurement: The 200+ customer figure reflects accounts with at least one active Harvester session per week within 6 months of GA — not just accounts with access. Submission rate improvement (+75%) was measured by comparing recruiter submission-to-open-role ratios 3 months pre and post feature launch across a cohort of 40+ active users. Feature NPS (71) was collected via an in-app survey prompt shown to users after their 5th Harvester session. The 18→3 minute figure is based on recruiter self-reported time-on-task from shadowing sessions, not system-logged data.
Lessons from scaling a feature to 200+ customers
The business case is the product spec for leadership. Engineering buy-in came from the ROI model — not the feature vision. Framing 90 minutes saved per recruiter as a 3–5× subscription ROI turned a feature request into a strategic initiative. I've used this framing for every 0→1 pitch since.
Distribution is a product decision, not a marketing problem. Bundling versus add-on wasn't a pricing call — it was a distribution call. Bundling gave us access to 100% of the customer base as an activation surface. That single decision compressed the 0→200 customer journey from years to months.
Sales demos are a product research tool. Running live demos with prospects exposed objections and confusion patterns I never would have seen in lab research. I used demo feedback to ship 3 UX improvements in the first 30 days post-GA. Treat sales as a continuous feedback loop, not a handoff.
AI moats are built through data flywheels, not model sophistication. The Harvester's ranking model wasn't state-of-the-art ML — it was a well-designed feedback loop. The moat came from recruiter behaviour data accumulating over time. Shipping fast and learning from real usage beat waiting for a perfect model.
Customer Success is a scaling team, not a support team. The Harvester health scorecard I built with CS turned reactive account management into proactive feature adoption. Accounts that received CS-driven activation outreach had 2.4× higher weekly active usage at 90 days than those who didn't.
Want to go deeper on this?
Happy to walk through the AI architecture decisions, the scaling playbook, or the metrics in detail.