Background Check Revamp

What Was Broken

A candidate applies, gets an offer, and then waits four months for a background check. Four months. In that time, they’ve probably accepted another offer, or lost interest, or both. The process was a chain of manual handoffs between vendors, Legal, and internal teams, with no single person able to see the full pipeline.

The 90th percentile was the real problem — most checks finished in 6–8 weeks, but the long tail of blocked cases dragged the average to four months.

The First Cut: Seeing the Board

Before we could fix the process, we had to measure it. SQL queries against the vendor databases, stitched together with VBA macros because nobody had built a real dashboard. The output went into QuickSight — color-coded by stage, flagged by delay, grouped by vendor.

What showed up was obvious once you could see it: three specific vendors were responsible for 60% of the delays. One of them had a broken API endpoint that nobody had reported because the fallback was manual email.

Rewriting the Rules

The fix wasn’t technical — it was procedural. We renegotiated SLAs with the two worst vendors, added an auto-escalation trigger at 30 days, and built a fallback path with a secondary vendor for the most common check types. The third vendor we just stopped using.

The verification workflows got re-engineered too. Instead of serial checks (A finishes, then B starts, then C), we parallelized everything that didn’t depend on the other. SQL tracked dependencies, QuickSight showed bottlenecks in real time, and the team stopped spending half their day chasing status updates by email.

Scaling the Team

Ten people couldn’t handle the volume at peak season, and four of them knew the whole process end-to-end. So we grew to 45 — not by hiring generalists, but by building specializations. One track for vendor management, one for exception handling, one for data quality. I mentored six of them into leads who now run their own teams.

The hardest part wasn’t the scaling — it was getting Legal and Compliance to agree that a 30-day SLA was actually safe. That took sitting in meetings, showing them the data, and earning trust that the new rules wouldn’t create risk.

The Numbers

90th percentile turnaround dropped from four months to one. Manual review time cut in half. The program touches 120K hires a year across India.

The takeaway: most process problems aren’t engineering problems. You have to sit inside the process long enough to see where it actually breaks.