You're Paid to Find the Right Person. Stop Drowning in the Wrong Ones.

Picture a typical Tuesday at a contingency recruiting firm carrying 20 open reqs. The morning starts with 40 new applications in the ATS from an Indeed posting that went live yesterday. Before a single recruiter makes a placement call, someone has to open those applications, eyeball the resumes, decide who clears the bar, and move them to the next stage. At most firms, that someone is the recruiter.

That's 45 minutes to an hour, minimum — and that's a light day. On a high-volume req like an inside sales role or a staff accountant search, it's longer. Now multiply that across three or four active reqs, add the LinkedIn InMail responses that came in overnight, the candidate who needs an interview confirmation resent, and the client who wants a status update by noon. It's 11 AM and no one has made a sourcing call yet.

The tools in play here are usually a mix of LinkedIn Recruiter, Indeed, an ATS (Bullhorn, Jobvite, Greenhouse — pick your poison), and a shared inbox or spreadsheet someone built in 2019 that everyone is afraid to touch. None of these systems talk to each other in any useful way. Data entry is manual. Follow-up is manual. Candidate status updates are manual.

Here's where AI actually intervenes: an automated screening layer sits between your job board inflows and your ATS. Every application gets read against a defined scorecard — not a keyword match, but a structured evaluation of experience relevance, tenure patterns, and role-specific must-haves. The output isn't a raw pile. It's a ranked list with a brief fit summary on each candidate, written back to the ATS record before the recruiter opens their laptop.

The recruiter still makes the call on who to advance. That judgment doesn't move. What moves is when they apply it — now it's the first thing they do, not the thing they finally get to after an hour of triage. (Source: LinkedIn, 2023 — the LinkedIn Future of Recruiting report found that sourcing and screening consume the largest share of recruiter time, ahead of client management and offer negotiation.)

The pitch you'll hear most often sounds like this: "Our platform uses AI to find you better candidates faster." It's vague by design, because the product underneath it is usually a résumé parsing upgrade dressed up in machine learning language. That's worth being skeptical of.

Here are the specific red flags worth watching for when a vendor sits down across from you:

• "We integrate with your ATS" — ask exactly what that means. Does it write data back to the ATS automatically, or does it require a human export-import step? A lot of "integrations" are glorified CSV imports.
• Proprietary candidate databases bundled with the automation — some vendors sell you access to their candidate pool as part of the deal. That's a data subscription product, not an automation product. They're solving a different problem than your workflow, and the two should be priced separately.
• AI scoring models you can't inspect — if a vendor can't explain what criteria their model is ranking candidates on, you don't own the logic. You're renting a black box. When a client asks why a candidate was ranked third, you need an answer that isn't "the algorithm said so."
• Compliance hand-waving — EEOC guidance on AI-assisted hiring tools is an active and evolving area. Vendors who dismiss this with "we're compliant" and no documentation should be walked out. The liability lands on the firm making the placement decision, not the software vendor.
• Overpromised implementation timelines — if a vendor says you'll be fully live in 48 hours, what they mean is their SaaS login will work in 48 hours. The actual configuration of your screening criteria, your ATS field mapping, your outreach sequences — that takes weeks of real work. Any firm that skips that work ships you a generic tool, not a configured system.

The vendors worth talking to are the ones who ask about your actual reqs before they demo anything. If the demo is the same for every recruiting firm, the product is built for the average firm — not yours.

Every recruiting firm with more than two years of history has the same expensive problem: a candidate database full of qualified people they've already screened, interviewed, and placed — who are now invisible. A new req comes in, the recruiter posts to Indeed, and a candidate they placed three years ago at a company they've since left sits uncontacted in Bullhorn.

Database reactivation automation is, in my opinion, the single highest-ROI implementation for most recruiting firms — and the most consistently overlooked one. Here's how it actually works.

The system connects to your ATS via API (Bullhorn, Greenhouse, Lever, and most major platforms have documented APIs — this is a solved technical problem). When a new requisition is created, the AI reads the req description and translates it into a structured search query: required skills, years of experience ranges, industry background, geography, and any hard exclusions your team defines. It runs that query against your full candidate database — not just active candidates, but your entire history.

The output is a shortlist of existing candidates ranked by fit, each with a brief summary of why they matched and when they were last contacted. A recruiter opening a new req on Monday morning doesn't start at zero. They start with five to ten warm names from their own database, before they've written a single job posting.

What changes on day one: recruiters stop saying "let me check if we have anyone" and start saying "here's who we already have." What changes by month three: the team starts to trust the database again. Candidates get re-engaged before they've had time to sign with a competitor. Reqs that used to sit open for three weeks because sourcing started cold now have initial outreach going out within hours of the req being created.

According to the Society for Human Resource Management, the average time-to-fill for professional roles in the U.S. has been climbing, with many firms reporting significant delays when sourcing must start from scratch. (Source: Society for Human Resource Management, 2023) Working your existing database before going external isn't just faster — it's cheaper per placement, and the candidates already know your firm.

The system Oaken builds for this uses pgvector for semantic similarity search across candidate records, Claude for generating fit summaries, and FastAPI to push the ranked list directly into the ATS as a saved search result. No separate dashboard. No new login. It shows up where recruiters already work.

The most common mistake recruiting firms make when they try to implement AI is starting with the wrong problem. They've heard about AI writing job descriptions, so they automate job description writing — which was never the bottleneck. Job descriptions take 20 minutes. Screening 60 resumes takes three hours. Starting with the flashy use case instead of the friction point is how you spend money and feel nothing.

The second failure mode is scope creep before the first thing works. A firm decides to automate screening, outreach, scheduling, reference checks, and ATS documentation all at once. Six weeks later, nothing is fully live because every system depends on decisions about the next one. The right approach is to pick one workflow, run it to completion, and let the team feel the difference before adding anything.

• Skipping the criteria conversation — AI screening is only as good as the scorecard behind it. Firms that skip the work of defining what "good" looks like for each req type end up with a ranking system built on guesses. The first two weeks of any implementation should involve sitting down with your best recruiters and making their implicit criteria explicit.
• Not involving the recruiters who will use it — implementations that are handed down from leadership without recruiter input get quietly ignored. Recruiters will keep doing it the old way if the new way doesn't feel like it was built for their actual workflow. The best implementations are built with the team, not for them.
• Treating the ATS as a barrier instead of the destination — the value of any automation is zero if it doesn't write back to where recruiters actually work. Firms that accept a side dashboard or a separate login are accepting a tool their team will stop using in 60 days.

There's also a data quality problem that surprises almost every firm. The candidate database that's going to power your reactivation automation has duplicate records, missing fields, and contact information that's three jobs stale. You can't automate your way around bad data. Some firms need two or three weeks of database cleanup before the automation can do anything useful — and that work is worth doing regardless of whether you automate, because your recruiters have been working around it manually for years.

The firms that get this right share one trait: they treat the first implementation as an experiment with a specific success metric, not a transformation of everything at once. (Source: Staffing Industry Analysts, 2023 — SIA research on technology adoption in staffing firms consistently finds that phased, use-case-specific implementations outperform broad platform rollouts on both adoption rate and measurable outcome improvement.)