925 Techies
← All case studies Staffing & Recruiting

From four months to four weeks: AI in high-volume hiring

How staffing firms and talent teams use AI to screen, match and schedule at volume, while recruiters spend their time on candidates and clients.

4 mo → 4 wkHiring cycle reported by Unilever
50,000 hrsCandidate time saved over 18 months
£1MReported annual cost savings

Figures: Unilever, 2016–2019. Source and caveats below.

01 The problem

A recruiter at a staffing firm can spend most of the week on work that never reaches a candidate conversation: reading resumes, re-keying data into the ATS, chasing availability and writing submittals. When a requisition draws several hundred applicants, good people are missed simply because nobody got to their resume.

Speed decides who wins the placement. The firm that puts three qualified candidates in front of the client first usually fills the role, and every day a billable seat stays open is lost margin for the agency and lost output for the client.

02 Where AI fits

A

Resume parsing and skills matching

Models read resumes and job descriptions for meaning, not keywords, and rank candidates against the requirements with a written rationale for each match.

B

Rediscovery of your own database

Semantic search across the ATS surfaces past applicants and silver medalists who fit a new requisition, before you pay for new sourcing.

C

Screening and scheduling agents

A conversational agent handles first-touch screening questions, work authorization checks and interview scheduling by text or email, around the clock.

D

Submittal and outreach drafting

Client-ready candidate summaries and personalized outreach generated from the resume and the recruiter's notes, for the recruiter to edit and send.

E

Demand forecasting

Models that predict which clients and skill sets will open requisitions, so benches are built ahead of demand.

03 Reference architecture

The layers we put in place for a typical staffing & recruiting deployment, from your systems of record up to the people who review the output.

  1. L1ATS / CRM (Bullhorn, Greenhouse, Workday)
  2. L2Resume and job parsing pipeline
  3. L3Embedding index of candidates and roles
  4. L4Matching and screening agents
  5. L5Recruiter review queue
  6. L6Bias audit and reporting layer

04 Real-world implementation

Public deploymentUnilever2016–2019

Unilever rebuilt its graduate hiring around online assessments and AI-analyzed video interviews for a pool of roughly 250,000 applicants a year. The company and its vendor HireVue reported the hiring cycle falling from about four months to as little as four weeks, more than 50,000 hours of candidate time saved over 18 months, over £1 million in annual savings, a 96% candidate completion rate and a 16% increase in the diversity of hires.

4 mo → 4 wkHiring cycle reported by Unilever
50,000 hrsCandidate time saved over 18 months
£1MReported annual cost savings

Source: Unilever / HireVue customer story; reported by Business Insider (2017). Unilever is not a 925 Techies client.

What to keep in mind: These figures are reported by the company and its vendor and have not been independently audited. Automated hiring tools are also regulated: New York City Local Law 144 requires annual bias audits, and EEOC guidance applies Title VII to algorithmic selection.

05 Business impact

Time to submit

Shortlists in hours rather than days mean you are first in front of the client more often, which is the main driver of fill rate.

Recruiter capacity

Removing manual screening and scheduling lets each recruiter carry more open requisitions without lowering candidate contact quality.

Sourcing spend

Filling from your existing database reduces job board and sourcing licence costs per placement.

Candidate experience

Every applicant gets a response and a next step, which protects your brand in a tight talent market.

KPIs we baseline and track
Time to submitSubmit-to-interview ratioFill rateRequisitions per recruiterAdverse-impact ratio

06 Implementation roadmap

  1. Weeks 1–3

    Assess

    Audit ATS data quality, map the recruiter workflow, pick one desk or job family as the pilot.

  2. Weeks 4–8

    Prove

    Matching POC scored against historical placements: would the model have ranked the people you actually placed?

  3. Weeks 9–18

    Build

    Integrate ranking, rediscovery and scheduling into the ATS. Add bias testing and recruiter feedback capture.

  4. Ongoing

    Operate

    Monitor match quality and adverse-impact ratios, retrain on new placements, extend to more desks.

07 Guardrails

  • Humans make every reject and every hire decision
  • Adverse-impact testing by gender, race and age before and after launch
  • Protected attributes and proxies removed from model inputs
  • Candidate notice and opt-out where local law requires it

08 Services involved

Start here

Exploring AI in staffing & recruiting?

We can run a proof of concept on your own data in about four weeks, measured against the KPIs above.