Giving recruiters their time back at a Sydney tech-recruitment agency, through three AI-assisted workflows that keep a recruiter in control and every candidate treated fairly.
Claude ProjectsFairness controlsCandidate privacyAdoption-led rollout
A demonstration built around a representative client scenario · figures are modelled estimates under stated assumptions.
The problem
Plateaued by admin, and burned by automation
Recruiters spend more time on drafting, notes and Bullhorn than on people.
Job ads, interview notes and client updates are rebuilt from scratch, inconsistently.
A previous automation attempt failed, tools dropped in without redesign or support.
The legacy: distrust, and private workarounds. Adoption is the real challenge.
So the goal isn't more tools, it's a recruiter-controlled redesign people trust, that's fair and private by design.
The approach
Three workflows, recruiter-in-control
Job-ad drafting Consistent, inclusive ads with a built-in inclusive-language check
Interview summarisation Structured, evidence-only summaries, no bias, no decisioning
Hiring-manager updates Professional client updates, confidential by default
The principle: AI drafts, a recruiter decides, and it never screens, ranks or selects candidates. The AI removes typing, not judgement.
Pilot 1 · Job-ad drafting
Current → future
Today, ~45 min/ad, inconsistent, inclusivity ad hoc:
Intake
Draft from scratch
Edit for tone
Inclusivity?
ad hoc
Post
With the system, ~18 min, 60% faster, inclusive by default:
Structured intake
Job-Ad Project
firm-standard draft
Inclusive-language check
Recruiter posts
Pilot 2 · Interview summarisation, AI in action
Evidence-only, with a fairness guard
Recruiter's rough notes (input)
"Strong on the React role, walked through a big migration they led, good on state management. Asked smart Qs about the team. Bit older, wasn't sure they'd fit our young team culture. Notice period 4 weeks."
STRUCTURED SUMMARY · fairness guard applied
Evidence vs requirements:
• React / front-end, led a large
migration; strong on state mgmt ✓
• Communication, asked informed
questions about the team ✓
Practical: notice period 4 weeks.
⚑ Fairness: an age-based "culture fit"
remark in the notes was NOT included,
not job-relevant. Summary is evidence
-only. No progression decision made.
The biased aside is dropped, not summarised. The tool sticks to job-relevant evidence, flags what it excluded and why, and leaves the decision to the recruiter, protecting the candidate and the agency at once.
Scenario results
All three clear the brief's targets
60%
Job ads ↓
60%
Interview write-ups ↓
60%
Client updates ↓
~1,900
Hours reclaimed / yr
Brief targets were ≥40% admin-time and 50% candidate-summary-prep, both cleared. ≈1.7 hrs/recruiter/week, ~6× the fee on the modelled assumptions. Percentages depend only on per-task time, so they're the robust part.
Scenario estimates under the stated assumptions.
Governance & what's delivered
Fair, private, recruiter-controlled, and reusable
The line we don't cross: AI never screens, ranks or decides on candidates; summaries are evidence-only with no protected attributes; candidate data stays in approved tools; a recruiter reviews and sends everything.