Case studies

Workflows we've taken off operators' plates.

Three engagements across real estate, recruiting, and HR — each shipped in eight weeks, each still in production. Numbers from week-12 reviews with each client.

Illustrative engagement profiles · pending sign-off before public release
Case 01 · Real estate

Brokerage prospecting agent for a 30-agent residential firm

[Client A] is a residential brokerage operating across Sydney and the Central Coast. Their top agents were spending one full day each week sourcing and contacting lapsed leads — the same work, week after week, with no consistent follow-through.

The brief

Reactivate lapsed buyer and seller leads at the scale of the whole brokerage — without piling more admin onto agents who already spend 10+ hours a week on prospecting that produces inconsistent follow-up.

What we built

An agent that pulls fresh REA data, ranks lapsed contacts by predicted intent, drafts a tailored outreach message per agent's voice, and queues sends through the brokerage's email and SMS providers. Every send is traced; agents see the agent's reasoning before approving.

How it shipped

  • One Zeroth engineer paired daily with the firm's top broker for 8 weeks
  • Eval suite written before tool code — coverage of message-tone, opt-out handling, REA data drift
  • Soft rollout to 5 agents in week 7, full firm in week 8
9.5h/wk
saved per agent on prospecting & outreach
measured week 12 vs baseline
+22%
lift in qualified inquiries in first 90 days
vs prior-quarter benchmark
~5mo
payback on the build fee
based on commission attribution
285h/wk
total operator hours redirected, firm-wide
9.5h × 30 agents
Stack Claude Sonnet 4.6 Zeroth platform · Modal REA scraper (Apify) Resend · Twilio SMS Supabase
We'd been told an agent like this would take six months and a team of four. Zeroth shipped it in eight weeks with one engineer paired with our top broker. By week 12 it had paid for itself in commission.
[Director of Sales] [Client A] · Sydney
Case 02 · Recruiting

Sourcing & reference-call agent for a boutique tech search firm

[Client B] is a 15-consultant exec search firm focused on senior engineering and product roles across APAC. Principals were burning Fridays writing up reference calls and Mondays cleaning up sourcing lists they'd built by hand.

The brief

Cut the cycle time from JD to first-pass shortlist, and stop principals from losing entire days to reference-call write-ups. Both tasks needed to remain consultant-quality — a generic AI résumé screener wouldn't cut it for retained search.

What we built

Two coupled agents. A sourcing agent reads a JD like a recruiter, expands Boolean queries, scrapes and ranks candidates, then writes a shortlist note per candidate. A reference agent transcribes calls, structures themes, and drafts a summary the principal edits — never sends.

How it shipped

  • Two agents, one paired engineer, 8 weeks — sourcing in production week 5, reference write-up week 8
  • Per-consultant prompt tuning: each principal's "voice" stored as a directive overlay
  • Evals over 60 historical placements before go-live
14h → 90min
JD to first-pass shortlist time
average over week-12 sample
3.2x
reference calls completed per consultant per month
write-up no longer the bottleneck
~4mo
payback on the build fee
based on extra completed searches
0
candidate emails sent without principal approval
agent drafts; consultant always sends
Stack Claude Opus 4.7 Zeroth platform · Modal LinkedIn scraper Whisper transcription Internal CRM (REST)
I used to spend Fridays writing up reference calls. Now I spend Fridays talking to candidates. The agent isn't doing the job — it's letting me do mine.
[Managing Partner] [Client B] · Melbourne
Case 03 · HR & People Ops

Policy QA & onboarding orchestration for a 120-client PEO

[Client C] is a Professional Employer Organisation serving ~120 Australian SMB clients. A six-person People Ops team was drowning in tier-1 policy questions while onboarding queues stretched to 11 days. Adding headcount wasn't the answer; the work was repetitive, not hard.

The brief

Take repetitive policy queries off a strained People Ops team and shorten an onboarding pipeline that had grown to 11 working days — without compromising the audit trail PEO clients rely on for compliance.

What we built

A policy-aware HR copilot in Slack that answers tier-1 questions with citations into the client's handbook, plus an onboarding orchestration agent that coordinates HRIS provisioning, contract dispatch, payroll setup, and day-1 checklists with human approval gates.

How it shipped

  • 8-week build for the policy QA copilot, 4-week extension for onboarding orchestration
  • Per-client policy corpora indexed with strict tenant isolation in Supabase
  • Every agent answer renders a citation panel showing the policy clause used
11 → 4days
average onboarding turnaround
trailing 90-day cohort vs baseline
61%
tier-1 HR queries auto-resolved with citation
up from 0% baseline
1.8FTE
capacity redirected from tier-1 to advisory work
measured week 16
100%
of agent answers carry a clause-level citation
policy QA design constraint
Stack Claude Sonnet 4.6 Zeroth platform · Modal Supabase pgvector Slack bot · HRIS APIs Per-tenant policy corpora
The agent doesn't replace the team — it lets them do the work they actually trained for. Our consultants stopped answering "where's the leave form" and went back to running employee relations cases.
[Head of People Operations] [Client C] · Brisbane

Got a workflow that looks like one of these?

30-minute call. We'll tell you whether agents help — and roughly what it would take.

Book a strategy call