AI for recruiting and staffing
AI for recruiting and staffing firms
First qualified submit wins the req. Sourcing, screening, and scheduling at machine speed. Judgment stays human. 5x ROI in 30 days, or we work free.
- Agency recruiting
- Temp staffing
- Internal TA
- Executive search
- Staff augmentation
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Speed-to-submit decides everything.
The first qualified candidate usually wins the req. Employers averaged $5,475 in cost per non-executive hire in 2025 (SHRM, 2025, 2,371 US organisations), which is what a client spends to fill the seat in-house. Sourcing by hand burns the days that decide whether they call you instead.
Recruiters spend the day scheduling.
Back-and-forth emails to find one time slot, multiplied by every candidate, every round.
Your bench goes cold.
Placeable candidates from last quarter sit forgotten in the ATS while you source strangers.
You pay to find the same candidate twice.
The developer you placed in 2023 sits in the ATS behind a dead email and a duplicate profile. Next month a sourcer buys them back from a job board at full price, as a stranger.
Silver medalists retire inside closed reqs.
The runner-up on the last search is the first call for this one. Nobody reopens a filled req to look, so the second-best person you already screened gets placed by a competitor.
Assignment end dates pass in silence.
A contractor rolls off Friday and signs with whoever calls first. The date sat in the ATS all along. Nobody ran the report, and redeployment is the margin you did not make.
A 360 recruiter's Thursday, req to submittal
She takes the req on the client call.
humanTen forty, the client on the line. She writes the intake as they talk: must-haves, the salary band, the two dealbreakers, the start date the hiring manager will not say out loud but means. The req goes in structured, because everything downstream searches against what she writes here.
The bench is searched before any board.
The match runs over the agency's own ATS first: past submittals, silver medalists from filled reqs, placements whose contracts are ending. Back comes a ranked shortlist with reasons attached, each hit citing its record, the 2024 submittal, the screen notes, the placement that ended in March. The boards are queued only for what the bench cannot answer.
She works the phones.
humanThe top of the list gets called, not messaged. Two are open to a move, one says ask me in the spring, and that answer is written back to the record so the spring req finds it. Fit, motivation and money are her read. Nothing on this desk decides a person is out.
Outreach drafts wait in the queue.
For the longer tail, messages draft themselves the way the anchor build did it, researched and specific to the person, not the persona. Consent state is checked in the data layer before a draft even exists. Nothing has been sent.
She approves every send.
humanShe reads the drafts the way she would read a junior's, cuts one that reads generic, tightens two, approves the rest. A reply pulls the thread straight back to her. This gate is the difference between an engine and spam.
Slots go out, interviews come back.
For the two live candidates, the system offers times from the client's and candidates' calendars, confirms, and rebooks the one who slips. Nobody on the desk composes a scheduling email today.
The submittal pack assembles.
Formatted CV in the client's template, screen notes against the req's criteria, right-to-work status, rate. It waits in the queue under her name, sourced field by field from the record.
Her name goes on the submit.
humanShe reviews the pack and sends it. First qualified submit on the client's desk while the other agency is still sourcing strangers. The call notes and the outcome file back to the ATS, so the next req in this stack starts warmer than this one did.
Recruiting & staffing, before and after
The manual path is dashed: Sourcing burns days per req, Scheduling ping-pong, Bench data goes stale. The system path replaces it, and a person approves before anything ships: Shortlists in hours, Interviews book themselves, Past candidates resurface.
Before: by hand
- 01Sourcing burns days per reqhuman
- 02Scheduling ping-ponghuman
- 03Bench data goes stalehuman
After: the system
- 01Shortlists in hours
- 02Interviews book themselves
- 03Your approvalhuman
- 04Past candidates resurface
The research
Mastercard cut interview scheduling time by more than 85% and scheduled 88% of interviews within 24 hours (Phenom, 2025, vendor case study).
Phenom · 2025
Proof from the pattern
The Breez build is the recruiting mechanic under a different label: thousands of prospects researched, qualified and approached with specifics. Pointed at candidates instead of customers, it is the same system.
We guarantee 5x ROI inside 30 days of deployment, in writing, measured against a baseline you sign before we build. If the system misses the bar, we keep working for free until it clears.
Who this is built for
Your database outruns the job boards
Fifteen years of placed candidates, submittals and interview notes sit in the ATS, and you paid to source every one of them once already. Every competitor searches the same LinkedIn. Nobody else can search your database. Indexed, it answers a new req before the boards even load, and the moat you always claimed to have starts acting like one.
The intake call ends with a call list
By the time the req is written up, the match has already run over past submittals, silver medalists and placements rolling off contract. You start the morning on the phone with the five closest people you already know instead of page four of a boolean search. The part of the job that wins fees is the part you now do all day.
Monday's orders fill from Friday's bench
The client calls Thursday needing six people on site Monday. The redeployment list is already drafted from assignment end dates, right-to-work checks current, and you approve the outreach before it goes. The order fills from workers you have already placed and paid, which is the fill your margin actually survives on.
Candidates hear back the same day
Loops get booked without a scheduling thread, screens are scored against the req's stated criteria, and the debrief lands with the scorecards attached. Nobody is ghosted by accident, which your employer brand notices before your dashboard does. Every hire decision stays with a person, and the trail shows it.
What stays human
- She takes the req on the client call
- She works the phones
- She approves every send
- Her name goes on the submit
The steps the day below leaves to a person, by design.
The numbers in recruiting and staffing firms
Mastercard cut interview scheduling time by more than 85% and scheduled 88% of interviews within 24 hours (Phenom, 2025, vendor case study).
Phenom · 2025
In 2025, 41% of talent-acquisition teams piloted AI interview scheduling and 23% standardized it (HeroHunt, 2025).
HeroHunt · 2025
What we build for recruiting and staffing firms
Fill from the bench
Candidate sourcing and shortlisting.
Profiles researched and matched against the req at volume. The recruiter reviews a shortlist instead of a haystack.
Bench re-engagement.
Past candidates resurfaced automatically the moment a matching req lands.
Redeployment before roll-off.
Assignment end dates watched, the call list drafted weeks out, the recruiter makes the calls. Contractors redeploy with you instead of signing with whoever phoned first.
Reach out and book
Personalized candidate outreach.
The outbound engine pointed at candidates: researched, specific to the person, CASL-compliant.
Interview scheduling automation.
Slots offered, confirmed and rescheduled without a recruiter playing calendar tennis.
Screen and submit
Screening and structured scoring support.
Consistent first-pass screens against the req's actual criteria. A recruiter makes every call.
Client submittal packs.
Formatted, complete and fast, so your candidate is the first qualified one on the desk.
How an answer is found in your own bench
A question runs against your own bench two ways at once: vector search for meaning and full-text search for exact wording. Reciprocal rank fusion merges both result sets, and the answer carries the source it came from.
- 01A question
- 02Vector search (meaning)
- 03Full-text search (exact wording)
- 04Rank fusion
- 05Answer, with its source
The economics
Before and after economics
Line
- Research per candidate
Before
Hours per properly researched person, done by hand, which caps the desk at whoever has hours left
After
Minutes per researched prospect in the anchor build, the same research mechanic this page points at candidates
- Interview scheduling
Before
Rounds of email to land one slot, per candidate, per round
After
Mastercard cut scheduling time by more than 85% and booked 88% of interviews inside 24 hours (Phenom, 2025, vendor case study)
- Manual desk load, modelled
Before
72 hours a week across the default desk: 6 recruiters at 12 manual hours each, $4,680 a week at $65 loaded cost
After
The calculator below prices your own desk, and the 5x ROI guarantee is measured against a baseline you sign before we build
- Submittal and proposal depth
Before
Packs assembled by hand at the end of the day, after the calls
After
The anchor build carried 40+ personalized data points per proposal, assembled in minutes, and the pack is the same move in a client template
Three sources only. The breez case frontmatter (minutes per researched prospect instead of hours, 40+ personalized data points per proposal), which is adjacent proof, the same research and outreach machine pointed at customers rather than candidates, and the rows say so where they cite it. The Phenom Mastercard figure already cited on this page's stat rail, a vendor case study and labelled as one. And this page's calculator defaults, 6 recruiters at 12 manual hours a week each at $65 loaded hourly cost, which is 72 hours or $4,680 a week of manual load. No row is a staffing client result, and the page's proof note says the same thing in the proof section.
Where the data comes from
ATS
Bullhorn, Loxo, Vincere or JobAdder on agency desks, Avionté or TempWorks where the desk runs temp pay and bill, Greenhouse or Lever inside internal TA teams. It holds the asset this whole build runs on: placed candidates, submittal history, screen notes, silver medalists and every assignment end date. The index points into it so the first search on any req runs over people you already know.
Where it stops. Candidate records stay in the ATS under your own permissions. The index references them, consent state is enforced in the data layer, and no automated decision ever writes a rejection back. Nothing about a person leaves your accounts, and nothing trains anyone's model.
CRM
The client side of the house, whether that is Bullhorn's CRM half or HubSpot: job orders, contacts, rate cards, MSAs and the win-loss history on past reqs. Intake becomes a structured req here, the submittal pack learns the client's format here, and the redeployment call list knows which clients buy contract talent again.
Where it stops. Client data is walled from candidate outreach copy, CASL consent state is checked before any message drafts, and nothing reaches a client or a candidate without a recruiter's approval. Fee terms and rate cards never appear in candidate-facing text.
Job boards
Indeed, ZipRecruiter and the LinkedIn seats you already pay for. On this build the boards are the supplement, searched after the bench answers, topping up only what the database cannot fill. Postings draft from the structured req, and board responses flow into the same ranked shortlist with the same reasons attached.
Where it stops. Searches and postings run inside each board's terms and your seat licences. Nothing auto-applies, nothing auto-posts without a person, and board data does not migrate into the corpus beyond what the licence allows.
How the system is built for recruiting and staffing firms
See the full capability mapRetrieval
Your own bench, past placements and closed reqs indexed together, so matching runs against people you actually know rather than a keyword search over a job board. The match cites the evidence in the record.
- pgvector
- Full-text BM25
- Reciprocal rank fusion
Agents and orchestration
Agents research, shortlist and draft outreach. A recruiter approves before anything sends. Screening output is a ranked shortlist with reasons attached, never a filter that silently drops people.
- agent-worker
- Proposal queue
- Ranked shortlists
Evaluation
Matching is graded against reqs you have already filled, where the placement is known. We also test for disparate outcomes across groups, because a matching system that has not been checked for bias is a legal problem.
- Eval graders
- Adverse-impact testing
- quality-worker
Models
Embedding models for candidate and role similarity. Frontier models for outreach and screening notes. No automated reject decisions, and no scoring model that cannot explain the factors behind a ranking.
- Embedding models via the same gateway
- Frontier models, one gateway, routed per task
- Explainable ranking
Data boundary
Candidate personal information stays in your accounts under Canadian privacy law, scoped per team, with consent state enforced in the data layer.
- Supabase row-level security
- Consent ledger
- No-training API terms
ATS, CRM, Job boards feed a hybrid index. The agent runtime works from that index, and every consequential action passes a human approval before it reaches Workflow builder, Task board, Ask your brain.
Your systems
- 01ATS
- 02CRM
- 03Job boards
The system
- 01Hybrid index
- 02Agent runtime
- 03Your approvalhuman
Where your team works
- 01Workflow builder
- 02Task board
- 03Ask your brain
What that means in practice
Agents and orchestration
AI that does the work instead of just answering: looks things up, calls your systems, completes multi-step tasks, and knows when to hand off to a human.
Where we stop. Multi-agent swarms are oversold; most jobs need one well-guarded loop. If a cron job and a script solve it, that is what we build, because 90 percent per-step accuracy compounds to 59 percent over five chained steps and no framework changes that arithmetic.
Retrieval and RAG
Connecting AI to your actual documents so it answers from your knowledge, accurately and with citations, instead of making things up.
Where we stop. A dedicated vector database is justified by scale, not by default. Most of RAG quality is won or lost in chunking and indexing strategy, not in the model choice, and we have walked clients back from RAG to plain search when that was the honest answer.
Evaluation and observability
How we prove the AI actually works: measured, monitored and regression-tested like real software, not vibes.
Where we stop. There is no engagement where we skip this. The honest variable is depth: a document pipeline gets faithfulness and extraction suites, an outbound agent gets human review sampling, a classifier gets a held-out test set. We size the harness to the risk, never to zero.
Built around your rules
| Regime | What it demands here | How the system complies |
|---|---|---|
| PIPEDA and BC PIPA | A resume is personal information. Consent, purpose limits and access and correction rights attach to every candidate record, and a candidate can ask what you hold and how it was used. | Candidate records stay in your ATS and your accounts, consent state lives in the data layer, and the audit trail can answer an access request with the actual retrieval and ranking history rather than a shrug. |
| Human rights law (BC Human Rights Code and equivalents) | Screening must not discriminate on protected grounds, directly or through proxies like a name, a postal code or a resume gap. | Criteria are structured from the req and auditable, adverse-impact testing runs before the first live shortlist, and a recruiter makes every decision about a person. Nothing is silently filtered out. |
| CASL | Commercial electronic messages need consent, sender identification and a working unsubscribe. | Consent state is enforced in the data layer, so an agent cannot draft to a person who opted out, and every message says who sent it and why it is relevant. |
| ESA agency licensing (Ontario and BC) | Temp agencies and recruiters operate licensed in Ontario, employment agencies hold a licence in BC, and charging a worker a fee for placement is prohibited in both. | The build changes nothing about the licence posture and touches no candidate money. Placement and submittal records accumulate in the ATS, where a licensing audit would expect to find them. |
| Ontario ESA job-posting AI disclosure | Since January 2026, publicly advertised Ontario postings from larger employers must disclose the use of AI to screen, assess or select applicants. | The system's role is narrow enough to disclose in one sentence, ranking against stated req criteria with a recruiter deciding, and we help write that sentence so it is true as printed. |
| Client-side automated hiring rules (NYC Local Law 144 and similar) | Clients hiring into jurisdictions with automated employment decision rules inherit bias-audit and notice duties for the tools in their process. | Decisions stay with recruiters by design, and the adverse-impact test results and the audit trail exist for the client whose counsel asks. |
The regimes that govern recruiting & staffing, what each demands, and how the system complies
Regime
- PIPEDA and BC PIPA
What it demands here
A resume is personal information. Consent, purpose limits and access and correction rights attach to every candidate record, and a candidate can ask what you hold and how it was used.
How the system complies
Candidate records stay in your ATS and your accounts, consent state lives in the data layer, and the audit trail can answer an access request with the actual retrieval and ranking history rather than a shrug.
- Human rights law (BC Human Rights Code and equivalents)
What it demands here
Screening must not discriminate on protected grounds, directly or through proxies like a name, a postal code or a resume gap.
How the system complies
Criteria are structured from the req and auditable, adverse-impact testing runs before the first live shortlist, and a recruiter makes every decision about a person. Nothing is silently filtered out.
- CASL
What it demands here
Commercial electronic messages need consent, sender identification and a working unsubscribe.
How the system complies
Consent state is enforced in the data layer, so an agent cannot draft to a person who opted out, and every message says who sent it and why it is relevant.
- ESA agency licensing (Ontario and BC)
What it demands here
Temp agencies and recruiters operate licensed in Ontario, employment agencies hold a licence in BC, and charging a worker a fee for placement is prohibited in both.
How the system complies
The build changes nothing about the licence posture and touches no candidate money. Placement and submittal records accumulate in the ATS, where a licensing audit would expect to find them.
- Ontario ESA job-posting AI disclosure
What it demands here
Since January 2026, publicly advertised Ontario postings from larger employers must disclose the use of AI to screen, assess or select applicants.
How the system complies
The system's role is narrow enough to disclose in one sentence, ranking against stated req criteria with a recruiter deciding, and we help write that sentence so it is true as printed.
- Client-side automated hiring rules (NYC Local Law 144 and similar)
What it demands here
Clients hiring into jurisdictions with automated employment decision rules inherit bias-audit and notice duties for the tools in their process.
How the system complies
Decisions stay with recruiters by design, and the adverse-impact test results and the audit trail exist for the client whose counsel asks.
Candidate data is handled to respect PIPEDA and BC PIPA, outreach is CASL-first, and every screening decision traces to a named human.
The objections
Candidates are drowning in AI recruiter spam. Automation will torch our reply rates.
Volume without research is what torched them. The anchor build won replies because every message carried research a person would have been proud of, done in minutes instead of hours, and the same mechanic writes to the person, not the persona. A recruiter approves every send, and a reply pulls the thread straight back to a human. The desks that lose this era are the ones blasting templates faster.
Our ATS is ten years of duplicates, dead numbers and half-filled records.
Expected. The first phase dedupes and indexes the database, then grades matching against reqs you already filled, where the placement is known. If the index cannot re-find the people you actually placed, we tell you that before anything is built on top of it. Most databases pass, because the submittal history and screen notes are far richer than the profile fields everyone stopped filling in.
Ontario makes us disclose AI in job postings now. Will that scare applicants off?
The disclosure is one sentence, and it reads better when it is specific. Ontario's ESA has required publicly advertised postings from larger employers to say when AI screens or assesses applicants since January 2026. You can print exactly what this system does, ranks against the req's stated criteria while a recruiter makes every decision, because that is what it does. The postings that should worry are the ones whose honest sentence would say something else.
Recruiting is a relationship business. The database is not the desk.
Correct, and the build never touches the relationship. It does the research, the scheduling and the pack assembly that keep a recruiter off the phone. What comes back is phone time, and the desk that spends it on candidates wins the req. Nothing here makes an MPC call, negotiates a fee or talks a counteroffer down. That was never the slow part anyway.
LinkedIn Recruiter already finds everyone.
It finds everyone for every agency at once. The same boolean string returns the same strangers to you and to the two competitors racing you to submit. Your ATS is the one database they cannot search, and it is full of people who have already answered your calls. Retrieval over your own bench is the only search a competitor cannot run.
How we build it for recruiting and staffing firms
Index your own bench first.
Matching runs against people you already know, not a keyword search over a job board. Accuracy is graded against reqs you have already filled.
Test for adverse impact before anyone is ranked.
A matching system that has not been checked for disparate outcomes across groups is a legal problem, not just a quality one. That test runs before the first shortlist.
Automate outreach, never rejection.
Research, shortlisting and drafting are automated. A recruiter approves every send and every decision about a person.
What we will not automate
Rejecting a candidate, and any score that cannot explain the factors behind it. Screening produces a ranked shortlist with reasons, never a silent filter.
Where your team works
Tour the platformWorkflow builder
Describe the workflow. Watch it assemble.
Say what should happen in plain language and the builder assembles the automation on a canvas you can read, run, and change.
Task board
People and agents, working the same board.
Every piece of work lives on one shared board, whether a person or an agent owns it. Handoffs between the two are explicit, so nothing falls in the gap.
Ask your brain
Ask a question. Get the answer and where it came from.
Every answer cites the document behind it, so you can check the source yourself. When the brain does not have the answer, it says so instead of guessing.
Cost per outcome
Every dollar of spend traced to the work behind it.
Outcomes delivered, cost per outcome, value attributed, return on spend. The same measurement the guarantee is settled against, live on one page.
By team
The same system, seen from the desk that runs it.
Run your numbers.
Your operations
Savings use the low end of our hours-reclaimed range. The math is conservative on purpose.
The math
Calculated at the low end of every range.
Every first build is covered in writing: 5x ROI in 30 days. Or we work for free.
The hard questions
The systems behind this
Free · 3-5 days
Know your number in five days.
We map your operations, find the highest-ROI automations, and hand you a ranked plan with the payback math attached. Yours to keep, whoever builds it.
Prefer to talk first? Book 15 minutes with James. No pitch deck.