A Thousand Applicants. Six People. Nothing Decided By A Machine.
Kestrel — a boutique recruitment agency in Australia
- Every application delivered
- The board has no knock-out question, so nothing is filtered before a person sees the count. The score ranks; it cannot refuse an application
- Seven gates, two of them the client’s
- The client approves the brief before an ad is written, and the offer before a letter exists. The agency holds the other five
- Zero screening calls
- The half-hour phone screen is out of the process. Interview 1 and Interview 2 are unchanged, and still face to face
How it used to work
One role, two dozen applications, and a phone that holds one conversation at a time.
24 applications, one role
everything through one phone
6 calls · 30 minutes each. Different questions, different notes, nothing that can be put side by side.
18 never called — not rejected, just further down the list.
Three hours of calls to learn what six minutes of answers would have told you
and the client saw none of it — they got the CVs and a paragraph
A role arrived over email and stayed there. The brief was whatever the client said on the phone, written up and sent back as a document that came back with tracked changes, or did not come back at all — and the ad went up anyway. Two dozen people applied to it. Six of them got a screening call: half an hour each, booked over three emails, run in whatever order people answered. The first candidate got the sharp version of the questions and the last one got the tired version, and neither conversation was written down in a form the client would ever see. The other eighteen were never called — not rejected, just further down a list that ran out before the day did.
A phone that holds one conversation
Every screen was half an hour of the recruiter’s day, plus the emails to book it. The queue decided who got called.
Eighteen of twenty-four never reached
Not rejected — just further down the list than the day was long. Nobody could say who had been read and who had not.
No two candidates asked the same thing
The questions drifted between calls, so the notes could not be compared — and the ranking had to be taken on trust.
The ad went up before the brief was agreed
The role brief lived in an email thread. Money was spent advertising a definition the client had never signed.
How it works now
An ad that posts on approval, two screens that propose, and seven checkpoints where a person says go.
The same six questions, the same limits — and nobody waiting for a slot in your day.
↑ the operator sits down — everything left of this happened unattended
All eight recorded on the same evening, at eight different times. None of it needed anybody watching.
Nothing here is a decision. Every score opens the second of video it came from, and a person still picks who the client meets.
Winning a client
Researched one at a time, when there was a gap
A weekly search against the agency’s own profile, researched, waiting at a gate
Agreeing the brief
An email thread and a document
The client edits it online and approves — no ad is written first
Advertising the role
Posted by hand, priced by guesswork
Written from the approved brief, priced live, published only on approval
Reading applications
Newest first, until the day ran out
Every one read against the keywords the client approved, and ranked
Screening a candidate
Half an hour on the phone, one at a time
Six timed questions, everyone at once, on their own evening
The ranking
A judgement the client had to take on trust
Three weights — and every score opens the second of video it came from
What the client sees
A CV stack and a paragraph
The answers themselves, in one link, with somewhere to respond
Who decides
Whoever had time to make the calls
Seven gates: the agency holds five, the client holds two
The screening call is the only thing this replaced. The rest of the change is shape.
A role now runs on rails. The client approves the brief in their browser — the description, the screening keywords and the questions — and only then is an ad written from it. The ad is priced against that client’s own account on the board and posted when a person approves the spend, so nothing is ever published or billed without a signature.
Then the applications arrive, all of them. The board carries no knock-out question, so nothing is filtered on the way in and the count on the screen is the real count. A resume screen reads every one against the keywords the client approved and puts fifteen forward. Those fifteen answer the same six questions on video — 90, 90, 60, 60, 45 and 45 seconds, one retake each — after typing their current and expected salary into a form, because nobody should spend 45 seconds of answer time saying a number. One rubric scores the answers on three weights: communication 30%, role fit 40%, values 30%. Six go to the client.
Both of those cuts are proposals. The screen ranked one applicant third and a person removed them; it ranked another forty-first and a person added them, and that candidate finished in the six the client is meeting. A third override did not pay off and is still on the screen. Nothing in the system reads a ranking and acts on it.
Seven checkpoints hold the process open, and two of them are not the agency’s. The client approves the brief before an ad exists, and approves the offer terms before a letter exists. Interview 1 and Interview 2 did not move — they are still face to face, still run by the client, and they now start from evidence instead of a CV.
A role, end to end
- WinFind and researchAutomated
Businesses matching the agency’s own client profile are found, researched, and written up with a discovery brief before anyone reads them.
Weekly
- ◆ Gate 1Human
Approve the targets
A person picks which are worth an email, and which list they join. Anything without a contact sits in a second queue with no approve button — enrolling it would mean emailing somebody the sequence cannot reach.
- Decides
- Who is worth approaching, and which campaign they join
- Nothing moves until
- Nothing is sent to a prospect
Whenever you sit down
- ◆ Gate 2Human
Approve the copy
The sequence is written and approved once, when the campaign is created. After that the agent decides only when it sends, never what it says.
- Decides
- The campaign copy, and the days it goes out on
- Nothing moves until
- No sequence runs
Once per campaign
- OutreachSequence, and stop on a replyAutomated + human
Approved targets get the copy that was approved. A reply stops that person’s sequence on its own, and a person reads every reply and decides whether it becomes a call.
Monday, Wednesday, Friday
- ◆ Gate 3 · your clientClient
This one is not yours.
The client approves the brief
They open it in a browser, edit the description, the screening keywords and the questions in place, and comment. Every change comes back with their name on it. The keywords matter most — they decide whose resume is read first.
- Decides
- The description, the keywords, the six questions and the rubric — edited, not just accepted
- Nothing moves until
- No ad is written
In the client’s own time
- ◆ Gate 4Human
Approve the ad and the spend
The ad is written from the approved brief and priced live against the client’s own account on the board. Approving it is what publishes it and what bills them.
- Decides
- The ad, the product and what it costs
- Nothing moves until
- Nothing is published, and nothing is billed
Per role
- ApplyEvery application arrivesAutomated
Applications are pulled down as they are made, with the resume attached. The board carries no knock-out question, so nothing is filtered on the way in and the count on the screen is the real count.
On arrival, at any hour
- ◆ Gate 5Automated + human
Confirm who records
The resume screen reads every application against the client’s keywords and proposes fifteen. A person confirms which fifteen — removing some it ranked highly, and adding some it ranked nowhere.
- Decides
- Which applicants are asked for answers
- Nothing moves until
- No invite goes out
Per role
- ScreenSalary, then six answers, scoredAutomated
Current and expected salary are typed into a form before recording unlocks. Then six answers upload as they are recorded, transcribe, and score against that role’s rubric within minutes. Fifteen candidates can be doing this at once.
On submission, at any hour
- ◆ Gate 6Human
Confirm the six
The rubric ranks everyone who finished and proposes six. A person reads the transcripts, writes the narrative, and overrules the ranking where it is wrong — the highest-scoring candidate on this role is not one of the six.
- Decides
- Who the client meets, and what the write-up says
- Nothing moves until
- The client sees nobody
Per role
- HireInterviews, face to faceHuman
The client watches the answers, then runs Interview 1 and Interview 2. Every stage move books the next meeting and syncs back, so nobody chases status by email. From here the role report is typed in by a person — nothing the agency runs can see a meeting room.
The client’s own process
- ◆ Gate 7 · your clientClient
This one is not yours.
The client approves the offer
Terms go to the client before a letter exists. Approving them is what writes the letter and sends it; the candidate signs it in the browser, and that signature is the last thing in the process that lands on its own.
- Decides
- The package, the start date and what the letter says
- Nothing moves until
- No offer letter exists
On a decision
Seven gates, and two of them are not the agency’s. No ad is written until the client has edited their own brief and pressed approve, and no offer letter exists until they have approved the terms. The two screens in between propose; they never promote.
See it working
Not screenshots — the agency’s back office, the candidate’s portal and the client’s own screens, on invented data. Click anything.
The back office
The dashboard every client, role and candidate runs through. Every screen clickable, on invented data.
- Two halves in one nav — winning the client, and filling the role — each carrying what is waiting behind it
- A role at every stage: client review, ad draft, advertising, screening, with the client, interviews, offer, filled
- A thousand applications narrowed to fifteen, then to six, with both overrules on the page
- A role report where six rows land on their own and three are typed in by a person
The candidate portal
The candidate’s side — the salary form, six timed questions with a real countdown, and the offer they sign.
- Current and expected salary first, in a form — the questions do not unlock until both are answered
- Six questions at 90, 90, 60, 60, 45 and 45 seconds
- A countdown that actually runs and stops the recording at the limit
- One retake each, and an offer signed in the browser at the end
What the client gets
The three screens that are not the agency’s: the brief they approve, the six they answer, and the offer they authorise.
- Edit the description, the screening keywords and the questions in place — every change is tracked back
- Approve it, and watch the ad become writable
- Then the six: progress, hold or decline each person, with their salary against the advertised range
- Then the offer — no letter exists until you approve the terms
The automation
What the automation actually is — seven agents, three clocks, seven gates. Invented figures.
- The contract: what it may do, and what it has no tool to do
- Fourteen tools that exist, and six drawn dashed because they do not
- Seven gates, grouped by which half of the business they belong to
- A week of runs, reported in numbers
Six things in there are worth finding. Go and look.
- 1A candidate who scored 8.4, whom the rubric put in its own top six, and who is not with the client. A person read the transcript, saw that her strongest answer was about a job she left in 2019, and held her. Her card records both the score and the decision.
- 2A brief that has been with the client four days. They have rewritten a question and left comments — and no ad has been written. Those four days would otherwise have been spent advertising the wrong role.
- 3A candidate who recorded two of six answers and stopped. He is not scored, not put forward and not chased. He shows as two of six, because a half-finished screen is a fact about him rather than an error to clear.
- 4A 7.4 that reads “consider” rather than “recommend”, and the sentence explaining why — she flagged her own payroll gap in question two. The system quotes her doing it, at 0:48, instead of deducting points silently.
- 5The applicant the resume screen ranked third, who never recorded a thing. Every keyword present, all of it earned inside a four-hundred-person shared-services centre, for a two-person practice.
- 6The applicant it ranked forty-first who finished in the six — and another added the same way who finished last of everyone. Both overrules are left on the screen, because a demo where every human override is vindicated is a sales deck.
Everything else screens cleanly. Each of these is the system doing its job on an awkward case — and four of the six are a person overruling, editing or waiting, which is the whole design.
What Kestrel still decides
- Nothing is published and nothing is billed without a person. The agent writes the ad and prices it; it has no tool that posts one nobody approved.
- The screens propose, they never promote. A thousand applications become fifteen and then six, and a person confirms both cuts before anybody is told anything.
- No applicant is filtered out on the way in. The board delivers every application, and there is no tool anywhere in the engine that removes one.
- The score is evidence, not a verdict. Nothing reads it to decide anything — it ranks a table for a person, and the person overrules it in both directions.
- The client holds two gates. No ad is written until the brief is approved, and no offer letter exists until the terms are.
- Candidate video lives in the Sydney region, every portal link is per-role and revocable, and the retention window is Kestrel’s to set.
Want this running for your business?