Skip to content
CogniYukti

Applicant tracking · agency and in-house

From a job description you half-trust to a hire you can defend.

Requisition to placement, with client work and your own hires in the same list.

CogniYukti reads the role before anyone is matched against it, shows the sentence behind every extracted field, and returns the interview quoted — so what reaches your shortlist is evidence with a source, not a score you have to trust.

  • Every extracted field shows the sentence behind it
  • The interview returned as evidence, not as a tick
  • Agency and in-house in the same list

Interview · round 2

32:10

  • 04:12Claim

    “I owned the migration — three teams, about eighteen months.”

  • 11:48Risk

    “My notice is ninety days, and I'd want to serve it in full.”

  • 19:03Not chased

    An answer that trailed off, and no follow-up question after it.

Each finding seeks the recording to the second it happened. Anything that cannot be verified against it is dropped rather than shown.

Illustrative

What an applicant tracking system is

A tracker knows which stage someone is in. It rarely knows why.

An applicant tracking system is the system of record for hiring — requisitions, candidates, applications, interviews, offers, hires. For an agency it also carries clients, submissions, placements, vendors and the fee. The category name is literal: it tracks. It knows where a candidate is and when they moved.

What it conventionally does not hold is the reasoning. That lives in a recruiter's head, in four sets of interview notes that disagree, and in an email thread with the client. Which is fine until somebody asks why — a client about a rejection, a hiring committee about a shortlist, an auditor about a decision made eighteen months ago.

It usually does not hold the disagreement either. On client work there are two truths about one candidate — where your side has got to, and what the client has actually said — and a single status field forces a recruiter to choose which half to record. Here they are separate axes, which is how a client's silence becomes a number instead of a grievance.

Reading a CV
Open it, skim it, type six fields into the tracker, hope you read the dates right
Fields come pre-filled, each showing the sentence it came from. You confirm or correct
Understanding the role
A two-line brief, and whatever you remember from the kick-off call
The JD is read into a ledger with the gaps named — and a link that asks the hiring manager about them
The interview write-up
Four sets of notes, typed afterwards, that disagree with each other
Quoted from the recording, with the panel's ratings checked against what was actually said
Defending a rejection
“We felt they weren't quite right for it”
The requirement, the line that evidenced it, and the one that evidenced nothing
Telling the client where things stand
A spreadsheet you rebuild every Thursday
They sign in and look. The status call becomes a conversation about the shortlist
Getting the fee invoiced
Re-enter the placement into an accounting tool and hope the terms match
The placement knows its fee terms. The invoice drafts from it, with GST applied

Before anyone is sourced

Most hiring goes wrong at the brief, not at the shortlist.

Forty words of job description, half of them boilerplate, and a kick-off call somebody remembers differently. Everything downstream inherits that — the search, the screening, the questions asked in the room.

CogniYukti reads the description into a ledger: section by section, with the evidence behind each reading attributed to the text it came from, a coverage score per section, and the gaps named rather than glossed. Then it offers a link that asks the hiring manager about exactly those gaps, and their answers become the next version.

What the brief said

  • 8+ years, data engineering
  • Python, SQL, cloud warehouse
  • Hybrid, two days on site

Enough to post. Not enough to search.

What nobody wrote down

  • Is regulated-industry exposure required, or nice to have?decides half the longlist
  • Team of one, or leading three?changes the seniority entirely
  • Is the budget the budget?the reason offers get withdrawn

These are asked before anyone is sourced, and the answers become part of the role — not a thread somebody remembers.

Illustrative

Reading a CV, and matching against it

Every field shows the sentence it came from.

Upload a resume and the fields come back populated — each one carrying its source quote and a confidence indicator, each one clickable through to the line in the document. The recruiter confirms or corrects before anything is saved. That review step is the feature: extraction you cannot check is extraction you have to re-do.

When a recruiter corrects one, the correction sticks. The next extraction in your workspace uses your phrasing, your clients' job titles, the way your market actually writes things down. The same is true in reverse for a job description — paste it in, thirteen canonical fields come back for review, and your edits teach it.

And before any of that costs you anything: one action reads every candidate you already hold against a role, sorts them into four bands, and names the requirement that blocked each person — hard blockers kept apart from preferences you could override. No AI call, no per-record cost. It is the pass you run first, and it answers the objection every buyer has to AI matching before the demo reaches the interesting part.

What the shortlist says4 requirements
  • Production migrationDirect evidenceCV · line 1
  • Ran it, not observed itConfirmed in interviewTranscript · 00:14
  • Led a squad of 6–8Direct evidenceCV · line 3
  • Regulated-domain exposureNothing evidences thisNot found

Every row names the line it came from — and the one that names nothing is the gap, stated rather than averaged away.

Job description

Senior Data Engineer · hybrid

  • Own the streaming layer end to end
  • Must have run a production migration
  • Lead a squad of 6–8 engineers
  • Regulated-domain exposure essential

Candidate CV

11 years · data platform engineering

  • Led the Kafka cutover in 2024
  • Reported to the Head of Platform
  • Ran a squad of 7 engineers

Interview transcript

Panel round 2 · 48 minutes

  • “We cut over on a Saturday, with two
  • people from SRE on the call.”
Illustrative
Illustrative

Requirement · production migration

Direct evidence — “Led the Kafka cutover in 2024.” Confirmed in the interview as owned rather than observed.

Requirement · leading engineers

Contextual inference — “Ran a squad of 7 engineers.” Never stated as a management title anywhere in the document.

Career shape

Platform depth in one domain, one step up every three to four years. Consistent with the role's trajectory, not a sideways move.

Gap · regulated-domain exposure

Nothing in the document evidences this, and the role marks it essential. Stated rather than averaged into a score.

A verdict, opened. Every row is answerable to a line in the document it came from — including the one that is answerable to nothing.
  • Per requirement, not per candidate

    A shortlist you can walk through line by line is a different object from a ranked list you have to trust.

  • Boolean search over your own corpus

    Plus saved searches that alert you when somebody new matches one.

  • Candidates arrive from everywhere

    Candidate-database search with import, ten job-board adapters, your own careers page, email intake, a browser extension that checks for duplicates before it writes, and referrals recorded as a chain.

  • Any role family, not only engineering

    The example above is a data engineer because technology roles are where most people expect this to work. The machinery is not role-specific: the same reading runs on a plant manager, a relationship manager, a staff nurse, an accountant or a field sales lead. What changes is the requirements it finds, not how it finds them.

The interview evidence layer

What was said in the room, not just that the room happened.

Every applicant tracker will help you schedule an interview. Almost none of them can tell you what was said in it — and that is where the decision actually gets made, usually from one person's memory and four lines typed afterwards.

A bot joins the call and records it, with consent captured on the file and a retention date attached. The write-up comes back quoted. Hover a finding below and it highlights the line it came from.

Interview · 32 min · recorded4 marks
  1. 04:12ClaimLed the migration end to end
  2. 11:48RiskNotice period longer than the brief allows
  3. 19:03Missed follow-upAnswer trailed off; nobody followed up
  4. 26:35Question craftTwo-part question — the second half went unanswered
Every mark carries the moment it happened. The one at 19:03 is about the interviewer, not the candidate — which is the half of an interview nobody else reports.
Illustrative

Nothing is invented

The recording is transcribed, and a finding that cannot be verified against it is dropped rather than shown. The failure mode is silence, not confidence.

The panel is checked too

Each panellist's rating sits beside what the interview itself produced, competency by competency, and the rows where they are furthest apart are marked before the debrief.

The interviewer is read too

Question by question: did it ask two things at once, and did the answer land, land halfway, or go somewhere else. Labelled as coaching on question craft, because that is what it is.

Before the round, too

A prep brief per interviewer: what the evidence already shows, what is still open, and what the last round did not get to.

CognityHire also builds Cognisense, an adaptive interviewing product in its own right. It is a companion, not a stage inside this module — the hookup is not built, and we would rather say that here than let a demo imply otherwise.

Closing it, and getting paid for it

The placement is a commercial object, not a status.

The terms live on the requisition — bill and pay rate, the fee this client negotiated down, exclusivity, submission cap, days promised to first shortlist. The recruiter working the role can see them without opening a contract, and the placement that comes out already knows what to bill.

Offers route for approval on their own numbers, with a time limit on each step, and are signed in place — with the document's fingerprint, the consent wording and the signer's circumstances recorded beside the signature. Retained work is held as tranches with their own triggers. If a candidate leaves inside the guarantee, the breach is one action: the credit note drafts and the commission already paid is reversed.

Rejections that read as respect

Rejection, hold and withdrawal letters, written from the reason your recruiter chose and the notes from the interview — never a score, never a generic verdict. On client work they sign on the client's behalf.

The fee has a life after the close

A slice is held back through the guarantee and released the night it closes clean. If it breaks, one action drafts the credit note and claws back commission already paid.

Drives, and what they were worth

Campus weeks, walk-in days and hackathons as events with cohorts under them, the institution recorded with its tier, and an audit trail behind every batch — so next year's calendar is a question with an answer.

Everyone else who needs to answer you

Three portals, and what each one is allowed to decide.

A login that only shows things is a nicer way of waiting. Your client shortlists, rejects — and cannot file that rejection without a reason — picks the interview slot, files a scorecard, and approves or disputes the invoice under a one-time code emailed to them. Almost every applicant tracker has a client login that shows a list. Almost none has the client signing off what you billed.

Your partner firms get a boundary with teeth: the first firm to submit owns that candidate for the window in their own agreement, and a second firm is told before they waste the call — seeing a name and nothing else. They also measure how long you take to answer them, on the same figures you see.

Candidates get the narrow one: where they stand, an offer to accept or decline, one-tap apply to another opening, and data requests that run immediately — including the refusal when a legal hold blocks one.

Your client

  • Their requisitions, and only theirs
  • Submissions, with two side by side
  • Interviews arranged, and the outcome
  • Their invoices
  • A tracker report on the day you said

Nothing from the client down the corridor.

A partner vendor

  • Only the requisitions you assigned
  • Their own submissions and placements
  • Their agreement and their team
  • Their invoices, raised here
  • Their own performance analytics

Candidate contact details stay yours until you decide otherwise.

The candidate

  • Where their application stands
  • Interviews, and self-scheduling
  • Their offer, signed in place
  • Other openings worth a look
  • Their own privacy controls

Your notes about them are not theirs to read.

All three carry two-way threads with unread counts — a reply from either side lands as a notification on the other.

For firms with a bench

Idle is a daily cost with a date on it.

Two facts decide a Monday: how long this person has been sitting, and what that has added up to. CogniYukti computes both from the daily cost and the bench start date, so the screen is current when you open it rather than when somebody last updated a file — sorted by cost accrued, because the alphabet has never been the right order for this.

And the person at the top of that list is a candidate for the requirement that came in this morning, already in the system with skills, rates and notice period on the record.

One consultant · off project 71 days£420 / day
Day 1£420
Day 14£5,880
Day 30£12,600
Day 71£29,820
Nobody decided to spend the last figure. It accrued while the bench was a spreadsheet somebody updated on Fridays.
Illustrative

Analytics

Seven suites, and the card that tells you something you did not ask.

Six of them answer a question about the hiring itself — the funnel, role health, recruiters, clients, vendors, and one person's history across every role and client you have. The seventh watches your own automations: which rules fire, which fail, and which are hitting a cap somebody set too tight.

On top of those sit anomaly sweeps that surface a change nobody queried, and a plain-language surface for the question you have not built a card for yet.

Nobody ran this queryflagged this morning

One reviewer has not opened a submission in nine days, and three roles are waiting behind them.

  • Roles affected3
  • Submissions held up14
  • Median wait9 days
A dashboard answers the question you thought to ask. This is the other kind — and it is the reason a stalled role gets noticed in September rather than at the quarterly review.
Illustrative

One product, both businesses

Mode is set per requisition, not per company.

A client search and your own office hire sit in the same list and behave differently. The client search gets the portal, the fee terms and the invoice; your own hire skips all of that and hands over to employee records at offer. An in-house team simply never turns the client-facing half on.

That matters on the day it changes. Firms hire contractors, open a small recruitment arm, or start placing into a group company. When that happens nothing is migrated — the commercial half switches on for the requisitions that need it, and everything already in the system stays put.

  • Your fields, on real objects

    Requisitions, postings and candidates take fields nobody else tracks — reportable and searchable, not a notes box.

  • DPDP and GDPR as screens

    Access and erasure runs, consent purposes, retention policies and legal holds — for the candidates you never hired, whose data you hold longest.

  • Rules instead of plugins

    When a stage changes, when a candidate goes quiet, when a submission ages past the SLA — do the thing. Automations and approvals are part of the product.

Questions we get asked

The ones that actually decide it.

We are not an agency. Is this built for one?

Mode is set on each requisition, so an in-house team runs the internal path end to end and never sees a client portal, a fee term or a submission cap. The interview evidence, the approvals, the role understanding and the handover to employee records are identical either way. You get the platform with parts switched off, not a reduced edition of it.

Do our clients have to pay for seats?

Client portal users are invited by you and are not your team's logins. What a portal user can see is configured per client.

Does the recording need the candidate's consent?

Consent is captured on the recording itself, with a retention date attached to the file, and recording can be switched off per interview at the point of scheduling. The compliance screens will tell you what is held, for whom, and until when.

What happens to candidates we never hired?

They stay on one person record under retention policies you set, with access and erasure requests handled as product surfaces. If that person applies again in two years, or turns up as a hiring manager at a client, it is the same record with the history still attached.

Does this only work for IT and technology roles?

No. The extraction reads the document rather than matching against a fixed taxonomy of job titles, so it runs the same way on a plant manager, a relationship manager, a staff nurse, a chartered accountant or a field sales lead. Technology roles appear most often in our examples because that is where buyers expect this to work, not because the product is limited to them. Bring a non-technology role to the demo and we will run it on the call.

Can we keep the job boards we already use?

Ten board adapters ship, alongside candidate-database search, email intake and your own careers page. They all land on one candidate record, which is the part that matters.

We already have an HR system. Does that make this pointless?

No — hiring works standalone. But the reason people move both is that the handover between them is where the re-typing lives. You can switch the people module on later without a migration, because the records are already shared.

How do we get our existing data in?

An importer maps spreadsheets to real records, de-duplicates against what is already there, and reports what it could not place rather than dropping it quietly.

Is the AI going to make decisions for us?

It does not advance anyone, reject anyone or send anything on its own. It extracts, quotes and flags; a person decides. Where it cannot support a claim with a line from the source, it drops the claim rather than softening it.

What does it cost?

We quote rather than publish. A four-person agency and a fifty-recruiter firm running three portals are different products in practice, and a price list would mislead more often than it would help. A session on a live role gets you a number.

Do you do market mapping or off-limits tracking?

No. We are strong once you are talking to people — the evidence from those conversations, the client-facing side of the search, and the commercial layer underneath. If mapping is central to how your firm works, say so on the call and we will tell you plainly where we fit and where we do not.

Can we phone-screen candidates from inside the product?

Not today. The dialler is wired for sales calls on contacts and leads; candidate calls are not on it yet. Interviews reach the evidence layer through the meeting bot instead.

Book a demo

Bring a live role and its terms.

Thirty minutes inside the product, on a role you are working right now — its terms, its brief, its shortlist. We set it up live on the call.

  • Paste a real job description and show me the gaps it names
  • Extract a CV and open the source behind three fields
  • Open a verdict and show me the requirement nothing evidences
  • Press a finding in an interview write-up and hear the moment
  • Show me what my client sees when they sign in
  • Take a placement to an invoice, then break it inside the guarantee