Salient

Worked example· Recruitment & staffing

How we would save a recruitment agency 20+ hours a week.

This is not a client case study. We have not built this for an agency yet, and pretending otherwise would be the first thing you should not trust us on. It is the architecture we would propose on day one, with the numbers we would hold ourselves to.

The example agency: five consultants, ten live roles at any time, perm and contract, running on a mainstream ATS with a lot of email around it.

Where the week actually goes

Not the billable hours. The hours nobody logs, because they do not feel like work worth logging.

Screening and shortlisting18Roughly 50 applicants per role, 3 to 5 minutes each, across 10 live roles.
Reformatting CVs to client templates5Copy, paste, anonymise, restyle. Once per submitted candidate.
Interview scheduling6Three-way availability by email, then the reschedules.
Candidate and client status chasing4The same four questions, asked by different people.
Weekly client reports3Assembled by hand from the ATS and someone's memory.
Total, across the team36hours per week, before anyone speaks to a candidate

What is left afterwards

Screening and shortlisting5Recruiter reviews a ranked shortlist with evidence, rather than reading every CV.
Reformatting CVs to client templates0.5Generated on submit, in the client's format, checked once.
Interview scheduling1Candidate books into real availability. Exceptions come to a human.
Candidate and client status chasing1Status changes push automatically. People ask less.
Weekly client reports0.5Drafted from pipeline data, edited before sending.
Total after828 hours a week returned to the team

The remaining eight hours are the part that should stay human: judgement calls, exceptions, and the conversations that decide whether someone is actually right for a role.

The architecture

01

Ingest

Applications land from the ATS, the inbox and the job boards into one queue. Nothing is deleted and nothing is auto-rejected.

02

Parse

Each CV becomes a structured profile: skills with evidence, tenure, location, notice period, salary expectation, right to work. Free text becomes something you can filter and audit.

03

Rank, never reject

The role spec becomes explicit criteria. Candidates are ranked against it with a written reason per criterion, and a confidence score. The recruiter sees a shortlist with reasoning, not a verdict.

04

Fairness controls

Name, photo, age, gender and address are stripped before any scoring. Criteria are fixed per role, applied to every candidate, and every decision is logged with its inputs. If you are ever asked why a candidate was not progressed, you have an answer.

05

Human handover

Low confidence, unusual profiles and anything touching a protected characteristic route straight to a person. The handover is a designed path, not a failure state.

06

Outputs

Client-formatted CV packs, scheduling links against live calendars, status updates that fire on pipeline changes, and a weekly report drafted from real data.

How we would prove it works

This is the part most AI projects skip, and it is the reason most of them quietly stop being mentioned.

Before anything goes live, we build a golden set from roles you have already filled. Fifty past roles, with the candidate you actually placed. Then we ask one question: does the system rank the person you hired inside the top five?

If it does not, it is not ready, and no amount of impressive-looking output changes that. The same suite runs on every change afterwards, so an improvement to one role type cannot quietly break another. You get the score, not a demo.

What it costs, and what it returns

Saved hours are the wrong way to value this. Agencies do not bank saved hours, they convert them into placements. So here is both.

Build, 6 to 8 weeks, fixed scope$24kIngest, parsing, ranking, fairness controls, evals, one ATS integration.
Operate, from month two$4.5kPer month. Monitoring, model cost control, eval regression, small changes.
Hours returned28Per week, across five consultants.
Break-even1Additional placement per month, at a $12,600 average fee on a $70k role at 18%.

Every number on this page is an assumption until we look at your data. The first thing we would do is replace them with yours, in a paid one-week audit, and tell you honestly if the case does not hold.

If any of this looks like your week, the call is worth 30 minutes.

Bring one role you filled recently and we will walk through where the hours went. If there is no real project here, we will say so on the call rather than in an invoice.

Book a strategy call