AI for Recruitment
Agencies
In a recruitment group, candidate data and consultant workflows already run through a specialist CRM. We build around that system, improving screening, communication and record quality without creating another place for consultants to update.
14.6 hrs
PER RECRUITER PER WEEK ON SEARCH (BULLHORN GRID 2026)
1 in 4
CANDIDATE PROFILES FAKE BY 2028 (GARTNER)
High-risk
UNDER EU AI ACT ANNEX III
Recruitment admin has changed faster than the job
A recruitment agency still gets paid for one thing: putting the right person in front of the right client, faster than the next firm. That part hasn't moved since 2005.
Everything wrapped around it has. Candidates apply with AI-written CVs. Some of those candidates aren't real. Your consultants compete with tools that've read every profile on the internet. The ICO and the EU AI Act now have opinions on how you screen.
Agencies that bend AI to the consultant's day are pulling away. Bullhorn's 2026 GRID report puts them at 3.5 to 4.5 times more likely to be growing revenue.
THE OLD DAY
- Boolean strings into LinkedIn Recruiter
- Reformatting CVs in Word until 8pm
- A database of 60,000 contacts, 800 active
- Compliance kept in someone's head
- Fake CVs you only catch on day one
THE AI DAY
- Briefing in, shortlist out, in minutes
- Branded CV ready in under a minute
- Dormant database worked overnight
- IR35, RTW and audit log on rails
- Synthetic candidates flagged at apply
Recruitment work worth automating
The same jobs appear across perm, contract, executive search and RPO. We automate them behind the CRM your consultants already use.
Parse and reformat
CV in, structured candidate record out, branded client-ready document attached. The 30 to 45 minutes per CV your consultants spend in Word becomes a click.
Search the database you forgot
Semantic search across the records you've paid to collect for a decade. Job spec in, ranked shortlist from your own CRM out, with the consultant who placed them last time named.
Screen without breaking the law
Skills-based scoring, named criteria, a written audit trail per decision. Ready for an ICO request, an Equality Act claim, or a client asking how you ranked their shortlist.
Catch the fakes
Synthetic CV detection at application. Liveness and identity checks before a video interview. The workflow includes sanctions, right-to-work and IR35 checks.
Talk to people, around the clock
Voice and chat agents for first-touch screening, availability checks and reference chasing. Consultants get a transcript and a summary. They make the call that matters.
Three risks to plan for
These three issues need different responses: candidate fraud controls, fair and reviewable screening, and a clear policy for consultant use of general AI tools.
iTutorGroup, $365,000.
The EEOC's first AI hiring discrimination settlement (consent decree, September 2023). The screening software auto-rejected female applicants aged 55+ and male applicants aged 60+. Over 200 applicants harmed. Recruitment is now Annex III "high-risk" under the EU AI Act, and the ICO's 2024 audit of UK recruitment AI tools led to 296 recommendations across the providers and agencies it looked at.
KnowBe4 hired a North Korean.
A security firm. Four video interviews, a clean background check, verified references. The new starter was a DPRK operative using a stolen US identity with an AI-augmented stock photo (KnowBe4, July 2024). Gartner says by 2028, 1 in 4 candidate profiles globally will be fake.
Mobley v Workday.
In May 2025, a US federal judge granted collective certification under the ADEA against Workday's screening AI. Applicants aged 40 or over who were rejected through it from 24 September 2020 onwards can opt in. The court let the case proceed on an "agent" theory under which the vendor may be directly liable alongside the employers using its software.
Sources: EEOC v iTutorGroup consent decree (E.D.N.Y.), KnowBe4 incident write-up (July 2024), Gartner, ICO AI in Recruitment Outcomes Report (November 2024), Mobley v Workday (N.D. Cal., May 2025 conditional certification).
How we start
We don't sell a platform. We build alongside your Bullhorn, Vincere, JobAdder, Mercury xRM, Voyager or Salesforce. Your data stays where it is.
Four phases, fixed scope, first working agent live. You can stop at any phase. Most don't.
BOOK A WORKING SESSIONConsultant-day audit
We shadow representative consultants and review the CRM process, record quality and administrative volume. The resulting shortlist ranks workflows by billing impact, readiness and compliance risk.
Wire up the company brain
CRM, mailbox, call notes, job specs, placement history, candidate documents and your compliance rules in one searchable layer. Without it, every agent you build is guessing. With it, every agent is reading from the same truth.
Launch the first agent
A contained workflow such as candidate assessment, CV processing or database reactivation runs on representative desks first. Consultants and the CRM owner review the results before wider use.
Add the next three, audited
Screening agent with a bias audit. Voice-first availability check. Compliance and RTW workflow. Each one logged, evaluated against your own historic placements, and documented in a format the ICO and your insurer will accept.
Works with your recruitment stack
Our AM2PM work connects candidate assessments, reminders, screening, e-sign and bulk record tasks to the recruitment CRM. We assess native CRM features and specialist products such as Daxtra or Textkernel before proposing custom work.
CRM / ATS
Bullhorn
CRM / ATS
Vincere
CRM / ATS
JobAdder
CRM / ATS
Mercury xRM
CRM / ATS
Voyager Infinity
CRM / ATS
RDB ProNet
CRM / ATS
Salesforce
PARSING
Daxtra / Textkernel
MULTI-POST
Broadbean / idibu / LogicMelon
SOURCING
LinkedIn Recruiter
COMPLIANCE
Xref / TrustID / Yoti
BACK OFFICE
Xero / Sage / Kingsbridge
AM2PM is a live recruitment example
Its recruiters build and send candidate assessments, receive graded reports in the CRM and use connected automations for reminders, screening, e-sign and bulk record work.
When this is worth discussing
We work best when there is a real operating problem, enough volume to measure and people from the affected teams who can make decisions.
Usually a good fit
- An established UK business, usually with annual revenue above £10m
- A repeated process with a known cost, delay, error rate or capacity problem
- A senior sponsor and a day-to-day owner who understand the work
- Access to the relevant staff, systems, sample records and security requirements
We may point you elsewhere
- A standard product already covers the process well
- The requirement is a one-off small build with no wider operating case
- There is no owner or access to the people and data needed to test the result
- The plan relies on AI making high-impact decisions with nobody responsible for review
Questions the buying team will ask
Bullhorn / Vincere already say they do AI. Why do I need you?
We first use the relevant features already available in the recruitment platform. Custom work is justified where your sectors, client templates, compliance rules, placement history or connected systems create a process the standard product cannot support.
Will my consultants resist this?
Not when the first thing they see is a CV they didn't have to reformat and a shortlist from a database they'd given up on. Consultants resist anything that adds admin to their day. The agents we build take admin off it.
What about the EU AI Act? We're UK-based.
Recruitment systems are named in Annex III as high-risk, but territorial scope and your role as provider or deployer still need checking. The European Commission's current implementation timeline puts those high-risk rules on 2 December 2027. We design for logged decisions, bias evaluation, human review of adverse actions and a deployer record, while also applying current UK data-protection requirements.
How do you stop the AI being biased?
Skills-based scoring against the criteria you set, not inferred protected characteristics. We backtest against your own historic placements and publish disparate-impact rates by gender, age band and ethnicity (where you collect it). If a model can't be shown to be fair, it doesn't go live. NYC Local Law 144-style bias audits if you sell into the US.
Half my CVs are AI-written now. Doesn't that break parsing?
It breaks keyword filtering. Modern LLM-driven parsing reads the document the way a consultant would and pulls out skills, dates and seniority signals reliably. The harder question is verification: is this person who they say they are? We score every CV for signs of synthetic generation and identity inconsistency, and flag the ones worth a second look before your consultant picks up the phone.
We're a 12-person agency, not Hays. Is this overkill?
The opposite. Hays can outsource back-office to Cognizant and call it AI strategy. A 12-person firm can't. You're the customer this is built for: small enough that one good agent gives every consultant an extra day a week, and you'll see it in next month's billings.
Where does our candidate data go?
It stays in your CRM and in UK or EU-hosted infrastructure we set up under your control. We use enterprise model endpoints with data processing agreements, no training on your data, and zero-retention configurations where they're offered. You see, and approve, every place a candidate record is read or written.
How much does it cost?
Audit is fixed-fee. First agent is priced per phase, scoped against the audit. Each agent after that is fixed per phase too. Optional monthly retainer once you're running, to evaluate, tune and add the next ones. You see the number before we touch a line of code.
Talk to us about your recruitment workflow
Show us the CRM, a representative desk and the administrative work that delays consultants. We will identify the first workflow with enough volume, data and ownership to measure properly.