AI for HR that is fair by design and useful by Monday
CV screening against written criteria, policy and payroll assistants, onboarding copilots and case triage for HR teams and HR software vendors, shaped by building and running ZekoHR, our own multi-tenant HR platform.
HR is a queue of documents and questions with a person's livelihood at the end of each one. AI shortens the queue; people keep the decisions.
CV screening against written criteria
Applications parsed and scored against the job's documented requirements, with the evidence for each score shown to the recruiter who builds the shortlist. Touches the ATS or ZekoHR recruitment module. See recruiting AI.
Policy and handbook assistant
Leave, expenses, parental policies, grievance procedure: answered from the current handbook with the paragraph cited, per country and per entity. Touches the document store and the intranet. See internal copilot.
Onboarding copilot
A new starter's first weeks guided step by step: documents to sign, systems to set up, people to meet, questions answered in context. Touches the HRIS, identity provider and document module. See AI onboarding and training.
Payroll and benefits queries
Why is my net pay different this month: explained from the employee's own payslip data after authentication, with the payroll team taking over anything that looks like an error. Touches payroll and the employee self-service app.
Case and grievance triage
Incoming cases summarised, categorised and routed to the right HR partner with the relevant policy attached, confidentially. Touches the case management module. The HR partner owns the case; the summary saves the first hour.
Expense and attendance anomaly checks
Duplicate claims, out-of-policy spend and attendance patterns that need a conversation, flagged for a manager rather than auto-rejected. Touches the expenses and attendance modules. See anomaly detection.
Compliance and data
Employee data, automated decisions and equal treatment
What we design for in HR
Employee and applicant data is personal data under GDPR and UK GDPR, and decisions with a significant effect on people cannot be left entirely to a machine. AI used for recruitment and employment decisions is classed as high-risk under the EU AI Act, which brings documentation, human oversight and transparency duties. Equality law means criteria must be job-related and outcomes monitored. In some countries employee representatives must be consulted before such systems are introduced. Applicant data has its own retention limits. This is engineering guidance, not legal advice.
We keep the human in the decision, show candidates and employees when AI is involved, log the evidence behind every score and give HR the outcome monitoring to spot problems early. The AI governance service produces the documentation a high-risk classification demands.
Reference architecture
Flagship workflow: recruiting pipeline with a recruiter in charge
The job specification and its scoring criteria are written and versioned before the first application arrives; they define what the screening step is allowed to consider. Applications from the ATS and job boards are parsed into a consistent structure.
Screening scores each candidate against the criteria with the supporting evidence quoted from the CV. The recruiter sees the ranked list and the evidence, adjusts and builds the shortlist. Their overrides are logged. A scheduling agent then arranges interviews across calendars and writes everything back to the ATS.
Underneath runs the fairness layer: criteria review before launch, pass-rate monitoring by stage and an audit record per candidate. This is the pattern inside ZekoHR's recruitment module, described in the ZekoHR case study.
The model ranks with evidence; the recruiter decides; the fairness layer watches both.
Track record
What we have built that applies
ZekoHR, our own HR platform
Payroll, attendance, recruitment, documents, expenses and grievances in one multi-tenant SaaS. We know HR data models, approval chains and payroll edge cases from running them.
Our production support bot answers from approved content and cites it, the behaviour a policy assistant needs when the answer affects someone's pay or leave.
HR, talent acquisition, legal and employee representatives review the candidate use cases and the data each one needs.
Week 2: criteria and controls
For screening, the criteria are written and reviewed; for assistants, the source documents are chosen. Oversight and transparency rules are agreed.
Weeks 3 to 4: prototype on past data
Screening replayed on a closed requisition and compared with recruiter decisions; the assistant tested by HR partners on real questions.
Day 30: decision pack
Agreement rates, hours saved, fairness monitoring design, governance documents and a rollout plan.
If you sell HR software, read about AI integration for SaaS products; otherwise book a call with the team.
FAQ
Frequently asked questions
Is AI allowed to screen candidates?
Screening against written, job-related criteria with a recruiter reviewing the shortlist is a common and defensible pattern; letting a model reject people on its own is not. Employment decisions are treated as high-risk in the EU AI Act and attract rules on automated decision-making under GDPR and UK GDPR, so we design for human review, transparency to candidates and monitoring of outcomes by stage. See EU AI Act readiness.
How do you check the screening is fair?
Criteria are written down and reviewed before any model sees a CV. In operation we monitor pass rates by stage and compare them across groups where you lawfully hold that data, and the recruiter's overrides are logged and fed back. We do not infer protected characteristics from CVs.
Can the assistant answer payroll questions from our actual system?
Yes, for the employee asking about their own record, after authentication. It reads from the payroll and HRIS through controlled APIs and explains the figures; it never changes them. Anything unusual goes to the payroll team with the context attached.
We are an HR software vendor. Can you add AI to our product?
That is what we did for ZekoHR, our own HR platform, and it is a large part of our work: embedding screening, assistants and document intelligence into multi-tenant SaaS with per-tenant isolation and cost control. See AI integration.
Ready to make AI work in production?
Tell us about your systems, your data and your constraints. We reply within one business day with a straight answer and a suggested next step.