Recruitment is full of repetitive admin: acknowledging applications, sorting CVs, chasing availability, booking rooms, sending the same rejection fifty times. It is also one of the places where automation can do real harm if it is aimed at the wrong target.
The distinction I give every client is simple: automate the admin, not the decisions.
Admin versus decisions
| Admin (automate it) | Decisions (keep a person accountable) |
|---|---|
| Acknowledging applications | Who is shortlisted |
| Extracting details from CVs | Who is rejected |
| Scheduling interviews | Who is hired |
| Sending reminders and updates | What an offer contains |
| Collecting onboarding documents | Anything involving a judgement about a person |
The left column saves hours and carries little risk. The right column is where fairness, reputation and the law come in.
What is safe to automate
- Acknowledgement. Every applicant gets an immediate, courteous confirmation. This alone improves how people feel about your company.
- Information capture. Pull name, contact details, experience and key skills out of CVs into a structured record.
- Scheduling. Offer available slots, book the interview, send the invitation and reminders, handle reschedules. The mechanics are covered in appointment booking automation.
- Status updates. Keep candidates informed at each stage without anyone drafting emails by hand.
- Closing the loop. A prompt, respectful response to unsuccessful candidates, sent by a person's decision but delivered automatically.
- Collecting documents. Requests for references, ID and signed paperwork once someone is hired.
Screening: help, do not decide
The tempting step is to let AI read every CV and rank them. It can genuinely help with a hundred applications, but only if it is used as an assistant to a reviewer rather than a gatekeeper.
A sensible role for AI:
- Summarise each application in a few lines against the job's stated requirements.
- Highlight evidence for and against each requirement, quoting the CV.
- Surface the information so a person can review faster.
What it should not do is silently discard people. Every application should be visible to a human reviewer, with the AI's summary attached as a convenience.
The fairness and legal risks
Hiring is a regulated area. Employment and anti-discrimination law applies to automated tools just as it does to people, and some jurisdictions have specific rules on automated hiring systems, including audit and notice requirements. The EU's AI rules treat recruitment tools as high-risk. I am not a lawyer; take advice for your market before relying on automated screening.
Bias
Models can pick up patterns that disadvantage people by gender, ethnicity, age, disability or background, sometimes through proxies such as a name, an address or a particular university. A system that "learns what a good candidate looks like" from past hires can simply reproduce past prejudice.
How to reduce it
- Score only against the written requirements of the role, not general impressions.
- Keep personal details out of the model's view where the task does not need them.
- Test on varied examples, including similar applications that differ only in name or background, and check the results match.
- Review outcomes regularly, looking for patterns in who is progressing.
- Keep a person accountable for every shortlist and every rejection.
- Be transparent with candidates about how their application is processed.
Privacy
CVs are full of personal data. Know where they are stored, who can read them, how long you keep them, and what you send to external AI services. Data protection rules apply, such as the Nigeria Data Protection Act or GDPR depending on where you and your applicants are.
A workflow that stays defensible
- An application arrives and a record is created. The candidate receives an acknowledgement.
- Details are extracted into structured fields and stored securely.
- An AI step writes a short, evidence-based summary against the role requirements.
- A person reviews every application, using the summary as a guide, and decides who proceeds.
- The chosen candidates are offered interview slots automatically, and bookings, reminders and rescheduling run themselves.
- Each candidate's status update is triggered by the reviewer's decision.
- The reasons for each decision are recorded by the person who made it.
After the hire
Onboarding is where automation is uncontroversial and valuable. The offer is accepted and the system collects documents, creates accounts, sends the first-week schedule, introduces the team and schedules check-ins. It is the same pattern as client onboarding, and for a new colleague it makes the first week feel organised. More examples of this kind are in 18 AI automation ideas that save 20+ hours a week.
Used this way, automation makes hiring faster and more considerate without taking away the part that should stay human. The candidates remember being answered promptly, and the people deciding are still the people responsible.