Recruitment teams often face more applications than they can review comfortably. AI assistance can organise information and support initial analysis, but a faster shortlist is not useful if it excludes suitable applicants for irrelevant reasons or cannot explain the basis of its recommendations.
Start by defining the role and the selection evidence. Without those foundations, a model may substitute patterns in past hiring data, polished writing or convenient keywords for the capabilities the job actually requires. The organisation remains responsible for the recruitment process it chooses.
Write the criteria before choosing the tool
Create a role scorecard with essential requirements, useful additional experience and evidence that can be assessed fairly. Distinguish a genuine requirement from a preference. A requirement for a particular qualification should have a job-related explanation, while experience described differently across CVs should still receive a consistent review.
Keep the criteria visible to recruiters and interviewers. Do not let each person ask the AI to make a shortlist using a different interpretation of “best candidate”.
Use assistance for bounded tasks
Suitable starting tasks can include summarising relevant experience, organising an application against the scorecard and drafting interview questions for review. Require the output to identify the evidence it used. If the source does not establish something, the tool should mark it as unknown rather than infer a qualification or employment history.
- Verify summaries against the submitted application.
- Check equivalent wording and transferable experience.
- Review applications the tool marks as incomplete.
- Avoid using protected or irrelevant personal characteristics as selection criteria.
- Keep interview and appointment decisions in an accountable review process.
Make human review meaningful
A reviewer must be able to examine the application, question the recommendation and change the outcome. Clicking approval without access to the underlying information is not a substantive review. Allocate enough time for borderline cases and keep a sample of both recommended and non-recommended applications for quality checks.
Kenya's Data Protection Act, section 35 addresses solely automated decisions with legal or similarly significant effects, including safeguards and exceptions. Assess the proposed workflow against that framework instead of assuming that any use of AI is either unrestricted or prohibited.
Explain the process to candidates
Tell applicants how their information will be used and provide an appropriate contact for questions or corrections. Be clear about where a tool assists the process. Candidates should not need to guess whether an unusual rejection resulted from an unreadable document, a missing field or a deliberate selection decision.
Use accessible application methods and provide a route for resolving submission problems. Formatting conventions, file conversion and language differences can affect what the tool extracts even when the candidate's experience is relevant.
Evaluate errors and consistency
| Check | Evidence to inspect |
|---|---|
| Extraction accuracy | CV facts compared with the summary |
| Criterion consistency | Similar evidence assessed against the same scorecard |
| False exclusions | Suitable applicants omitted from recommendations |
| Reviewer independence | Recorded overrides and their reasons |
Use a reviewed sample and appropriate safeguards when analysing outcomes. Small numbers and incomplete information do not justify confident claims that a process is fair. Document the limits of the assessment and investigate patterns that need closer attention.
Keep records proportionate and secure
Agree which application records and review notes need to be retained, who may access them and when they should be deleted. Review provider terms before sending candidate data to an external service. A hiring tool should not become a new uncontrolled store of identity information and private employment histories.
Trinity's Targeted Recruitment service can help establish a structured selection process. The Recruitment Effectiveness tool offers a starting point for examining the workflow before adding more automation.
