Record Keeping That Defends an HR Decision at Tribunal
When employment decisions are challenged, whether in an employment tribunal, a discrimination claim, or a data protection investigation, the organisation’s records determine whether it can defend the decision. Record-keeping practices in particular are critical when AI is involved in hiring, promotion, or termination decisions.
What Records Must Be Kept
An organisation that uses AI in HR decisions must keep records that demonstrate the decision was lawful, fair, and non-discriminatory. These records include: documentation of the AI system (what it does, what data it uses, how it was trained and tested), records of decisions made by or influenced by the system, and records of how the organisation responded to individual concerns or disputes.
For each AI hiring decision, the organisation should keep: the candidate’s application materials, the score or recommendation produced by the AI system, any notes about how the decision was made, whether a human reviewed the automated decision, what the human’s conclusion was, and if the candidate requested review or provided additional information, what that was and how it was considered.
For each employee affected by an AI performance rating or monitoring system, the organisation should keep: the raw data analysed by the system, the score or assessment produced by the system, what the assessment was used for, whether the employee was informed of the assessment, whether the employee requested review or provided alternative information, and the final decision made about the employee.
For all AI systems used in HR decisions, the organisation should keep: the impact assessment (DPIA) conducted before deployment, results of bias testing (pre-deployment and ongoing), records of updates to the system, reports of any incidents or complaints, responses to complaints, and evidence of actions taken to remediate problems.
Retention Periods
Personal data should not be kept longer than necessary. However, HR records and data protection documentation must be kept long enough to defend decisions if challenged. In practice, this means keeping records for at least as long as employment records are kept (typically three to six years after employment ends) and as long as the statute of limitations for discrimination claims (typically three years from the date of the alleged discrimination, but extending to six years if the organisation has concealed the discrimination).
Documentation That Proves Compliance
Documentation of compliance is worth more than the absence of complaints. An employer who can produce evidence of pre-deployment bias testing, post-deployment monitoring, reasonable adjustments, transparency to affected individuals, and human review processes is in a far stronger position to defend a hiring or performance decision than an employer who can say only “the AI said so.”
For a hiring decision challenged on discrimination grounds, the organisation should be able to produce: the job specification the AI system was matched against, the bias testing results that show the system does not discriminate on the basis of protected characteristics, the candidate’s application materials and the score they received, a description of how the score was used in the hiring decision, evidence that the candidate was informed about the AI system and given an opportunity to request human review, and records of the hiring manager’s independent judgment applied to the candidate’s file.
For a performance decision challenged on discrimination grounds, the organisation should produce: the job criteria the performance assessment measures, evidence that the criteria are applied consistently across all employees, bias monitoring results showing the assessment system does not systematically disadvantage protected groups, the employee’s raw performance data and the assessment score, evidence that the employee was informed of the assessment and given access to their data, and records of any conversation or review that occurred in response to the employee’s concerns.
Privacy and Confidentiality
Record-keeping must balance transparency with privacy. An employee has the right to access their own data, but other employees’ data should be protected. Performance assessment systems should not compare employees publicly or share one employee’s data with other staff members.
When keeping records of AI decisions, the organisation should minimise the collection of sensitive personal data. If an AI hiring system is tested for bias regarding disability, the testing need not require collecting disability status on every candidate; it can be conducted on a subset of candidates with documented consent.
Litigation Readiness
From a practical perspective, an organisation that uses AI in HR should assume that decisions will be challenged and should keep records accordingly. This means keeping documentation of system design and testing, records of individual decisions, records of responses to disputes, and records of remedial action if problems are discovered.
A litigation-ready record set tells a coherent story: the organisation identified a business need, evaluated available solutions, selected an AI system, tested it for bias and fairness, deployed it with human oversight, monitored for problems, responded to complaints, and took action to remedy any issues discovered.
An organisation that cannot tell that story should not expect to win a discrimination case based on an AI hiring or performance decision.