A person makes every employment decision.
Employers and governments use Epitome's assessments to find patterns, estimate fit and decide where to act. Users can see how every output was built, and a person accepts, adjusts or rejects each recommendation.
On this page
Our positionThe EU AI ActUses we don't permitHigh-risk usesData we useData we never useExplainabilityHuman oversightFairness monitoringModel governanceShared responsibilityOur position
Epitome is a decision-support system. Hiring recommendations support recruiters and hiring managers. Development and skills-gap outputs inform planning. Authorised users can override or adjust any recommendation, and the platform records their reasons in an audit trail.
The EU AI Act
The EU AI Act takes a risk-based approach that matters especially in employment. It entered into force on 1 August 2024. Prohibited practices have applied since 2 February 2025. Obligations for high-risk employment systems apply on the Act's phased timetable, and we update client deployment documentation as the timetable changes.
The Act's scope in employment is broad. It covers candidates, contractors and trainees as well as employees, and a feature's classification depends on how a client uses it as well as on what it does technically.
Uses we don't permit
Clients may not use Epitome for:
- biometric emotion recognition of employees, candidates, contractors or other workers;
- webcam, video, voice or other biometric analysis to infer emotions or intentions at work;
- biometric categorisation to infer race, religion, political views, trade union membership, sex life or sexual orientation;
- social scoring in employment or workforce decisions;
- manipulative or deceptive techniques designed to distort behaviour or impair informed decisions.
High-risk uses
High-risk uses need tighter governance, fuller documentation and careful deployment controls. Under the Act, many employment uses of AI are likely to be high-risk. In Epitome, these include:
- ranking or filtering candidates for recruitment or selection;
- using skills or competency results in promotion, progression, reward or termination decisions;
- using workforce analytics to monitor or evaluate individual workers;
- using role-fit or matching results to decide who gets work opportunities or tasks.
High-risk does not mean prohibited. We configure these features with each client accordingly.
A scheduling assistant may fall outside the high-risk category when it performs a narrow procedural step and does not materially influence the decision. A ranking tool used in selection does not.
Data we use
- Information people give us directly, such as assessment responses, profile details and uploaded CVs.
- Data the client provides, such as employee records, role definitions and organisation structures.
- Public reference frameworks, including ESCO, O*NET and SkillsFuture.
- Market and benchmark data, anonymised and used only in aggregate.
Data we never use
- Social media data scraped about candidates or employees.
- Candidate data bought from brokers or aggregators.
- Browsing history or web activity as an assessment input.
- Personal characteristics inferred from IP address or location.
- Biometric data such as facial recognition or voice analysis.
- Protected characteristics inferred from names, photos or similar proxies.
Explainability
Users can see how each output was produced: the factors considered, how much each contributed, and the data or benchmark behind it. In a fit score, that means the competency evidence, psychometric fit, team contribution and growth trajectory components and the weight given to each. Where a result is less reliable, for example because responses were rushed or uniform, we mark it with a data-quality notice.
Human oversight
Recommendations are presented for a person to accept, adjust or reject, and the reason for any change is saved to the audit trail. Where a client needs specific escalation, approval or override steps, we build them into the product configuration and the client's process.
Fairness monitoring
Protected characteristics are excluded from decision factors. Where clients lawfully hold demographic data, we support adverse-impact monitoring, including four-fifths-rule analysis and statistical testing, along with a channel for raising concerns.
Model governance
We take each AI capability through design, development, validation, deployment and operation, and keep a model inventory recording its purpose, inputs, oversight requirements, known limitations and owner.
Shared responsibility
Epitome documents each feature's intended purpose, boundaries, explanations and controls, and supports audits. Clients remain responsible for how they deploy the platform: their lawful basis, notices to individuals, final employment decisions and any collection of demographic data for fairness analysis. Using a feature outside its documented purpose may change a client's obligations under the Act.