Using AI scores to choose who is made redundant in Great Britain

When an employer automates work or cuts roles, it may use software to score staff for selection: performance ratings, productivity figures, skills profiles or attendance data. This page sets out, for workers and employers in England, Wales and Scotland, the statutory rules that apply to that kind of selection, as the text stood on 31 October 2026. It covers what counts as redundancy, how selection is tested for fairness, and when collective consultation is required. For the rules in the EU and the United States, see AI and layoffs: consultation and notice rules.

The short answer: replacing work with AI can be a genuine redundancy, and redundancy is a potentially fair reason for dismissal. The employer still has to act reasonably, and the method used to pick who goes is a central part of that. Where 20 or more employees at one establishment may be dismissed as redundant within 90 days, the employer must consult representatives and disclose in writing the proposed method of selection, which includes any scoring tool.

Is it a redundancy?

An employee is dismissed by reason of redundancy if the dismissal is wholly or mainly attributable to the employer ceasing the business, or ceasing it at the employee's place of work, or to the fact that the requirements of the business for employees to carry out work of a particular kind, or to do it at that place, have ceased or diminished or are expected to (Employment Rights Act 1996, section 139(1)). "Cease" and "diminish" mean permanently or temporarily and for whatever reason (section 139(6)). If an AI system now does work that people used to do, the need for employees to do that kind of work may have diminished, which fits the definition. A label is not enough, though: if the real reason is something else, such as performance, the rules for that reason apply instead; see unfair dismissal when AI data is behind the decision.

Fair reason, fair process, fair selection

Redundancy is one of the potentially fair reasons for dismissal (section 98(2)(c)). Whether the dismissal is fair then depends on whether, in the circumstances, including the employer's size and administrative resources, the employer acted reasonably in treating it as a sufficient reason, decided in accordance with equity and the substantial merits of the case (section 98(4)). Employees generally need 2 years of continuous employment to claim (section 108); this falls to 6 months where the effective date of termination is on or after 1 January 2027 (SI 2026/559).

When software scores are part of selection, the questions that bear on reasonableness include:

• Is the pool right? Who was considered for redundancy, and why those people.
• Are the criteria objective and checkable? A score the employer cannot explain or verify is hard to defend. Know what each input measures and over what period.
• Is the data fair to everyone? Figures can be skewed by absence, part time hours, parental leave or disability. Selection because of a protected characteristic, or using criteria that put a group at a disadvantage without justification, can also be discrimination; see AI and the Equality Act 2010.
• Did each person see and challenge their scores? Individual consultation gives people the chance to correct errors before the decision.
• Was alternative work considered? Including roles created by the new system, and changed hours or home working; see flexible working requests.
• Is the work moving to another business? If a provider takes the activities over, TUPE may apply instead; see TUPE when work moves to an AI provider.

Some reasons for selecting a person for redundancy are automatically unfair, for example where the real reason for selecting them was a protected disclosure (section 105). See raising concerns about AI at work.

Collective consultation: 20 or more at one establishment

• When it applies: where an employer proposes to dismiss as redundant 20 or more employees at one establishment within a period of 90 days or less, it must consult the appropriate representatives of the employees who may be affected (Trade Union and Labour Relations (Consolidation) Act 1992, section 188). The Employment Rights Act 2025 adds a second test based on a "threshold number" across the employer to be set by regulations; on 31 October 2026 that change was in force only for the purpose of making those regulations (S.I. 2026/3, regulation 2(14)).
• When it starts: in good time, and at least 45 days before the first dismissal takes effect where 100 or more are proposed, otherwise at least 30 days (section 188(1A)).
• What it covers: ways of avoiding the dismissals, reducing the numbers and mitigating the consequences, with a view to reaching agreement (section 188(2)).
• What must be disclosed in writing: among other things the reasons for the proposals, the numbers and descriptions of employees affected, and the proposed method of selecting the employees who may be dismissed (section 188(4)). Where software scoring is part of that method, describing it is part of the disclosure.
• Who represents staff: a recognised independent trade union, or otherwise employee representatives, either existing ones with authority to be consulted or ones elected for the purpose (section 188(1B)).

If the employer does not consult

A complaint can be made to an employment tribunal, which may make a protective award: an order to pay remuneration for a protected period of up to 180 days, set according to the seriousness of the employer's default (section 189). The maximum was raised from 90 to 180 days from 6 April 2026, subject to transitional provisions (S.I. 2026/323). The complaint can be made before the last of the dismissals takes effect, or generally within 6 months beginning with that date where it is on or after 1 October 2026; earlier cases keep 3 months (section 189(5)).

Practical steps for employers

• Write down why the roles are no longer needed, separately from how people will be chosen.
• Use selection criteria a person can explain and check. If a tool produces scores, test them for errors and for effects on different groups before you rely on them.
• Count the proposed dismissals per establishment early, and start collective consultation in time if 20 or more are proposed within 90 days.
• Show each employee their scores and the data behind them, and record their response and any corrections.
• Look for alternative roles, including new work around the AI system. See also consulting employees before an AI rollout.

For employees

Ask how the pool was drawn, what the selection criteria are, whether any software produced or informed your scores, and what data it used. Ask to see your own scores and point out anything wrong or missing, such as time on leave. Ask about other roles (if you are pregnant, on maternity, adoption or shared parental leave, or recently back, see the priority right to a suitable vacancy) and, once under notice, about time off to look for work, and see statutory redundancy pay when your role goes to automation. Once dismissed, you can ask for written reasons for the dismissal. If you are in a union, it may be the representative the employer must consult. This page is general information, not legal advice.