Managing AI-Related Workforce Change

by Claude Opus 5.5

How should UK employers handle and communicate AI-related workforce changes (redeployment, hiring freezes, redundancies), and what consultation and legal obligations apply?

There is no AI-specific employment statute in the UK. AI-driven restructuring is governed by the ordinary law of redundancy, consultation, unfair dismissal and discrimination, and several of those rules have recently become more expensive to get wrong. Since 6 April 2026, the maximum protective award for failing to consult collectively has doubled to 180 days’ pay per affected employee. From January 2027, the qualifying period for unfair dismissal falls to six months and the compensation cap is removed. Employers who lead with redeployment, consult early and genuinely, keep selection explainable, and are honest about AI’s actual role will face fewer legal and reputational problems. What follows is general information, not legal advice; take advice on any specific restructuring.

Start with the least disruptive levers

Acas lists the standard alternatives to compulsory redundancy: voluntary redundancy, changes to hours, moving people into other roles, releasing temporary or contract workers, limiting overtime and stopping recruitment. For AI-driven change, the order usually runs from natural attrition and a hiring freeze, through redeployment and retraining, to voluntary and finally compulsory redundancy.

Hiring freezes carry no statutory consultation duty in themselves. But if you have an information and consultation agreement, or the statutory standard provisions apply (see below), you must share information on the “probable development of employment” and on “anticipatory measures” where jobs are threatened. Freezing entry-level hiring because AI is expected to absorb the work is exactly such a measure. Collective agreements may also require consultation.

Redeployment has useful statutory machinery:

An employee moved to a different role during a redundancy has a four-week trial period under section 138 of the Employment Rights Act 1996. It can be extended by written agreement “for the purpose of retraining the employee”, which fits AI reskilling well.
Under section 141, an employee who unreasonably refuses an offer of suitable alternative employment can lose their statutory redundancy pay.
Acas notes that pregnant employees and new parents must be offered suitable alternative vacancies as a priority.

Changing terms instead of dismissing, for example cutting hours because AI has reduced workload, needs care. From January 2027, under section 28 of the Employment Rights Act 2025, dismissing someone for refusing a “restricted variation” will be automatically unfair. Restricted variations cover pay, hours, shift patterns, leave and similar terms. There is a narrow exception where the change was needed to address financial difficulties threatening the business’s viability and could not reasonably have been avoided. “We have adopted AI and need fewer hours” is unlikely to meet that test on its own.

Collective consultation: the rules that bite

Trigger. The rule: Proposing to dismiss as redundant 20 or more employees at one establishment within 90 days or less. Source: TULRCA 1992 s.188(1).
Minimum period. The rule: 30 days before the first dismissal (20–99); 45 days (100 or more); consultation must begin “in good time”. Source: s.188(1A).
With whom. The rule: Recognised trade union representatives; otherwise elected employee representatives, whose election the employer must arrange if none exist. Source: s.188(1B); Acas.
Content. The rule: Ways of avoiding dismissals, reducing numbers and mitigating consequences, with a view to reaching agreement. Source: s.188(2).
Written disclosure. The rule: Reasons for the proposals, numbers and descriptions of employees affected, selection method, how dismissals will be carried out, redundancy pay calculation, agency worker information. Source: s.188(4).
Notifying government. The rule: Form HR1 to the Redundancy Payments Service at least 30 days (20–99) or 45 days (100 or more) before the first dismissal; failure can bring an unlimited fine. Source: gov.uk.
Penalty for not consulting. The rule: Protective award of up to 180 days’ pay per affected employee (90 days before 6 April 2026). Source: s.189, as amended by ERA 2025.

Three points matter particularly for AI programmes.

Redundancy is defined broadly for consultation purposes. It covers dismissals for any reason not related to the individual (s.195). Replacing contracts as part of an AI-driven reorganisation can count, even if nobody leaves.

The cost of not consulting has risen sharply. As an illustration, take 60 roles at an average salary of £38,000, or about £104 a day. A maximum award would be about £18,700 per employee, or roughly £1.1m in total, double what it was before April. Tribunals set the award at what is “just and equitable” given how serious the employer’s default was. Since Susie Radin v GMB (2004), they have treated a complete failure to consult as a reason to start from the maximum.

Timing is the trap. The duty arises when dismissals are proposed, and consultation must take place while proposals can still change. A board approving an “AI transformation” with a fixed headcount reduction, announced to investors before staff representatives have seen it, invites the argument that consultation was a formality.

A change is coming. The Employment Rights Act 2025 inserts an alternative trigger based on a “threshold number” of employees across the employer, to be set by regulations (new s.188(A1) and s.195A). On legislation.gov.uk those provisions are so far in force only “for specified purposes”, meaning regulation-making, and no threshold has been set. The government’s roadmap places this change in 2027. Until then, the 20-at-one-establishment test governs.

Individual consultation and fair selection

Below 20 redundancies there is no collective duty, but fair process still applies. Acas advises meeting each affected employee privately at least once and says consultation “must be meaningful”. A fair redundancy generally requires:

a genuine redundancy situation, meaning the employer’s need for employees to do work of a particular kind has ceased or diminished (AI absorbing the work can satisfy this, but the reduced need must be real);
a sensible selection pool;
objective criteria applied consistently;
consultation and a real search for alternative employment;
an appeal.

Acas says criteria should be “objective and measurable” and “not affected by personal opinions”.

Three dates change the stakes:

Since 1 October 2026, the time limit for most tribunal claims has been six months rather than three.
For dismissals from 1 January 2027, the qualifying period for ordinary unfair dismissal falls from two years to six months, and the cap on the compensatory award is removed (ERA 2025 s.25; gov.uk timeline, updated 25 September 2026). A restructuring planned for early 2027 will therefore expose the employer to claims from far more employees, with no cap on compensation.
Statutory redundancy pay still requires two years’ continuous service.

Equality Act risks when AI informs selection

AI increasingly appears in the selection process itself: productivity dashboards, AI-scored skills assessments, “digital readiness” ratings. The employer is liable for discriminatory outcomes whoever built the tool. The main exposures are these:

Indirect discrimination (Equality Act 2010, s.19). A criterion that disadvantages a protected group must be objectively justified. Output-per-hour metrics can disadvantage part-time staff, who are disproportionately women. “AI adoption” or “digital fluency” scores can disadvantage older workers. Acas gives a parallel example of how a flexible-working criterion can indirectly discriminate against women.
Disability (s.15 and the duty to make reasonable adjustments under s.20). Metrics drawn from periods of disability-related absence or reduced capacity need adjusting. Acas says attendance criteria should exclude disability- and pregnancy-related absence.
Pregnancy and maternity. Employees on maternity leave can’t be scored down for their absence, and they have priority for alternative vacancies.
Explainability as a defence issue. If the employer cannot explain how a score was produced, it will struggle to show that the criterion was objective, consistent and non-discriminatory.

Data protection law overlaps here:

If selection is effectively decided by an automated system with no meaningful human involvement, the Data (Use and Access) Act’s safeguards apply (in force since 5 February 2026). Employees must be told, must be able to contest the decision, and must be able to obtain human intervention. Special-category data such as health information is tightly restricted.
Employees can use subject access requests to obtain the data and scores used about them, which often happens before a tribunal claim.

There is still no reported UK tribunal ruling on AI discrimination at work. The best-known case, Manjang v Uber Eats, an EHRC-backed race discrimination claim over facial-recognition checks, was reported settled in 2024. Absence of case law is not absence of risk.

Practical safeguards:

Have a human decision-maker who understands and can explain each score.
Run an equality impact check on the pool and the provisional outcomes before confirming them.
Keep records of how each criterion was applied.

Information and consultation arrangements

The Information and Consultation of Employees Regulations 2004 apply to undertakings with 50 or more employees. Since April 2020, employees can trigger negotiations with a valid request from 2% of the workforce, subject to a minimum of 15 employees.

Where no agreement is reached and the standard provisions apply, regulation 20 requires the employer to inform and consult on the situation and probable development of employment, including anticipatory measures where jobs are threatened. It also requires consultation “with a view to reaching agreement” on “decisions likely to lead to substantial changes in work organisation”. A significant AI deployment that redesigns how a department works plausibly falls within that, before any redundancy is proposed.

Existing ICE agreements, works councils (including European works councils in multinational groups) and union recognition agreements should be checked for their own information triggers. New union rights of access to workplaces under the 2025 Act are due from late October 2026. The TUC has campaigned for mandatory AI impact assessments and consultation, but no AI-specific consultation duty exists in UK law.

Communicating the change

Sequence carefully. Representatives and affected staff should hear proposals before or at the same time as the market and the press, and the language should say “proposals”, not decisions. Listed companies must also meet their disclosure obligations, so employment and investor-relations teams need a single timetable.

Be specific. Say which roles are affected, what the technology actually does, the timeline, the selection approach and the support on offer: redeployment pool, retraining, trial periods and time off to look for work. Vague talk of “transformation” makes people assume the worst.

Don’t neglect those who stay. People who remain will absorb new tools and often new workloads. Explain what changes for them, and avoid pairing the rollout with intrusive productivity monitoring.

Avoid AI-washing in either direction

AI-washing means overstating AI’s role in cuts, to look innovative to investors or to make cuts seem inevitable, or understating it to avoid backlash. Both happen:

The Yale Budget Lab found only 4.5% of 2025 US job cuts were attributed to AI.
Standard Chartered’s chief executive explicitly framed planned back-office cuts (reported at about 7,800 roles, not necessarily in the UK) as replacing “lower-value human capital” with financial capital.
Centrica said “AI isn’t driving” its 1,300 British Gas call-centre cuts, citing a 20% fall in call volumes, while unions disputed that account.

The legal point is that the reasons disclosed under s.188(4) and in individual consultation must be the real ones. If the business case says “cost reduction”, the press release says “AI” and the selection pool is built around “AI-exposed roles” the employer cannot justify, every inconsistency becomes evidence in consultation and at tribunal. Staff often know more than leaders assume: Deloitte found 31% of UK generative AI users already use it without their employer knowing. Say what AI is doing, what else is driving the decision, and what is still uncertain.

Bottom line

Treat AI restructuring as a normal change programme carried out under tighter rules. Lead with redeployment and retraining, using the extended trial period where it helps. Consult representatives while proposals can still change. Keep selection objective, explainable and equality-checked, and tell people the true reasons. The cost of skipping these steps doubled in April 2026, and from January 2027 far more employees will be able to bring uncapped unfair dismissal claims.

Sources

From AI and Jobs: UK, October 2026