Predictive layoffs refer to using data analytics, AI, machine-learning models, business forecasts, productivity scores, attendance data, performance indicators, or other automated tools to predict that certain employees may become “surplus” and should therefore be selected for layoff/retrenchment.

The important legal point is that the use of a predictive model does not by itself make a layoff or retrenchment lawful. The employer must still comply with the applicable labour law, prescribed procedure, compensation requirements, and rules governing selection of workers.

As of 2026, India's principal framework is the Industrial Relations Code, 2020 (IRC). Section 70 requires, for covered workers with at least one year of continuous service, one month's notice/relevant wages in lieu, retrenchment compensation, and prescribed notice to the appropriate Government/authority. Section 71 ordinarily follows the “last come, first go” principle unless there is an agreement to the contrary or reasons are recorded for departing from it.

1. Meaning of predictive layoffs

A predictive-layoff system might analyse:

• declining business revenue;
• projected workload;
• employee productivity;
• performance ratings;
• absenteeism;
• skills and qualifications;
• automation exposure;
• departmental costs;
• predicted future demand; or
• other HR and business data.

For example, an employer's AI system might predict that a particular department will require 30% fewer employees next year and generate a list of employees considered “redundant.”

The prediction is only an input into the employer's decision. It should not automatically replace legally required human decision-making and statutory procedures.

2. Section 70 – statutory conditions

Under Section 70 of the Industrial Relations Code, a worker with at least one year of continuous service cannot ordinarily be retrenched unless the statutory requirements are satisfied.

These include:

1. One month's written notice stating the reasons, or wages in lieu of notice;
2. Retrenchment compensation of the prescribed amount—generally 15 days' average pay for every completed year of continuous service or part exceeding six months; and
3. Notice to the appropriate Government/authority in the prescribed manner. 

Therefore, an employer cannot simply say:

“Our algorithm predicted that your position is no longer required, so your employment ends today.”

The predictive result does not replace the statutory requirements.

3. Section 71 – selection of employees

This is particularly important for predictive layoffs.

Section 71 provides that, where workers belong to a particular category and there is no contrary agreement, the employer should ordinarily retrench the last person employed in that category, unless the employer records reasons for selecting someone else.

Consequently, an AI-generated ranking such as:

Employee A – 92% redundancy risk
Employee B – 87% redundancy risk
Employee C – 81% redundancy risk

does not automatically establish that Employee A can lawfully be selected.

The employer should be able to explain the legally relevant basis for departing from the statutory selection principle.

4. Algorithmic discrimination

A predictive model can create legal problems if the underlying data or variables indirectly discriminate against protected groups.

For example, a model might heavily rely upon:

• historical promotion records;
• previous performance ratings;
• salary levels;
• career interruptions;
• attendance;
• location;
• age-related variables; or
• other correlated characteristics.

If the model systematically disadvantages a particular category of workers, the employer may face equality, discrimination, employment-law, or privacy-related challenges depending upon the applicable law and circumstances.

The employer therefore needs to examine both the output and the methodology used to generate the output.

5. Human review and natural justice

Indian Supreme Court decisions concerning termination emphasise that employment termination can have serious civil consequences and that arbitrary procedures may be legally problematic.

This becomes especially relevant when an employee is terminated because of an opaque automated assessment.

An employee should, where applicable, have an opportunity to challenge factual errors—for example:

• incorrect performance data;
• wrongly attributed absenteeism;
• outdated qualifications;
• inaccurate productivity measurements;
• incorrect employment history; or
• errors in the data used by the predictive system.

6. Important case laws

1. State Bank of India v. N. Sundara Money, (1976) 1 SCC 822

The Supreme Court gave the expression “retrenchment” a broad interpretation under the Industrial Disputes Act. Termination for a reason other than the statutory exclusions could fall within retrenchment.

The case emphasised that statutory retrenchment protections cannot be avoided merely by describing termination differently.

Relevance: An employer cannot necessarily avoid retrenchment protections by describing an AI-generated termination as a “business decision,” “role optimisation,” or “predictive workforce adjustment.”

2. Punjab Land Development and Reclamation Corporation Ltd. v. Presiding Officer, Labour Court, (1990) 3 SCC 682

A Constitution Bench confirmed the broad interpretation of “retrenchment,” holding that termination by the employer for any reason, subject to statutory exclusions, can constitute retrenchment.

Relevance: If predictive analytics ultimately causes termination of a worker covered by retrenchment provisions, the employer must examine whether the termination falls within the statutory concept of retrenchment rather than relying upon the technological terminology used.

3. D.K. Yadav v. J.M.A. Industries Ltd., (1993) 3 SCC 259

The Supreme Court held that termination under a standing-order mechanism could not simply operate automatically without appropriate procedural safeguards. The Court connected termination affecting livelihood with fairness, natural justice and Articles 14 and 21.

Relevance: An automated or predictive system should not be treated as an unquestionable decision-maker. Where applicable, the affected employee must have appropriate procedural safeguards.

4. Uptron India Ltd. v. Shammi Bhan, (1998) 6 SCC 538

The Supreme Court considered an “automatic termination” provision and held that management discretion could not be exercised capriciously. The Court emphasised objective consideration of the relevant circumstances and an opportunity of hearing.

Relevance: A predictive system that automatically places an employee on a termination list presents a similar concern: automation should not convert a managerial prediction into an automatic termination.

5. Regional Manager, SBI v. Rakesh Kumar Tewari, (2006) 1 SCC 530

The Supreme Court examined Section 25G and explained the conditions surrounding the “last come, first go” principle. The Court recognised that the category of workers, contrary agreement and reasons for departing from the normal rule are relevant factual matters.

Relevance: A predictive model's ranking should not casually replace the statutory selection framework. If an employer chooses a different worker from the ordinarily applicable sequence, the reasons should be legally defensible and properly recorded.

6. Delhi Transport Corporation v. D.T.C. Mazdoor Congress, 1991 Supp (1) SCC 600

The Constitution Bench considered a rule permitting termination of permanent employees without adequate safeguards. The majority treated an unfettered termination power as inconsistent with Article 14 principles applicable to the statutory corporation.

Relevance: The case is significant for predictive HR systems because a termination mechanism should not confer uncontrolled or arbitrary discretion merely because an automated system has produced a recommendation.

7. Pramod Jha v. State of Bihar, (2003) 4 SCC 619

The Supreme Court explained the purpose of retrenchment safeguards, including providing the employee with time to seek alternative employment and ensuring payment of retrenchment compensation at the appropriate stage.

Relevance: Predictive identification of future redundancy does not eliminate the employee-protection objectives underlying retrenchment legislation.

7. Special issue: large industrial establishments

The IRC contains an additional regime for establishments to which Chapter X applies. Section 79 provides additional conditions for retrenchment, including three months' notice/pay in lieu and prior Government permission for covered establishments.

Thus, before implementing predictive layoffs, an employer must determine:

• whether the establishment falls within the relevant statutory category;
• whether the workers satisfy the statutory definition;
• whether the required workforce threshold applies;
• whether Government permission is required;
• whether notice requirements have been satisfied; and
• whether compensation and other statutory obligations have been fulfilled.

8. Data and privacy considerations

Predictive layoffs may involve extensive employee data. If an employer collects information about employees and feeds it into an AI/analytics system, it should also consider applicable privacy and data-protection requirements.

The Supreme Court's privacy jurisprudence recognises privacy as a constitutionally protected right, although the precise constitutional application differs depending upon whether the employer is a State/public authority or a private employer.

For a private employer, statutory data-protection and employment-law obligations may therefore be particularly important, while constitutional challenges may arise differently when State action or a State instrumentality is involved.

9. When can predictive layoffs become legally problematic?

A predictive-layoff system becomes particularly vulnerable where:

• the algorithm automatically terminates employees;
• employees are not given required statutory notice;
• retrenchment compensation is not paid;
• the employer ignores the applicable statutory selection rule;
• the model contains materially inaccurate employee data;
• the employer cannot explain the selection decision;
• similarly situated workers are treated inconsistently;
• the system produces discriminatory effects;
• the employer uses irrelevant personal characteristics;
• the employer uses predictive scores as a substitute for legally required reasons; or
• required Government permission/procedure is bypassed.

10. Practical compliance framework

An employer using predictive analytics for workforce reduction should ideally follow this sequence:

Business forecast → Human/legal review → Validate employee data → Identify applicable labour-law regime → Determine legally permissible selection criteria → Document reasons → Individual/statutory notice → Compensation → Government notification/permission where applicable → Grievance opportunity → Final termination → Re-employment/re-skilling obligations where applicable.

The 2026 framework also contains a re-skilling fund mechanism for retrenched workers, in addition to retrenchment compensation.

Conclusion

Predictive layoffs are not inherently unlawful merely because predictive analytics or AI is used. The legal issue is how the prediction is converted into an employment decision.

Under India's current labour framework, an employer must still comply with the applicable retrenchment provisions, including notice, compensation, selection requirements, Government procedures/permissions where applicable, and other statutory protections. The Supreme Court's jurisprudence further indicates that employment termination cannot be insulated from legal scrutiny simply because the employer relies on an automated, contractual, or managerial mechanism.

Accordingly, a predictive score should ordinarily be treated as decision-support information, not as an automatic legal justification for termination.