Automated decision-making at the workplace: the legal framework in Georgia
Enterprises increasingly automate decisions on candidate selection, employee evaluation, scheduling and pay. The Labour Code of Georgia knows no stand-alone institution for automated decision-making, but such decisions are fully governed by the Code's general rules. This page explains the three pillars that matter most: the allocation of the burden of proof in discrimination disputes, the consequences of violating the non-discrimination principle, and the special statutory guarantees with which a system-generated decision most often collides — the leave regime connected with pregnancy, childbirth, childcare and adoption.
The burden of proof in a discrimination dispute
When an algorithmic decision is legally challenged, the decisive rule concerns the burden of proof. In a dispute connected with the prohibition of discrimination, the burden rests on the employer if the candidate or the employee points to facts or circumstances that create a basis for a reasonable presumption that the employer violated the requirement of the non-discrimination ban. In practice this means that the weight of documenting and explaining the outcomes, criteria and operation of an automated assessment shifts to the employer once a prima facie case is indicated. It is precisely for this reason that the process of using algorithmic tools must be recorded and interpretable: the logic and the results of a system's decisions are the material with which the employer will defend its position — or fail to. Documentation discipline around automated decisions is therefore not a technicality but the core of litigation readiness.
Consequences of violating the non-discrimination principle
The Labour Code ties a violation of the non-discrimination principle directly to administrative liability. An employer's violation of the ban on discrimination — including direct and indirect discrimination, harassment and sexual harassment at the workplace — of the principle of reasonable accommodation, or of the rule on equal pay for equal work, results in a warning or a fine applied under the procedure established by paragraph 1 of Article 77 of the Code, at three times the amount of the respective fine. If the same act is committed repeatedly within 1 calendar year from the imposition of the administrative penalty for that violation, the fine doubles — it is imposed at twice the amount of the fine for the respective violation.
The regime for harassment cases deserves separate mention: imposing liability on the offending employee does not release the employer from the corresponding liability. The employer may itself be held liable if it became aware of a fact of harassment or sexual harassment and did not notify the labour inspection of that fact and did not take appropriate measures to prevent the conduct. In the context of automated systems this means that becoming aware of a fact flagged or recorded by a system — and remaining passive about it — can itself become a ground for liability. Silence after a system surfaces a problem is not a neutral option.
Leave guarantees that an automated decision must respect
Automated scheduling and evaluation systems collide most often with the special guarantees the Code protects during pregnancy, childbirth and childcare. At her own request, an employee is granted paid leave for pregnancy and childbirth of 126 calendar days, and in the case of a complicated delivery or the birth of twins — 143 calendar days. Leave for childcare amounts to 604 calendar days, or 587 calendar days in the case of a complicated delivery or twins; of that leave, 57 calendar days are paid. The childcare leave may be used, in full or in part, by the child's mother or father, and the employee must warn the employer 2 weeks in advance of taking it. The paid parts are used consecutively over 183 or 200 calendar days.
In the case of adoption of a newborn, an employee who has adopted a child under 1 year of age is granted, at his or her own request, leave amounting to 550 calendar days from the child's birth, of which 90 calendar days are paid. These numbers are not an informational courtesy: any automated rota, headcount-balancing or payroll module must be aligned with these minimum standards, otherwise the system's "decision" contradicts the law. Where an algorithm allocates shifts or scores availability, the statutory floor travels with the employee and cannot be optimised away.
Practical steps and our service
We explain to clients how automated decision processes can be positioned within the Labour Code framework: at which stages human involvement is required, how to defuse the burden-of-proof risk, and how a system can be configured to respect the leave guarantees. We assist in drafting internal rules, preparing disputes and producing legal assessments. An early legal review of an automated tool is substantially cheaper than defending an unexplainable decision after the fact.
Where the human decision remains and where the system only recommends
The Labour Code does not prescribe a level of automation, but the norms described above make clear that final responsibility always rests with the employer. In sound practice the output of an algorithmic system is treated as a recommendation, while the decision is taken by a human who can explain why that particular outcome was accepted and verify that it does not conflict with the requirements of the Labour Code. With the burden-of-proof rule in view, every automated decision should be backed by an interpretable record: which data were used, which rule led the system to its result, and who reviewed the result at the human level. This approach directly serves the defence of the employer position in a dispute and at the same time reduces the risk that a system output contradicts the principle of the non-discrimination ban. We help put this process into documentary form so that it stays aligned with the Code requirements as the technology and the team evolve.