Civil Law and UAE Algorithmic Damage Calculation Methodologies

1. Introduction

Algorithmic damage calculation refers to the use of mathematical models, artificial intelligence, statistical methods, machine learning, automated valuation systems, or other computational tools to estimate the amount of compensation payable for civil harm.

In the UAE, algorithmic calculation can potentially be used in disputes involving:

• personal injury;
• loss of income;
• business interruption;
• contractual losses;
• construction delays;
• financial losses;
• intellectual-property infringement;
• reputational damage;
• data and privacy harm;
• algorithmic discrimination;
• cyber incidents;
• AI-related losses;
• insurance claims; and
• complex commercial disputes.

The important legal principle is that an algorithm calculates or estimates damages; it does not itself determine the legally recoverable amount. The court or arbitral tribunal remains responsible for determining whether damage exists, whether it was caused by the defendant, and what compensation is legally justified.

Because the UAE's new Civil Transactions Law took effect on 1 June 2026, older cases decided under the 1985 Civil Transactions Law should be understood as important historical and doctrinal authorities, rather than automatically treated as decisions under the new Code.

2. Meaning of Algorithmic Damage Calculation

Traditional damage assessment can be represented as:

Wrongful act → Damage → Evidence → Judicial assessment → Compensation

Algorithmic damage assessment adds another stage:

Wrongful act → Damage → Data → Algorithm/model → Estimated loss → Judicial assessment → Compensation

For example, an AI model might calculate:

• expected future income;
• lost customers;
• business interruption;
• diminution in property value;
• lost market share;
• probability-adjusted future profits;
• medical or rehabilitation costs;
• financial loss attributable to a cyberattack.

The model can assist the court, but legal entitlement and mathematical estimation remain separate questions.

3. Fundamental Principle: Compensation Must Correspond to Legally Recognized Damage

The central civil-law objective is generally to place the injured party, so far as legally possible, in the position they would have occupied absent the wrongful conduct.

However, compensation is not simply:

Whatever amount an algorithm produces.

The court must first establish:

1. a legally recognizable injury;
2. a wrongful act or legally compensable breach;
3. causation;
4. the extent of actual damage;
5. appropriate evidence;
6. the legally recoverable category of loss.

Only then can an algorithm assist in quantification.

4. UAE Civil-Law Framework

The UAE's previous Civil Transactions Law contained the foundational civil-liability provisions, including Article 282 concerning liability for causing harm.

The new Federal Decree-Law No. 25 of 2025, effective 1 June 2026, modernizes the civil-law framework while retaining core concepts relevant to damages, including:

• unlawful harm;
• causation;
• contractual responsibility;
• good faith;
• abuse of rights;
• judicial assessment of loss.

Accordingly, algorithmic calculation should be treated as a method of proof and valuation, rather than as a new autonomous source of liability.

5. Main Types of Algorithmic Damage Calculation

A. Historical-loss model

The system calculates losses that have already occurred.

Example:

A cyberattack prevented a business from operating for 20 days.

The algorithm may analyse:

• historical daily revenue;
• seasonal variations;
• normal expenses;
• cancelled orders;
• actual recovery.

B. Forecasting model

The system estimates what would probably have happened without the wrongful conduct.

Example:

A company claims that a breach caused it to lose future customers.

The model may compare:

Actual performance

against

Expected performance without the wrongful event.

This is commonly called a counterfactual analysis.

6. Counterfactual Damage Calculation

Counterfactual modelling is particularly important.

The basic formula can be expressed as:

Loss = Expected position without the wrong − Actual position after the wrong

For example:

• expected profit: AED 10 million;
• actual profit: AED 7 million.

Potential loss:

AED 3 million

But the court must still determine whether the entire AED 3 million difference was caused by the defendant.

Other causes could include:

• market decline;
• inflation;
• new competitors;
• management failure;
• supply-chain disruption;
• unrelated economic events.

Therefore:

Statistical difference ≠ legally compensable damage.

7. Causation Before Calculation

This is one of the most important principles.

An algorithm may accurately calculate a financial difference but still fail to establish legal causation.

Suppose:

• Company expected revenue = AED 50 million.
• Actual revenue = AED 40 million.
• Algorithm calculates AED 10 million loss.

The defendant may respond:

“The AED 10 million decline resulted from an economic recession rather than our breach.”

The court therefore needs two separate inquiries:

Question 1

Did the defendant cause compensable harm?

Question 2

If yes, how much was caused by the defendant?

Algorithmic modelling primarily assists with Question 2.

8. Loss of Profits

Loss-of-profit claims are especially suitable for algorithmic analysis.

The model may examine:

• historical sales;
• market growth;
• customer retention;
• pricing;
• seasonal trends;
• competitor performance;
• economic conditions;
• projected demand.

A possible model is:

Expected profit − actual profit = preliminary loss

But the court may discount the amount where the projection is highly speculative.

The claimant must establish a sufficiently reliable basis for the projected profit.

9. Loss of Chance

Algorithmic methods may also be used where the claimant lost a commercial or economic opportunity.

For example:

A company claims that wrongful conduct prevented it from winning a major contract.

The algorithm could estimate the probability that the contract would have been obtained.

Suppose:

• potential contract value = AED 20 million;
• estimated probability of success = 40%.

The mathematical expected value is:

AED 20 million × 40% = AED 8 million.

But this does not automatically mean the claimant is entitled to AED 8 million.

The court must determine whether UAE law recognizes the relevant loss and whether the probability is sufficiently established.

10. Business Interruption

AI and algorithms can be particularly useful for business-interruption damages.

A model can consider:

• historical revenue;
• average daily sales;
• seasonal adjustments;
• fixed costs;
• variable costs;
• customer demand;
• industry benchmarks;
• duration of interruption.

For example:

Expected net profit during interruption − actual net profit = estimated interruption loss.

Again, the calculation must be supported by reliable evidence.

11. Construction Delay

Construction disputes are especially suitable for algorithmic analysis.

An algorithm may analyse:

• project schedules;
• critical path;
• progress reports;
• weather;
• labour records;
• material deliveries;
• variation orders;
• concurrent delays.

It may determine:

Which delay caused how much additional time and cost?

This is important because UAE construction disputes frequently involve concurrent causes.

An algorithm should therefore not simply assign the entire delay to the defendant.

12. Concurrent Causes

Suppose:

• contractor delay = 30 days;
• employer variation = 20 days;
• extraordinary weather = 10 days.

An automated system may identify all three factors.

The legal question becomes:

Which losses are attributable to the defendant?

This requires legal and factual analysis in addition to mathematical modelling.

13. Personal-Injury Damages

Algorithmic systems can estimate:

• medical expenses;
• rehabilitation costs;
• future treatment;
• lost earning capacity;
• future income;
• life expectancy;
• inflation;
• discount rates.

For example:

Annual earning loss × expected working years

can produce an initial estimate.

But the court must account for uncertainties such as:

• career progression;
• unemployment risk;
• retirement;
• disability;
• future medical development;
• actual earning capacity.

14. Present-Value Calculations

Future damages often need to be converted into present value.

A simplified formula is:

PV = Future Loss ÷ (1 + r)ⁿ

where:

• PV = present value;
• r = discount rate;
• n = number of years.

Algorithms can perform complicated discounted-cash-flow calculations.

However, the selection of the discount rate is a legal/economic judgment, not merely a programming decision.

A model can calculate the consequences of a 3%, 5% or 7% rate, but the court must decide which assumption is appropriate.

15. Inflation

Long-term damage calculations may require inflation assumptions.

An algorithm may calculate:

Future cost = Current cost × (1 + inflation rate)ⁿ

For example, if future medical expenses are expected to rise, the model can project those costs.

But again, the court must determine whether the inflation assumption is sufficiently supported.

16. Discounting and Double Recovery

Algorithmic models must avoid double counting.

For example:

• lost profit already includes lost sales;
• a separate “lost customer” calculation may include the same loss.

If both are awarded independently, the claimant could receive compensation twice for the same damage.

Therefore:

Algorithmic sophistication does not eliminate the prohibition against double recovery.

17. Algorithmic Calculation of Reputational Damage

Reputation-related harm is particularly difficult to quantify.

AI may analyse:

• media exposure;
• social-media reach;
• customer sentiment;
• brand-value changes;
• sales before and after publication;
• customer churn;
• search visibility.

But correlation does not necessarily prove causation.

For example:

Brand value fell by AED 5 million after a defamatory publication.

That does not automatically prove that the publication caused the entire AED 5 million reduction.

Other factors may have contributed.

18. Data and Privacy Damage

Algorithmic models may also estimate economic consequences of data misuse.

Possible components include:

• remediation costs;
• notification costs;
• system restoration;
• lost contracts;
• customer churn;
• regulatory expenses where legally recoverable;
• business interruption.

The claimant must distinguish actual legally recoverable damage from purely theoretical consequences.

19. AI-Related Damage

Suppose an AI system makes a harmful decision causing:

• financial loss;
• contractual loss;
• property damage;
• reputational injury.

An algorithmic damage model could determine:

Actual outcome − expected lawful outcome = preliminary economic difference.

But the court must still determine:

• whether the AI operator is legally responsible;
• whether the damage was foreseeable;
• whether the AI output caused the loss;
• whether another actor contributed;
• whether the loss is too remote;
• whether the claimant mitigated the damage.

20. Case Law

Because UAE reported jurisprudence specifically concerning AI-generated damage calculations is still developing, the following authorities are principally foundational analogies dealing with civil liability, causation, expert evidence, contractual damages, and judicial evaluation of evidence.

Case 1 — UAE Federal Supreme Court, Civil Appeal No. 79/2020

Principle

The Federal Supreme Court addressed the legal significance of admission and the circumstances in which an admission can establish a right.

Relevance to Algorithmic Damage Calculation

A computer-generated financial figure should not automatically be treated as an admission by a defendant.

For example, an AI-generated calculation appearing in a defendant's accounting system does not necessarily prove that the defendant accepts the claimant's damage figure.

The claimant may need to establish:

• who generated the calculation;
• whether it was authorized;
• whether the underlying data was accurate;
• whether it was intended to represent the defendant's position.

Relevance: Evidentiary and attribution analogy.

21. Case 2 — UAE Federal Supreme Court, Commercial Appeal No. 215/2020

Principle

The Federal Supreme Court emphasized the importance of properly reasoned expert evidence.

A court cannot simply adopt an expert conclusion without adequately examining its reasoning and addressing relevant arguments.

Relevance

This is highly important for algorithmic damages.

Suppose an expert presents:

“Our AI model calculates the claimant's loss at AED 15 million.”

The court should examine:

• model design;
• input data;
• assumptions;
• methodology;
• error margins;
• alternative explanations;
• causation.

An AI calculation should therefore be treated as expert evidence requiring judicial evaluation, not as an automatic answer.

22. Case 3 — Dubai Court of Cassation, Case No. 266/2008

Principle

The case concerned construction delay and concurrent causes.

Relevance

Algorithmic damage calculation frequently involves separating multiple causes.

For example:

Defendant's breach + claimant's delay + market conditions = total economic loss.

An algorithm can identify correlations and allocate time or financial effects, but the court must determine the legally attributable portion.

This makes the case particularly relevant to algorithmic causation and apportionment.

23. Case 4 — Dubai Court of Cassation, Case No. 1/2006

Principle

This case also involved concurrent delay and allocation of responsibility among contributing causes.

Relevance

It demonstrates why a damages model should not simply compare:

Before event vs after event.

A proper model should ask:

What would have happened if the defendant's wrongful conduct had not occurred, while all other independent factors remained?

This is essentially a legal version of counterfactual modelling.

Relevance: Strong causation analogy.

24. Case 5 — Abu Dhabi Court of Cassation, Case No. 55/2016

Principle

The case concerns the abuse-of-rights doctrine under the former Article 106.

Relevance

Damage calculations can become problematic where a party uses a lawful power abusively.

For example:

A contractual or technological power is deliberately exercised to cause disproportionate financial harm.

An algorithm could calculate the resulting economic damage, but the court must first determine whether the underlying conduct was legally abusive.

Thus:

Damage calculation follows legal liability; it does not create liability.

25. Case 6 — Dubai Court of Cassation, Appeal No. 440/2016

Principle

The case addressed contractual termination and the relationship between contractual rights, contractual stability and good faith.

Relevance

Where algorithmic systems automatically calculate termination charges, penalties or financial losses, the court may need to examine:

• the contractual basis;
• good faith;
• reasonableness of the calculation;
• contractual limitations;
• actual damage.

An automated contractual calculation is therefore not necessarily conclusive merely because it is specified by software.

Relevance: Contractual damages analogy.

26. Case 7 — Dubai Court of Cassation, Appeal No. 313/2007

Principle

The case concerned contractual termination powers and their legal limitations.

Relevance

Algorithmic systems increasingly exercise contractual functions automatically.

For example:

• calculating early termination charges;
• determining account losses;
• applying contractual penalties;
• calculating service credits.

The fact that a computer automatically calculated the amount does not remove the need to establish the underlying legal entitlement.

Relevance: Contractual-power analogy.

27. Case 8 — UAE Federal Supreme Court, Penal Cassation No. 1422/2022

Principle

The Court emphasized that evidence must have sufficient probative force and that the court should examine and scrutinize the evidentiary material.

Relevance

This principle is directly useful to algorithmic valuation.

A damages model should be tested for:

• reliability;
• completeness;
• consistency;
• methodology;
• underlying data;
• logical assumptions.

The court should not treat mathematical sophistication as equivalent to evidentiary reliability.

Relevance: Strong evidentiary analogy.

28. Case 9 — UAE Federal Supreme Court, Penal Cassation No. 1093/2019

Principle

The trial court has authority to evaluate evidence and determine which evidence is reliable and probative.

Relevance

The court can therefore compare:

• algorithmic valuation;
• human expert testimony;
• accounting records;
• contracts;
• market data;
• bank statements;
• historical financial information.

An algorithm does not automatically have priority over conventional evidence.

Relevance: Evidentiary analogy.

29. Case 10 — UAE Federal Supreme Court, Penal Cassation No. 660/2023

Principle

The Court recognized the trial court's authority to draw reasonable conclusions from evidence, provided the inference is logically supported by the record.

Relevance

This is important for statistical damages.

An algorithm may identify:

“The defendant's conduct probably caused 70% of the revenue decline.”

The court must still determine whether the underlying evidence makes that inference sufficiently reliable for legal purposes.

Relevance: Causal-inference analogy.

30. Algorithmic Damage Calculation Framework

A practical UAE methodology can be structured into eight stages.

Stage 1 — Establish liability

Determine:

• breach;
• wrongful act;
• contractual violation;
• abuse of rights;
• other legal basis.

Stage 2 — Establish damage

Identify:

• actual financial loss;
• lost profits;
• additional expenses;
• property loss;
• personal injury;
• reputational harm;
• other legally recoverable injury.

Stage 3 — Establish causation

Determine:

What part of the damage resulted from the defendant's conduct?

Stage 4 — Collect reliable data

Potential sources include:

• accounting records;
• invoices;
• bank statements;
• contracts;
• tax records;
• sales databases;
• employment records;
• medical records;
• project schedules;
• market statistics.

Stage 5 — Construct the counterfactual

Determine:

What would probably have happened without the wrongful act?

This is often the most important stage.

Stage 6 — Apply the model

Possible methodologies include:

• regression analysis;
• discounted cash flow;
• event studies;
• before-and-after comparison;
• benchmark analysis;
• synthetic-control analysis;
• Monte Carlo simulation;
• probability-weighted valuation;
• machine-learning forecasting.

Stage 7 — Test assumptions

The court or expert should examine:

• sensitivity;
• error margins;
• alternative assumptions;
• missing data;
• outliers;
• model bias.

Stage 8 — Judicial determination

The final amount is determined by the court or tribunal, not by the algorithm.

31. Common Algorithmic Methodologies

A. Regression analysis

Regression may identify the relationship between:

• wrongful event;
• sales;
• profits;
• prices;
• customer numbers.

Useful for business-loss claims.

B. Event-study methodology

An event study examines changes surrounding a specific event.

Example:

Publication of harmful information → immediate market-value change.

This may be useful in commercial and securities-related disputes.

However, other simultaneous events must be controlled for.

C. Difference-in-differences

This compares:

• affected group;
• unaffected group;

before and after an event.

It can help estimate the incremental impact of wrongful conduct.

D. Monte Carlo simulation

Where multiple future variables are uncertain, the model generates thousands of possible scenarios.

It may produce:

• expected loss;
• median loss;
• probability distribution;
• confidence intervals.

This can be useful for complex commercial disputes.

E. Discounted Cash Flow

Common in:

• business valuation;
• lost-profit claims;
• shareholder disputes;
• commercial damages.

Future cash flows are converted into present value.

32. Sensitivity Analysis

A sophisticated damages model should test how the result changes when assumptions change.

For example:

This demonstrates that the alleged loss may depend heavily on assumptions.

The court can therefore determine whether the claimed amount is robust or speculative.

33. Probability-Weighted Damages

Some claims depend upon uncertain future events.

Example:

• potential contract = AED 50 million;
• probability of obtaining it = 30%.

Expected value:

AED 50 million × 30% = AED 15 million.

But legal recoverability depends on the applicable UAE principles concerning certainty, causation and loss.

The probability figure itself must be supported by evidence.

34. Avoiding Speculative Damages

One of the most important safeguards is:

The more uncertain the underlying assumptions, the less persuasive the algorithmic calculation becomes.

A sophisticated AI model cannot transform speculation into proof.

For example:

If a claimant has no historical sales records, an AI system should not be permitted to create artificial historical performance merely to calculate lost profits.

35. Data Quality

The principle “garbage in, garbage out” is especially important.

A model may be mathematically perfect but legally unreliable if the input data is:

• incomplete;
• inaccurate;
• manipulated;
• biased;
• outdated;
• selectively chosen.

Therefore, the court should examine both:

Model validity

and

Data validity.

36. Explainability

Where AI is used to calculate damages, the opposing party should, where legally appropriate, have an opportunity to understand:

• what data was used;
• what methodology was applied;
• what assumptions were made;
• how the result was produced.

A completely unexplained “black-box” figure may be difficult to challenge effectively.

37. Expert Evidence

Algorithmic damage disputes will frequently require experts.

An expert should ideally explain:

1. methodology;
2. data sources;
3. assumptions;
4. model architecture;
5. error rate;
6. sensitivity;
7. alternative models;
8. causation assumptions;
9. limitations.

The expert's role is to assist the court, not replace its legal judgment.

38. Difference Between Mathematical and Legal Certainty

This distinction is crucial.

Mathematical certainty

The algorithm accurately performs its calculations.

Legal certainty

The evidence establishes that the calculated loss is legally attributable to the defendant and recoverable under law.

A model may have:

99.9% computational accuracy

while still producing a legally irrelevant result if the underlying causal assumption is wrong.

39. Algorithmic Damages and Mitigation

The claimant generally should not unnecessarily allow losses to increase after discovering the wrongful conduct.

For example:

A business discovers a cyberattack but waits six months before implementing available protective measures.

The defendant may argue that some later losses resulted from failure to mitigate.

An algorithm can model:

Loss with reasonable mitigation − actual loss

to help determine the incremental amount.

40. Multiple Defendants

Where several defendants contributed to the damage, the model can help separate:

• Defendant A's contribution;
• Defendant B's contribution;
• independent market factors;
• claimant's contribution.

However, mathematical apportionment does not automatically determine the legal allocation of liability.

The court must apply the applicable UAE rules concerning multiple causes and responsibility.

41. Algorithmic Damages in Arbitration

Arbitration is particularly suitable for sophisticated damages modelling.

Parties may present:

• forensic accountants;
• economists;
• data scientists;
• valuation experts;
• AI specialists.

The tribunal may need to evaluate competing models.

For example:

Claimant model: AED 50 million

Respondent model: AED 8 million

Tribunal: determines which assumptions and evidence are legally and factually persuasive.

42. Confidentiality and Trade Secrets

Algorithmic valuation may involve commercially sensitive information.

The court or tribunal may need to protect:

• source code;
• proprietary models;
• customer data;
• pricing information;
• confidential business records.

Possible approaches include:

• confidential expert review;
• redacted disclosure;
• protective measures;
• controlled access;
• independent expert analysis.

43. Algorithmic Bias in Damage Calculation

AI itself can introduce bias.

For example, a model estimating future income may systematically undervalue certain categories of workers.

Potential causes include:

• biased training data;
• historical discrimination;
• inappropriate variables;
• flawed assumptions.

A legally responsible damages model should therefore be tested for statistical and substantive bias.

44. Algorithmic Damage Calculation in Different Areas

45. Important Legal Safeguards

A UAE court considering an algorithmic damages calculation should ideally ask:

1. Is the underlying damage legally recoverable?

2. Is the data authentic?

3. Is the methodology scientifically or economically reasonable?

4. Are the assumptions transparent?

5. Is causation established?

6. Have alternative causes been considered?

7. Has mitigation been considered?

8. Is there double counting?

9. Is the model biased?

10. Is the result excessively speculative?

46. Relationship Between AI Calculation and Judicial Discretion

The court should maintain the final decision-making authority.

The correct structure is:

Algorithm → Evidence → Expert explanation → Judicial assessment → Legal award

Not:

Algorithm → Automatic compensation

This distinction protects judicial independence and ensures that mathematical modelling remains subordinate to legal standards.

47. Key Problems

The major problems likely to arise in UAE litigation include:

1. Black-box models
2. Insufficient data
3. Model bias
4. Conflicting expert models
5. Causal uncertainty
6. Speculative future profits
7. Double recovery
8. Data privacy
9. Trade secrets
10. Cross-border data
11. Manipulated inputs
12. AI hallucination
13. Overconfidence in statistical results
14. Difficulty explaining complex models to courts

48. Core Legal Formula

A useful conceptual framework is:

Recoverable Damages = Legally Recognized Loss × Causal Attribution × Evidentiary Reliability − Avoidable/Non-Attributable Loss

This is not a statutory UAE formula. It is an analytical framework for understanding how algorithmic calculations should interact with civil-law principles.

49. Conclusion

Algorithmic damage calculation methodologies can become extremely valuable in UAE civil litigation because modern disputes increasingly involve enormous datasets and complex economic relationships.

However, the fundamental rule should remain:

An algorithm can quantify damage, but it cannot independently establish legal liability.

The proper sequence is:

1. Establish legal responsibility →
2. Establish actual damage →
3. Establish causation →
4. Identify the counterfactual →
5. Collect reliable data →
6. Select an appropriate valuation methodology →
7. Test assumptions and alternative causes →
8. Avoid double recovery →
9. Apply mitigation principles →
10. Allow the court or tribunal to determine the final compensation.

The UAE cases concerning concurrent causes, expert evidence, contractual obligations, abuse of rights and judicial assessment of evidence provide the principal doctrinal foundation. The authorities such as UAE Federal Supreme Court Civil Appeal No. 79/2020, Commercial Appeal No. 215/2020, Penal Cassation Nos. 1422/2022, 660/2023 and 1093/2019, Dubai Court of Cassation Cases Nos. 266/2008 and 1/2006, and Abu Dhabi Court of Cassation Case No. 55/2016 are particularly useful by analogy.

The emerging principle is therefore:

In UAE civil law, algorithmic valuation should be treated as sophisticated evidence and an expert methodology—not as an autonomous legal determination of damages. The ultimate award must remain grounded in legally recognized harm, causation, reliable evidence, and judicial assessment.