Choose the measurement method for the obligation being measured
Ind AS 37 paragraphs 36–44 require a best estimate of settlement expenditure, with probability-weighted outcomes for large populations, a midpoint for an equally likely continuous range, and consideration of other outcomes even when a single obligation’s most likely result is the starting point. A provision is not automatically the largest possible loss or a mechanical average for every case. This article checks ICAI’s 2025–26 compendium text on 27 October 2026 and focuses on the estimation model after the recognition conditions have been assessed. ICAI Ind AS 37: paragraphs 36–44 and 59.
Define the population before assigning probabilities
For warranties or similar obligations, identify the sales cohorts still exposed at the reporting date, their contractual coverage and defect experience. Use mutually exclusive outcomes that together cover the modelled possibilities. Apply probabilities to matching cost units: a per-unit repair estimate needs the relevant unit count; an estimate stated for the entire population should not be multiplied by units again. Group materially different products rather than averaging away different failure patterns. ICAI Ind AS 37: paragraphs 36–44 and 59.
Support both probabilities and settlement costs
Keep the experience data, engineering or legal assessment, repair-cost assumptions and reasons for changes. A historical percentage may require updating for known defects or changed settlement conditions. Paragraph 38 recognises management judgement informed by experience, experts and relevant additional evidence after the reporting period. Evidence should relate to the obligation and reporting-date conditions; a convenient round percentage is not a substitute. ICAI Ind AS 37: paragraphs 36–44 and 59.
Handle a single obligation with judgement
Paragraph 40 allows the individually most likely result to be the best estimate, but requires considering other possible outcomes. A distribution concentrated on higher or lower alternatives can affect the appropriate amount. Retain the most likely outcome, the alternatives and the reason for the chosen settlement estimate. Merely entering a weighted average into a spreadsheet does not establish that it measures the amount rationally required to settle or transfer that individual obligation. ICAI Ind AS 37: paragraphs 36–44 and 59.
Avoid duplicating risk adjustments
Paragraphs 42–44 address uncertainty without permitting excessive provisions. If a probability-and-cost model already reflects the relevant adverse outcomes, an additional flat contingency percentage needs an independent rationale. Trace each risk to where it enters the model. Likewise, do not select an adverse cost estimate and then deliberately inflate its probability for the same uncertainty. Preserve a risk map that distinguishes missing risks from risks already modelled. ICAI Ind AS 37: paragraphs 36–44 and 59.
Keep measurement, tax and roll-forward distinct
Paragraph 41 measures the provision before tax, with tax consequences addressed under Ind AS 12. Paragraph 59 requires review at each reporting-period end. The examples assume recognition conditions met, time-value effects immaterial, no reimbursement asset and no separate tax calculation. Their updated estimates concern the same outstanding population unless otherwise stated; settlements, new sales and discount unwinding would need separate roll-forward lines in an actual file. ICAI Ind AS 37: paragraphs 36–44 and 59.
Applicable primary references
ICAI Ind AS 37: paragraphs 36–44 and 59
Six worked illustrations
All facts and amounts below are fictional. Assumptions describe the model, not additional statutory rules.
Illustrative example 1: Weight outcomes once for a warranty population
A company has 10,000 covered units. Assume 80% need no repair, 15% need a ₹400 repair and 5% need a ₹2,000 repair. Probabilities total 100%. Expected cost per unit is ₹160; the provision estimate is ₹16,00,000. Do not apply another 20% claims factor after the probabilities have already included no-claim cases.
Illustrative example 2: Keep dissimilar product groups separate
Product A has 2,000 units, with 10% requiring repairs of ₹1,000 and 90% none: expected cost ₹2,00,000. Product B has 500 units, with 20% requiring repairs of ₹3,000 and 80% none: expected cost ₹3,00,000. Combined estimate is ₹5,00,000, supported by two populations rather than an unweighted average of failure rates.
Illustrative example 3: Use a midpoint only for the stated distribution
Assume evidence supports a continuous settlement range of ₹8 lakh to ₹14 lakh, with every point equally likely and no other relevant adjustment. The midpoint is (8 + 14) ÷ 2 = ₹11 lakh. The arithmetic is not sufficient if evidence instead indicates a skewed distribution.
Illustrative example 4: Distinguish a single-obligation judgement from a formula
A single remediation obligation has a 60% outcome of ₹12 lakh and 40% of ₹18 lakh. The mathematical expected value is ₹14.4 lakh. Assume the documented engineering assessment, considering the remedial scope and higher-cost alternatives, independently supports a best settlement estimate of ₹15 lakh. On those assumed findings the chosen amount is ₹15 lakh, ₹3 lakh above the most likely outcome. The example does not prescribe ₹15 lakh or the weighted average for every individual obligation.
Illustrative example 5: Re-estimate the same outstanding cohort
For the unchanged 10,000-unit cohort in the first illustration, updated supported probabilities are 85% nil, 12% at ₹500 and 3% at ₹2,500. Expected cost per unit is ₹135 and the new estimate ₹13,50,000. With no settlements or new obligations in this isolated model, the reduction from ₹16,00,000 is ₹2,50,000.
Illustrative example 6: Find a duplicated risk margin
A population-level model assigns 70% to total settlement cost of ₹10 lakh and 30% to ₹30 lakh, producing ₹16 lakh. Assume those outcomes and probabilities already capture the identified adverse-cost risk and no additional risk is evidenced. Adding a flat 20% creates ₹19.2 lakh, an unsupported extra ₹3.2 lakh in this model. The reason to remove it is duplication, not a rule that all risk adjustments are prohibited.
Ten-step verification table
| Step | Action | Evidence |
|---|---|---|
| 1 | Confirm a recognised present obligation | Recognition memorandum |
| 2 | Define the measured population | Cohort and coverage records |
| 3 | Choose the estimation approach | Population, range or individual obligation |
| 4 | Check outcome completeness | Mutually exclusive probabilities totalling 100% |
| 5 | Match cost and volume units | Per-unit or whole-population basis |
| 6 | Support assumptions | Experience and specialist evidence |
| 7 | Review single-obligation alternatives | Reasoned best-estimate conclusion |
| 8 | Map each risk once | Probability, cost and additional adjustments |
| 9 | Separate tax and time-value effects | Applicable measurement analysis |
| 10 | Reconcile to the closing provision | Review and roll-forward evidence |
Sixteen practitioner FAQs
1. Does this guide decide whether every uncertain item is a provision?
No. Recognition conditions must be assessed before applying these measurement controls.
2. When is expected value used?
Paragraph 39 describes probability-weighted measurement for a large population of items.
3. Should no-claim cases be excluded from a warranty model?
Include all relevant possible outcomes, including nil cost, in the supported probability distribution.
4. Must probabilities total 100%?
For the complete mutually exclusive model used here, yes; otherwise explain missing or overlapping outcomes.
5. Can I multiply a total-population cost by unit count?
No. First identify whether the cost input is per unit or already relates to the whole population.
6. Why separate product cohorts?
Different coverage, failure patterns and repair costs can make one blended assumption misleading.
7. When is a midpoint appropriate?
For the stated continuous range where every point is equally likely, subject to the actual evidence and relevant adjustments.
8. Is a midpoint appropriate for a skewed range automatically?
No. Equal likelihood is an essential assumption of that shortcut.
9. Is the most likely amount always correct for a single obligation?
No. Paragraph 40 requires consideration of other possible outcomes.
10. Is expected value mandatory for every single lawsuit or remediation?
No. Document why the chosen amount represents the best settlement estimate for that obligation.
11. Can specialist advice determine the amount without management review?
Management must evaluate its relevance and incorporate supported judgement, rather than filing an unexplained expert number.
12. Can probabilities simply be made more conservative?
Do not deliberately overstate probabilities to duplicate uncertainty already incorporated in adverse costs.
13. Are additional risk adjustments always forbidden?
No. Identify the separate risk and justify why it has not already been captured.
14. Is the provision measured net of tax?
Paragraph 41 requires measurement before tax; address tax consequences separately under Ind AS 12.
15. Why does the revised example exclude settlements?
It isolates a change in estimate; actual utilisation and new obligations need their own roll-forward lines.
16. When should the estimate be reviewed?
At each reporting-period end, using the then-current best estimate and relevant supporting evidence.
Conclusion
Build Ind AS 37 provision estimates using appropriate populations, supported probabilities and single-obligation judgement, with controls against duplicate risk margins.