Sampling vs. 100% Testing: Defensibility & Audit Evidence [2026]
Published: March 26, 2026 | Category: Audit Procedures | Read Time: 13 minutes | Author: CORAA Team
Introduction
Audit sampling is grounded in statistical theory: test a representative sample, extrapolate error rates to the population.
Per SA 530 (Audit Sampling), issued by ICAI, sampling is an acceptable approach when properly designed.
But there's an assumption buried in SA 530: "Sampling is practical."
That assumption breaks down when AI makes comprehensive testing feasible. If you can test 100% of entries in the same time it takes to design a sample, why sample?
This guide compares the two approaches and when to use each.
Sampling: The Traditional Approach
How Audit Sampling Works
Step 1: Define Population
- Total GL entries: 20,000
- Scope: All entries in audit period
Step 2: Calculate Sample Size
- Using SA 530 tables or statistical formulas
- Risk of incorrect acceptance: 5% (typical)
- Expected error rate: 0.5%
- Tolerable error: 2%
- Result: Sample size = 500 entries
Step 3: Select Sample
- Random selection (or systematic)
- Ensure representativeness
- 500 entries from 20,000
Step 4: Test Sample
- Auditor tests all 500 entries
- Results: 1 error found (0.2% error rate)
Step 5: Extrapolate
- Error in sample: 1 of 500 = 0.2%
- Extrapolate to population: 0.2% × 20,000 = 40 entries
- Conclusion: Approximately 40 errors in population (range: ±20 at 95% confidence)
Step 6: Evaluate
- Extrapolated error (40) < Tolerable error (2% × 20,000 = 400)
- Conclusion: Population likely free of material error
Sampling Risk
Sampling Risk = Risk that auditor's conclusion based on sample differs from conclusion if entire population were tested
Example of Sampling Risk:
Sample tested: 500 entries
- Errors found: 1 (0.2% error rate)
- Conclusion: Population ~99.8% error-free
Actual population (all 20,000 entries):
- Actual errors: 200 (1% error rate)
- Auditor missed 95% of errors (because sample wasn't representative)
Root cause: 1,000 errors happened to be concentrated in untested 95% of population.
Advantages of Sampling
✓ Practical for manual testing. Testing all 20,000 entries manually would take 300+ hours.
✓ Provides acceptable audit evidence (per SA 530, when properly designed).
✓ Statistically defensible. Using tables/formulas, sample sizes are predictable.
Disadvantages of Sampling
✗ Inherent sampling risk. Even with proper design, conclusions could be wrong.
✗ Extrapolation uncertainty. If 1 error found in 500, actual error rate could be 0%, 1%, or 5%.
✗ Concentrated errors not detected. If all errors happen to be in untested 95% of population, they're missed.
✗ NFRA concern. Inspection reports note: "Sampling-based procedures missed material errors subsequently identified in post-audit review."
100% Testing: The Modern Approach
How 100% Testing Works
Step 1: Define Population
- Total GL entries: 20,000
- Scope: All entries in audit period (same as sampling)
Step 2: Define Scanning Rules
- Rule 1: Flag entries >10% of account average
- Rule 2: Flag entries exactly round numbers (₹1,00,000)
- Rule 3: Flag duplicate patterns
- Rule 4: Flag timing anomalies (weekends, post-close)
Step 3: Scan 100% of Entries
- AI runs rules on all 20,000 entries
- Time: 5-10 minutes (automated)
Step 4: Flag Anomalies
- Entries matching rules flagged
- Result: 250 entries flagged (1.25% of population)
Step 5: Auditor Investigates
- Auditor reviews all 250 flagged entries
- Results: 8 errors found
Step 6: Conclude
- 100% of entries scanned
- 8 errors identified and investigated
- No material unadjusted errors remain
- Zero sampling uncertainty
Advantages of 100% Testing
✓ No sampling risk. Every entry assessed; no extrapolation uncertainty.
✓ Concentrated errors detected. All anomalies found regardless of where they cluster.
✓ Defensible to NFRA. "Tested 100% of entries; no sampling risk."
✓ Feasible with AI. What took 300 hours manually takes 5 hours with automation.
Disadvantages of 100% Testing
✗ Requires AI/automation tool. Manual 100% testing is impractical.
✗ False positives possible. Automated rules might flag items that aren't errors.
✗ Auditor judgment still required. Must investigate all flagged items.
Comparison Table: Sampling vs. 100% Testing
| Aspect | Sampling | 100% Testing |
|---|---|---|
| Coverage | 2-5% of population | 100% of population |
| Extrapolation | Yes (required) | No (not needed) |
| Sampling Risk | 5% (inherent) | 0% (eliminated) |
| Detection of Concentrated Errors | Low | High |
| Manual Effort | High (300+ hours) | Low (automation) |
| Concentrated Errors Missed Risk | High | Low |
| NFRA Defensibility | Moderate ("sampling based on...") | Strong ("100% scanned...") |
| Cost per Entry | ₹0.30 (high) | ₹0.01 (low) |
Real-World Examples
Example 1: Embedded Lease Discovery
Scenario: 50 supply contracts; 1 embedded lease (CNC equipment rental, ₹25L/year, 5-year term; missed under Ind AS 116)
Sampling Approach (5% sample):
- Sample 3 contracts (5% of 50)
- Likelihood both contracts sampled: <10%
- Probability embedded lease included in sample: ~5%
- Result: 95% chance lease is missed
100% Scanning Approach:
- Scan all 50 contracts for lease keywords
- NLP flags "equipment rental" language
- Auditor reviews flagged section
- Result: Lease identified with 95%+ confidence
Example 2: Year-End Cutoff Error Detection
Scenario: 15 revenue entries dated Dec 31, 11:59 PM (suspiciously late)
Sampling Approach (5% sample):
- Sample 500 of 20,000 entries
- Expected to find: 0.375 of the 15 entries (38% chance of finding any)
- Result: 62% chance cutoff error is missed entirely
100% Scanning Approach:
- Scan all 20,000 entries for timing anomalies
- Flag all post-close entries
- Auditor reviews all 15 Dec 31 entries
- Result: All cutoff anomalies identified
Example 3: Duplicate Payment Detection
Scenario: ₹50L vendor payment recorded twice in AP
Sampling Approach:
- Sample 1,000 of 50,000 AP entries (2% sample)
- Likelihood duplicate is in sample: ~2%
- Result: 98% chance duplicate is missed
100% Scanning Approach:
- Automated duplicate detection on all 50,000 entries
- Flags exact duplicates (same vendor, amount, date)
- Result: 99%+ chance duplicate is found
When to Use Sampling
Sampling is appropriate when:
✓ Population is small (<1,000 entries) AND comprehensive testing is impractical
✓ Procedure requires judgment (not rule-based)
✓ AI/automation tool is unavailable
Example:
- Testing lease payment terms (requires judgment on whether lease is finance vs. operating)
- 50 leases identified; sampling 10 for detailed testing
When to Use 100% Testing
100% testing is appropriate when:
✓ Population is large (1,000+ entries)
✓ Procedure is rule-based (thresholds, matching, duplication)
✓ AI/automation tool is available
✓ Comprehensive coverage is desired (eliminates sampling risk)
Example:
- Testing GL entries (rule-based: unusual amounts, timing)
- 20,000 GL entries; scan 100%
Example:
- Testing bank reconciliation (rule-based: matching deposits to credits)
- 5,000 transactions; match 100%
Audit Approach: Hybrid Strategy
Best Practice: Use Both
100% Testing (Automated):
- GL entry anomaly scanning (100% coverage)
- Bank reconciliation matching (100% coverage)
- Duplicate detection (100% coverage)
- Contract language screening (100% coverage)
Sampling (Auditor Judgment):
- Testing lease classification (50 leases; sample 10 for detailed testing)
- Testing related party pricing (20 RP transactions; sample 5 for pricing verification)
- Testing disclosure accuracy (complex technical areas)
Result:
- Comprehensive anomaly detection (via 100% scanning)
- Focused auditor judgment (on high-risk items)
- Efficient use of auditor time
NFRA Defensibility
NFRA View: Sampling
NFRA Inspector:
"Auditor tested 5% sample; extrapolated to population. Sampling uncertainty acknowledged (±2% at 95% confidence). Conclusion based on statistical inference rather than observed evidence."
NFRA Concern: With comprehensive testing feasible, why rely on extrapolation?
NFRA View: 100% Testing
NFRA Inspector:
"Auditor tested 100% of GL entries via automated scanning; 250 flagged entries reviewed by auditor; 5 errors identified and corrected. No material unadjusted errors remain."
NFRA Approval: Comprehensive evidence; no extrapolation; defensible approach.
Documentation: Sampling vs. 100% Testing
Sampling Workpaper
Procedure: Revenue Testing (Sample)
OBJECTIVE: Verify revenue transactions fairly stated
POPULATION:
- Revenue GL entries: 20,000
- Period: Jan 1 - Dec 31, 2026
SAMPLE DESIGN:
- Sample size: 500 entries (2.5%)
- Selection method: Random
- Confidence level: 95%
- Expected error: 0.5%
RESULTS:
- Entries tested: 500
- Errors found: 1 (cutoff error, ₹12L)
- Error rate: 0.2%
CONCLUSION:
- Extrapolated error: 40 entries (~0.2% of population)
- Error within tolerable limit (2% = 400 entries)
- Population likely fairly stated
- Adjustment proposed: ₹12L
100% Testing Workpaper
Procedure: Revenue Testing (100% Automated Scan)
OBJECTIVE: Verify revenue transactions fairly stated
POPULATION:
- Revenue GL entries: 20,000
- Period: Jan 1 - Dec 31, 2026
SCANNING RULES:
- Rule 1: Entries >10% of account average → Flag
- Rule 2: Round number entries → Flag
- Rule 3: Post-close entries (after Dec 31) → Flag
- Rule 4: Weekend transactions → Flag
RESULTS:
- Entries scanned: 20,000 (100%)
- Entries flagged: 250 (1.25%)
- Entries reviewed by auditor: 250
- Errors found: 8
CONCLUSION:
- 100% of entries scanned and assessed
- 8 errors identified and reviewed
- No material unadjusted errors remain
- Adjustments proposed: ₹45L
- No sampling uncertainty
Sampling Error vs. 100% Scanning
Potential Scenarios
Scenario A: Errors Randomly Distributed
Sampling works well:
- Sample captures representative errors
- Extrapolation reasonably accurate
- Auditor conclusion likely correct
100% Testing:
- Captures all errors (not just representative)
- Higher detection rate
Scenario B: Errors Concentrated in Untested Entries
Sampling fails:
- Sample captures few or no errors
- Auditor concludes "population fairly stated"
- In fact, ₹50L errors exist in untested portion
100% Testing:
- Finds all concentrated errors
- Correct conclusion
Scenario C: Few Errors, Widely Distributed
Sampling works:
- Sample likely captures errors
- Extrapolation reasonably accurate
100% Testing:
- Also finds errors
- More efficient (no sampling)
Key Takeaways
-
Sampling risk is inherent. Even well-designed samples can be unrepresentative.
-
100% testing eliminates sampling risk entirely. Every entry assessed; no extrapolation.
-
AI makes 100% testing feasible. What took 300 hours manually now takes 5 hours.
-
NFRA prefers 100% testing. When comprehensive coverage is available, sampling becomes harder to defend.
-
Use hybrid approach. 100% automated scanning for rule-based procedures; sampling for judgment-based procedures.
-
Concentrated errors are the risk. If errors cluster in untested portion of sample, they're missed entirely.
-
Defensibility matters. 100% testing = stronger evidence = better NFRA defensibility.
Related Blog Posts
- 100% Ledger Testing: From Sampling to Comprehensive Coverage
- SA 530 vs. Continuous Auditing: From Sampling to Monitoring
- AI in Audit Procedures: Complete Framework
- Manual vs. AI Audit Procedures: Comparison & Impact
About CORAA
CORAA helps Indian audit firms transition from sampling to comprehensive 100% testing. Eliminate sampling risk, improve detection accuracy, and strengthen NFRA defensibility with AI-powered audit procedures.
Learn more: Visit our website