ISA 520 Analytical Procedures: Setting Expectations and Investigating Variances

You've built an expectation for revenue. The client's actual figure is 8% higher. Your file says "in line with expectations" and moves on. If that sounds familiar, it's the single most common analytical procedures deficiency flagged in European quality reviews. The question isn't whether you ran the procedure. It's whether the expectation was precise enough to mean anything.

ISA 520 analytical procedures require auditors to develop an independent expectation of a recorded amount using reliable data, set a threshold for acceptable variance, and investigate any difference that exceeds that threshold by obtaining corroborating evidence beyond management inquiry (ISA 520.5 and ISA 520.7).

Table of contents

1. Planning analytics versus substantive analytics
2. Developing the expectation under ISA 520.5
3. How reliable is your data? The ISA 520.A4-A5 assessment
4. Setting the threshold: ISA 520.A12-A14 in practice
5. Investigating variances under ISA 520.7
6. Worked example: Bakker Industrial B.V.
7. Reviewer checklist for analytical procedures files
8. Common mistakes
9. Related content

If you're looking for a ready-to-use dual-threshold variance analysis tool, the ciferi analytical review tool builds the expectation, sets both percentage and absolute thresholds, and flags variances that need investigation. The rest of this post walks through the ISA 520 requirements that sit behind it.

Planning analytics versus substantive analytics

ISA 520 covers analytical procedures used as substantive procedures during the audit and those performed near completion under ISA 520.6. Planning-stage analytical procedures (identifying risks and directing attention) fall under ISA 315. The distinction matters because the documentation bar is different.

Planning analytics require you to identify unusual relationships. That's it. You spot something odd, you adjust your risk assessment, you move on. Substantive analytics under ISA 520 require you to develop a sufficiently precise expectation, define what "significantly different" means before you look at the actual figure, and then corroborate every variance that exceeds your threshold. The second category produces nearly all the inspection findings.

Most teams treat both the same way. They run a quick comparison, note whether the number looks reasonable, and document a one-line conclusion. That approach works for planning. For substantive analytics, it fails every time.

A third category exists: completion-stage analytical procedures under ISA 520.6. These are performed near the end of the audit as an overall review, not as substantive testing. Their purpose is to confirm that the financial statements are consistent with the auditor's understanding of the entity. They are less precise than substantive analytics but serve a different function. If a completion analytical flags something unexpected (a margin shift, an unusual cost trend, a balance that moved in a direction inconsistent with the auditor's knowledge of the business), it triggers a reassessment of whether sufficient appropriate evidence has been obtained. Completion analytics are not optional. ISA 520.6 makes them mandatory on every engagement.

Developing the expectation under ISA 520.5

ISA 520.5 requires you to develop an expectation of the recorded amount or ratio. The word "develop" does real work here. Copying last year's balance and comparing it to this year's balance is not developing an expectation. Neither is asking management what they expect and writing down their answer.

An expectation is an independently derived estimate of what the balance should be, built from data that is at least partly independent of the accounting records being tested. ISA 520.A5 gives examples: prior-period results adjusted for known changes, industry data, budgets (with caution), relationships between financial and non-financial data.

The precision of your expectation determines whether the procedure has any detection power. A vague expectation (revenue should be "broadly similar" to last year) will only catch enormous misstatements. A precise expectation (revenue should be approximately €47.2M based on prior-year volume of 12,400 units, a 3% price increase effective April, and the loss of one customer accounting for €1.1M) can catch misstatements at a lower level.

Two factors control precision. The first is disaggregation. Testing total revenue as one number is less precise than testing revenue by product line, by quarter, or by customer segment. ISA 520.A13 recognises this directly. The second factor is the quality of the inputs. An expectation built from non-financial data (units shipped, headcount, square metres of rental space) combined with financial data (average selling price, cost per FTE, rent per square metre) is more precise than one built from financial data alone.

Document the inputs, the logic connecting them to the expected figure, and why you believe those inputs are reliable. The ciferi analytical review tool structures this documentation with separate fields for each input and a calculated expectation before the actual figure is visible.

A useful discipline: write the expectation sentence before looking at the actual figure. "Based on 12,400 units at an average price of €3,810, I expect revenue of approximately €47.2M." If you develop the expectation after seeing the actual result, you're rationalising the number, not testing it. Some audit software forces this sequence by hiding the actual figure until the expectation is entered. For teams working in Excel, a simple discipline is to enter the expectation on a separate tab from the trial balance data, then link them only after the expectation is complete.

The level of disaggregation also depends on what risk you're addressing. If the fraud risk relates to fictitious revenue in one product line, a total-revenue analytical procedure provides no evidence. You need the procedure disaggregated to the product line where the risk sits. ISA 520.A13 makes this point implicitly: the auditor's expectation should be sufficiently precise to identify misstatements at the relevant level. Total-level analytics serve planning and completion. Account-level or sub-account-level analytics serve substantive testing.

How reliable is your data? The ISA 520.A4-A5 assessment

ISA 520.A4 lists four factors that affect whether you can rely on data used in an analytical procedure: the source of the information, its comparability, the nature and relevance of the information, and the controls over its preparation. ISA 520.A5 adds that information from sources independent of the entity is more reliable than information generated internally.

In practice, this means you need to answer four questions before using any dataset as an input to your expectation.

Where did this data come from? If it came from the client's own system with no independent corroboration, you're using the accounting records to test the accounting records. That's circular. Industry statistics from an external database, published price indices, or independently verified non-financial data (units shipped per warehouse records reconciled to delivery confirmations) carry more weight.

Has the data been audited or reviewed? Prior-year audited figures are more reliable than unaudited management accounts. But even audited figures need adjustment for known changes (acquisitions, disposals, new contracts, lost customers). Using last year's audited revenue without adjusting for a contract that ended in June makes the prior-year figure useless as a comparator.

Is the data comparable? Comparing this year's staff costs to last year's is misleading if the entity restructured in Q3. Comparing gross margin by product line is misleading if the product mix shifted. You need to know whether the data measures the same thing in both periods.

Were controls operating over the preparation of this data? If you're using management's budget as an input, you need some basis for believing the budget was prepared carefully. A budget that the board approved and monitored monthly is more reliable than one that was prepared for the bank in January and never revisited.

Document your reliability assessment for each input. One sentence per factor is sufficient, but all four factors need to be addressed.

Setting the threshold: ISA 520.A12-A14 in practice

ISA 520.A12 states that the auditor's determination of what constitutes a significant difference from the expectation is influenced by materiality and the desired level of assurance. ISA 520.A14 notes that smaller differences may be appropriate when the analytical procedure is providing the primary substantive evidence for an assertion.

The threshold is the number below which you won't investigate a variance. Set it too high, and the procedure has no detection power. Set it too low, and you spend time investigating immaterial variances that have no impact on the financial statements.

Most firms default to performance materiality as the threshold. That's a reasonable starting point but not always right. If the analytical procedure is the only substantive procedure for a balance (common for staff costs, rent, depreciation), ISA 520.A14 implies the threshold should be lower than performance materiality. If the analytical procedure supplements other substantive procedures, a threshold closer to performance materiality may be acceptable.

A dual-threshold approach works well for most engagements. Set both a percentage threshold and an absolute threshold. A variance must exceed both to be flagged for investigation. The percentage threshold catches proportionally large movements. The absolute threshold catches movements that are large in euros regardless of percentage. For a €40M revenue line with performance materiality of €600,000, a 5% percentage threshold and a €500,000 absolute threshold would mean that a €1.2M variance (3%) would not trigger investigation (below the percentage threshold), and a €520,000 variance (1.3%) would not trigger investigation (below the absolute threshold in relative terms, above it in absolute terms but below the percentage gate). Only variances exceeding both gates need work.

Document your threshold before you compare the expectation to the actual figure. Reviewers look for this. If the threshold appears to have been set after the comparison (especially if it sits just above the actual variance), the file looks manufactured.

Where the analytical procedure is used as the primary substantive test for a balance, the threshold should reflect the level of assurance the procedure needs to provide. For a balance tested only by analytics (no detailed testing of transactions), the acceptable variance should be well below performance materiality. A common approach: set the threshold at 50-75% of performance materiality for sole-source analytical procedures, and at 75-100% of performance materiality when the analytical procedure supplements detailed testing. Document the rationale for whichever level you choose. The point is not to pick the right number from a table. The point is to demonstrate that you considered the detection risk for that specific balance and set the threshold accordingly.

What reviewers check on analytical procedures files

Inspection teams focus on four areas in analytical procedures working papers. The first is whether the expectation is genuinely independent. An expectation that reproduces the client's own forecast, without adjustment or independent inputs, provides no audit evidence. The expectation must contain at least one input that does not come from the accounting system being tested.

The second is whether the threshold was documented before the actual figure was entered. File metadata (cell edit timestamps in Excel, version history in audit software) makes this verifiable. If the threshold and the actual figure were entered at the same time, the reviewer will assume the threshold was reverse-engineered.

The third area is the quality of the investigation when variances exceed the threshold. A one-line management explanation with no corroboration fails ISA 520.7. Reviewers expect to see the explanation, the evidence obtained to test it, and the residual variance after corroboration. If the residual is material, they expect to see it on the summary of unadjusted differences.

The fourth area is completion-stage analytical procedures under ISA 520.6. These are separate from substantive analytics. They are performed near the end of the audit as an overall review of the financial statements. The purpose is to identify previously unrecognised risks of material misstatement. If the completion analytics flag an unusual relationship that was not identified during planning or fieldwork, the auditor must evaluate whether additional procedures are needed. Reviewers check that completion analytics were performed after all audit adjustments were posted and that any unusual findings were followed up.

Investigating variances under ISA 520.7

When a variance exceeds your threshold, ISA 520.7 requires you to investigate it by inquiring of management and obtaining corroborating evidence. The standard is explicit: inquiry alone is not sufficient.

This is where most files break down. The typical pattern is a management explanation ("we had a strong Q4 due to a large one-off contract") followed by a conclusion that the variance is satisfactorily explained. No corroboration. No testing of the explanation. No reconciliation of the explanation back to the numbers.

The investigation should follow a sequence. First, get the management explanation. Second, assess whether the explanation is plausible given what you already know about the entity. Third, corroborate the explanation with evidence. If management says Q4 revenue included a €2.1M one-off contract, inspect the contract, check the revenue recognition, and verify that €2.1M accounts for the variance. Fourth, recalculate. If the variance was €2.4M and the one-off contract explains €2.1M, you still have an unexplained €300,000. Decide whether that residual is below your threshold or requires further work.

If the variance remains unexplained after investigation, ISA 520.7 requires you to treat it as a potential misstatement and evaluate its effect under ISA 450. Do not bury unexplained variances in the analytical procedures working paper. They belong on the summary of unadjusted differences.

The ciferi analytical review tool includes separate documentation fields for the management explanation, the corroborating evidence obtained, and the residual variance after corroboration.

Worked example: Bakker Industrial B.V.

Bakker Industrial B.V. is a Dutch manufacturer of industrial packaging with €52M revenue, €38M cost of sales, and performance materiality of €520,000. The audit team uses a substantive analytical procedure on staff costs (€8.4M in the prior year, audited).

1. The team develops an expectation. Prior-year staff costs were €8.4M. The entity hired 12 additional warehouse staff (average cost €42,000 each) and gave a 3.2% salary increase effective January. Expected staff costs: (€8.4M x 1.032) + (12 x €42,000) = €8,669,000 + €504,000 = €9,173,000.

   Documentation note: "Expectation based on audited PY staff costs of €8.4M, adjusted for 3.2% CLA increase (confirmed to collective labour agreement dated 14 November 2024) and 12 new hires per HR headcount report reconciled to payroll register. Non-financial inputs used: headcount change from HR system, verified to employment contracts on file."

2. The team assesses data reliability. The prior-year figure is audited. The CLA increase is confirmed to the signed agreement (external document). The headcount change comes from the HR system (internal, but reconciled to employment contracts). All four ISA 520.A4 factors documented.

   Documentation note: "Data reliability: PY audited (high). CLA increase per signed agreement (external, high). Headcount per HR system reconciled to 12 employment contracts inspected (medium-high, corroborated). Budget not used as input."

3. The team sets a dual threshold. Staff costs are tested only through this analytical procedure (no detailed testing planned). Threshold set at 4% and €350,000 (below performance materiality of €520,000, reflecting that this is the sole substantive procedure for the balance per ISA 520.A14).

   Documentation note: "Threshold: 4% AND €350,000. Rationale: sole substantive procedure for staff costs. Threshold set below PM of €520,000 per ISA 520.A14 to achieve sufficient precision."

4. Actual staff costs are €9,510,000. Variance: €337,000 (3.7%). The variance exceeds the absolute threshold (€350,000 is not exceeded; €337,000 is below €350,000). Neither gate is breached. Conclusion: no investigation required.

   Documentation note: "Actual €9,510,000. Variance €337,000 (3.7%). Below both the 4% percentage threshold and the €350,000 absolute threshold. No investigation required. Staff costs accepted as not materially misstated."

A reviewer opening this file sees the expectation, the inputs, the reliability assessment, the threshold rationale, and the comparison. Every number traces to a source document. The dual-threshold approach prevented unnecessary investigation of an immaterial variance while keeping the procedure precise enough to catch a material one.

Reviewer checklist for analytical procedures files

1. Confirm the expectation uses at least one input independent of the entity's accounting records (ISA 520.A5). If every input comes from the general ledger, the procedure is circular.

2. Check that the data reliability assessment addresses all four ISA 520.A4 factors. One sentence per factor is enough, but all four must be present.

3. Verify the threshold was documented before the comparison was performed. If the file metadata shows the threshold cell was edited after the actual figure was entered, raise it with the engagement team.

4. For variances exceeding the threshold, confirm corroborating evidence exists beyond management inquiry (ISA 520.7). A management explanation without corroboration is insufficient.

5. Check that unexplained residual variances after investigation are carried to the summary of unadjusted differences under ISA 450.

6. For near-completion analytical procedures under ISA 520.6, confirm the procedure was performed after all audit adjustments were posted. Running it on pre-adjustment trial balance figures defeats the purpose.

Common mistakes

• Setting the threshold after seeing the actual variance. The AFM has flagged files where the threshold appears calibrated to sit just above the actual difference, suggesting reverse-engineering rather than genuine expectation-setting. Document the threshold before running the comparison.

• Using prior-year unadjusted figures as the expectation. If last year's file had audit adjustments, the opening balance should reflect those adjustments. Using the pre-adjustment figure from the prior year means your expectation is already misstated.

• Accepting management's explanation without corroboration. ISA 520.7 requires corroborating evidence. A note saying "per discussion with CFO, increase due to new contracts" is inquiry, not evidence. Inspect the contracts.

Related content

• Analytical procedures glossary entry. Covers the ISA 520 definitions and the distinction between planning, substantive, and completion-stage analytical procedures.
• Analytical review tool. The dual-threshold variance analysis tool referenced throughout this post, with built-in documentation fields for expectation inputs, reliability assessment, and corroboration.
• ISA 450 misstatement accumulation: tracking factual, judgmental, and projected misstatements. Where unexplained variances from analytical procedures end up when they can't be cleared.