The analysis looks convincing. Revenue, cost of sales and gross profit reconcile in both years. Every subtraction works. The commentary reports that gross margin improved from 30% to 35%—a five-percentage-point gain that appears to justify confidence in pricing and production performance.
The source says the opposite. During extraction, the annual columns were reversed. FY 2025 values were labelled FY 2026 and FY 2026 values were labelled FY 2025. The arithmetic remained perfect because each displaced column still added up internally. The real result is a five-percentage-point margin decline.
The numbers add up—and the conclusion is wrong
A CFO receiving an Income Statement in Excel or PDF usually starts with an apparently simple task: compare periods, calculate growth and margins, then explain the material movements. The risk is that a mathematically valid model can be semantically wrong. A swapped header, misunderstood unit or ambiguous label may preserve every equation while reversing the business conclusion.
In this synthetic example, the incorrect extraction produces the following calculated margins:
| Calculated measure | Labelled FY 2025 | Labelled FY 2026 | Reported change |
|---|---|---|---|
| Gross margin | 360 ÷ 1,200 = 30% | 350 ÷ 1,000 = 35% | +5 percentage points |
Nothing in the formula reveals the error. Gross profit divided by revenue is correct for each extracted column. Only a comparison with the source headers can prove that 30% belongs to FY 2026 and 35% belongs to FY 2025.
How do you analyse an income statement from Excel or PDF?
To analyse an Income Statement from Excel or PDF, first identify the statement, entity, reporting periods, currency and unit directly from the source. Extract each financial value with its row and column location, confirm that the period header applies to that column, and preserve the original label. Recalculate accounting relationships, compare every material extracted value with the source, calculate supported KPIs, label unresolved ambiguity and only then write the financial conclusion.
- Source row → extracted value: retain the original label, file and precise location.
- Period and unit confirmation: bind the value to its column header, currency, scale and reporting scope.
- Validation: test both source agreement and deterministic financial relationships.
- KPI → finding → decision: calculate only from confirmed values and preserve the evidence path.
This method applies whether the same synthetic statement arrives as a structured XLSX workbook or as a PDF with a reliable text layer. The acquisition method differs; the evidential requirements do not.
Start with the source statement, not the extracted table
The following two-year Income Statement is entirely synthetic. Values are EUR thousands. Expenses are presented as positive amounts and deducted in the displayed relationships.
| Income Statement | FY 2025 | FY 2026 |
|---|---|---|
| Revenue | 1,000 | 1,200 |
| Cost of sales | 650 | 840 |
| Gross profit | 350 | 360 |
| Gross margin | 35% | 30% |
The XLSX version contains a worksheet named Income Statement. Cell A1 states the unit “EUR '000”; B2 and C2 contain the period headers FY 2025 and FY 2026. Revenue is on row 4, cost of sales on row 5 and gross profit on row 6. In the equivalent text-layer PDF, the table appears on page 2: the label is column 1, FY 2025 is column 2 and FY 2026 is column 3.
The source structure matters because a value is not self-describing. “360” is not yet gross profit, FY 2026 or EUR 360 thousand. Those meanings come from the value’s relationship to the row label, column header, unit declaration and statement context.
Trace one number from the source to the margin finding
Take FY 2026 gross profit. The selected source value is 360. Its complete lineage must survive extraction, validation, calculation and interpretation.
| Stage | Recorded evidence | Control question |
|---|---|---|
| Source row | Gross profit | Is this the displayed subtotal, not an inferred label? |
| XLSX location | Income Statement!C6 | Does column C inherit header FY 2026 from C2? |
| PDF location | Page 2, table row 6, column 3 | Does column 3 sit beneath FY 2026 in the text layer? |
| Extracted value | 360 | Does the extracted token equal the source token? |
| Confirmed meaning | FY 2026 Gross profit; EUR '000; positive presentation | Are period, line item, unit and sign explicit? |
| Arithmetic check | 1,200 − 840 = 360 | Does the subtotal reconcile? |
| Calculated KPI | 360 ÷ 1,200 = 30% | Is the denominator confirmed FY 2026 Revenue? |
| Finding | Gross margin declined by 5 percentage points | Does the comparison use 30% − 35%, not reversed periods? |
The XLSX cell and PDF position are alternative representations of the same synthetic record, not two independent confirmations. If both files were generated from one underlying report, agreement between them does not prove economic correctness. It proves consistent representation. The ledger or reporting model remains the higher-level evidence for the underlying financial amount where that assurance is required.
This is the practical role of Financial Data Lineage: the final statement remains inspectable through the KPI, confirmed canonical value, extracted token and exact source location.
Arithmetic validation and source validation answer different questions
| Control | Test | What it proves | What it cannot prove |
|---|---|---|---|
| Arithmetic validation | Revenue − Cost of sales = Gross profit | The displayed relationship reconciles within each period | That the period, unit or label was assigned correctly |
| Source validation | Extracted value, row, period header, unit and sign agree to the source location | The analytical value preserves the source meaning | That the source itself is complete or economically correct |
For FY 2025, 1,000 − 650 = 350. For FY 2026, 1,200 − 840 = 360. Both checks pass. They would also pass after the columns were reversed because the three values moved together. Arithmetic validation is necessary, but it is not an identity check.
Source validation asks a different sequence: which header governs the cell; whether a merged heading spans the expected columns; whether the unit applies to the whole table; whether parentheses mean negative values; whether the label is a source label or an inferred mapping; and whether the file contains comparative, budget or restated columns that could be mistaken for periods.
The Financial Data Validation control layer provides the wider control architecture: source-value preservation, duplicate and omission checks, subtotal recalculation, cross-statement relationships and explicit exceptions.
An unclear period or ambiguous label must stop the affected conclusion
Suppose the PDF text layer returns two numeric columns but the header extraction is incomplete. Or suppose the label “Direct costs” appears where the analytical taxonomy expects “Cost of sales”. The system may retain the observed value and propose an interpretation, but it should not silently present a period comparison or gross-margin finding as confirmed.
| Observed issue | State | Permitted output | Required resolution |
|---|---|---|---|
| Period header missing or detached | Review required | Values shown without a year-on-year conclusion | Inspect the visual header or obtain a clearer source |
| Label could exclude part of cost of sales | Review required | Source label and candidate mapping disclosed | Confirm accounting scope or supporting schedule |
| Unit absent | Blocked for amount-based findings | Ratio only if numerator and denominator demonstrably share scale | Confirm currency and scale |
| Critical source mismatch | Blocked | No affected KPI or management conclusion | Correct extraction or replace the source |
Review status belongs to the specific field, KPI and dependent conclusion. It does not require rejecting the entire document when unaffected values remain usable. Equally, a high document-level confidence score must not conceal a material uncertainty in the period header.
Once the controller confirms that column B is FY 2025 and column C is FY 2026, the dependent margins and comparison are recalculated. The prior “improvement” is invalidated rather than manually overwritten without history.
The verified result changes the direction of management attention
| Measure | FY 2025 | FY 2026 | Change |
|---|---|---|---|
| Revenue | 1,000 | 1,200 | +200 / +20.0% |
| Cost of sales | 650 | 840 | +190 / +29.2% |
| Gross profit | 350 | 360 | +10 / +2.9% |
| Gross margin | 35.0% | 30.0% | −5.0 percentage points |
Revenue grew by EUR 200 thousand, or 20%. Cost of sales grew by EUR 190 thousand, or 29.2%. Gross profit therefore increased by only EUR 10 thousand, while gross margin declined from 35% to 30%.
The source supports the direction and magnitude of the performance change. It does not yet prove the cause. Lower realised price, adverse product mix, input-cost inflation, production inefficiency, classification change or a combination of these could produce the same aggregate pattern. Those are hypotheses requiring additional evidence.
False confidence usually enters before the ratio is calculated
| Failure | Why it looks valid | Consequence | Control |
|---|---|---|---|
| Swapped period columns | Each column still reconciles | Trend direction reverses | Bind every value to its source header |
| Unit ignored | Ratios may remain correct | Amounts and materiality are misstated | Retain currency and scale |
| Ambiguous cost label mapped automatically | The wording appears similar | Gross margin scope changes | Preserve source label and review mapping |
| PDF reading order assumed | Text is extractable | Headers detach from values | Validate spatial row and column relationships |
| Calculated KPI treated as a source value | The number appears in the analysis | Lineage becomes unclear | Label source values and derived metrics separately |
| Causal commentary written from totals | The hypothesis is commercially plausible | Management acts on an unproven driver | Request driver-level evidence |
Income statement analysis from Excel benefits from explicit cells and formulas, but hidden columns, merged headers, stale formulas and inconsistent sheet conventions still require control. Income statement analysis from PDF benefits from stable page evidence, but reading order and table geometry must be reconstructed correctly. Neither format removes the need to verify extracted financial data.
The management note separates the finding from the investigation
| Element | Management statement |
|---|---|
| Verified observation | Revenue increased by 20%, gross profit increased by EUR 10k and gross margin declined by 5 percentage points. |
| Evidence status | FY 2025 and FY 2026 values agree to the source locations, units and period headers; gross-profit subtotals recalculate. |
| Interpretation | Revenue growth carried weaker gross-margin economics. The aggregate Income Statement does not establish the cause. |
| Required investigation | Controller to quantify price, product-mix and cost-of-sales contributions before pricing or cost action is approved. |
| Decision boundary | Do not treat the margin movement as a pricing failure or production-cost failure until the driver bridge reconciles to the EUR 190k cost increase and the revenue change. |
The controller needs a sales-volume and realised-price bridge by product and customer; product and channel mix; discounts, rebates and returns; standard versus actual material prices; material usage, yield, scrap and rework; labour and production-overhead variances; inventory and cost-of-sales cut-off; foreign-exchange effects where relevant; and any classification or accounting-policy changes between periods.
Those schedules should reconcile back to revenue, cost of sales and gross profit. Only then can management distinguish a pricing action from a portfolio, procurement or operational response.
When the verified finding must be presented alongside cash, forecasts and owned actions, the natural next step is the monthly CFO reporting pack example.
Use this checklist on the next Excel or PDF statement
| Check | Question |
|---|---|
| Source identity | Is the entity, statement type, version and reporting scope explicit? |
| Periods | Does every extracted column retain the exact source header and order? |
| Currency and unit | Are currency, scale and sign convention attached to every amount? |
| Field lineage | Can a material value be opened at its XLSX cell or PDF page, row and column? |
| Source agreement | Do extracted labels and values equal the source tokens? |
| Arithmetic | Do gross profit and other supported subtotals recalculate? |
| KPI definition | Are numerator, denominator, period and percentage-point convention explicit? |
| Ambiguity | Are unclear periods, labels or units routed to review rather than guessed? |
| Interpretation | Is the verified observation separated from causal hypotheses? |
| Decision evidence | Do the requested supporting schedules match the action management may take? |
The operational standard is simple: interpretation may propose what a source structure means; deterministic controls should test values and relationships; material ambiguity should remain visible for human review. The resulting analysis is useful not because the ratio was calculated quickly, but because the decision can be traced back to what the file actually said.
