Gains:
- Understand what analytical procedures are within the scope of BDS 520 and how to use artificial intelligence for expectation generation, ratio analysis and deviation explanation drafting.
- Ability to interpret horizontal-vertical analysis, ratio analysis and trend breakdowns with artificial intelligence support and identify deviations that need to be examined
- Ability to understand that the deviation explanation suggested by artificial intelligence is a hypothesis and cannot be considered as an audit result without being supported by evidence.
One of the most powerful weapons of an experienced auditor is "expectation". Before looking at the financial statement, he knows roughly what he needs to see because he knows the business: "This company grew by 15% last year, the number of personnel is stable, so I would expect personnel expenses to increase approximately as much as inflation." Then it compares the reality with the expectation. If there is a deviation from expectation, there is either an explanation or a problem. The systematic form of this way of thinking is called analytical procedures, and BDS 520 organizes it.
The analytical procedure is to establish an expectation using reasonable relationships between financial and non-financial data and compare it with actual and look for significant deviations. It is used in three places: in planning (to see the risk), in substantive testing (for substantive verification) and at the end of the audit (for an overall reasonableness check). In this unit, you will learn how to use artificial intelligence in expectation formation, ratio analysis and deviation explanation draft production; But we will discuss why the explanation proposed by AI is a hypothesis, not evidence.
Tools of the analytical procedure
The main analytical tools used by the auditor are:
- Horizontal analysis: Examining the change (amount and %) of an item from year to year. "Sales increased by 40%; why?"
- Vertical analysis: Ratio of items to a whole (e.g. the ratio of each expense to revenue). "While cost of goods sold was 62% of revenue, this year it became 71%; why did the margin narrow?"
- Ratio analysis: Calculating meaningful ratios (gross profit margin, inventory turnover rate, receivables collection period, current ratio) and examining its trend and comparison with the industry.
- Trend analysis: Seeing multiple periods together and catching breaks.
- Reasonableness test: Establishing an independent expectation (e.g. “interest expense ≈ average debt × interest rate”) and comparing it with reality.
AI does all of these calculations in seconds, puts them in a spreadsheet and flags deviations. Moreover, it can suggest possible explanations for deviations. Here is the critical point: the explanation produced by AI is an initial hypothesis. “Margin has narrowed because raw material prices have increased” may be reasonable, but the auditor cannot conclude without corroborating it with contracts, invoices, management disclosure, and—especially in the case of unexpected deviations—independent evidence.
The more sensitive the expectation, the stronger the procedure
An important subtlety of BDS 520: the reliability of an analytical procedure depends on how precisely and independently the expectation is established. If you build the expectation directly from the data provided by the business, you fall into circular logic (validating the same data with the same data). A strong expectation is based on independent inputs as much as possible (non-financial data, external market data, your own account). AI can help you build an expectation model — “what inputs do I need to independently calculate interest expense?” — but the auditor is responsible for the independence of the inputs and the appropriateness of the model.
Tip: Predetermine the "insignificance threshold" when interpreting the deviation. How much deviation from expectation will be considered "investigative"? Set this threshold upfront, consistent with materiality, so you don't have to chase every little fluctuation the AI produces.
Establishing strong expectations with non-financial data
The analytical procedure is at its most powerful when you build the expectation from non-financial data, not just financial data. For example, you can independently estimate a hotel's room revenue by multiplying the number of room-nights occupied by the average room rate; you can approximate the energy expense of a manufacturing business from the number of units produced and typical consumption per unit; You can predict a school's education income from the number of students and average fee. Such expectations do not fall into the circularity trap because they are independent of the business's own accounting record and provide a real cross-check. AI helps you establish the inputs and formula for this independent expectation model; but the auditor is responsible for ensuring that the inputs are truly independent and reliable (number of students, room-night, production quantity correct?). If the difference between expectation and reality is large, this indicates either an accounting error or a problem with the expectation input; Both need to be investigated.
Step by step analytical procedure
- Set expectations. Write down what you expect for the item under review, with independent input if possible.
- Prepare data, verify completeness. Is the actual data you will be comparing accurate and complete?
- Calculate with AI. Make horizontal/vertical/ratio/trend calculations; Show the steps of each account.
- Identify deviations. Flag deviations that exceed the threshold you set beforehand.
- Generate hypotheses but verify. Ask the AI for possible explanations; test each one with proof.
- Document the result. Expectation, fact, deviation, explanation and confirmatory evidence.
three mini cases
Case 1 — Finding the real cause of the deviation. One auditor found personnel expenses increased by 38%; However, the number of personnel and inflation explained this only by 22%. He had the AI ask about divergence hypotheses; YZ listed options such as "raise, new hire, bonus, severance pay". The auditor examined the payroll records: the difference was due to a lump sum premium payment paid during the year but not spread out as an expense over the correct period, and there was a seasonality error. AI gave hypothesis; Payroll evidence and the auditor's investigation revealed the truth.
Case 2 — Mistaking a hypothesis for evidence. A team member asked AI about a 40% increase in an expense item; AI "probably is due to inflation," he said. The team member wrote this as a comment directly on the worksheet and capped the pen. The responsible person asked: inflation was 25% at that time, 40% was not explained; Additionally, the item was a rental expense with a fixed contract and did not automatically increase with inflation. The real reason was a new field added to the contract. Lesson: The explanation of AI is a hypothesis; There can be no conclusion without being tested with evidence.
Case 3 — Poor expectation. To test the reasonableness of interest expense, an auditor established his expectation directly from the interest expense recorded by the business (circular). Naturally, it turned out that "expectation = reality" and no deviations appeared. The responsible person reminded that the expectation should be established independently: average debt balance × average interest rate. Upon independent calculation, it turned out that the actual interest expense was significantly lower than expected, and no interest was accrued on a debt. Lesson: independence of expectation determines the strength of the procedure.
Weak prompt / Strong prompt
Weak prompt:
Look at this chart, tell me if there is anything abnormal.
The problem: no expectations, no threshold, no context. AI either randomly flags “anomaly” or generates made-up explanations; the result lacks control value.
Powerful prompt:
Your role: you are an analytical procedure assistant to an independent auditor. You will calculate and SHOW deviations and suggest possible explanatory HYPOTHESES; verification and conclusion are mine.Context (anonymous): Production company. Revenue current year 240M, previous year 200M (20% growth). The number of personnel is fixed.Data:[account; previous year; current year] lines (revenue, COGS, personnel expense, rent, interest expense, ...)Task:1) Make horizontal analysis (amount and % change) for each item; show calculation step.2) Vertical analysis: proportion each expense to revenue; compare for two years.3) Apply the following threshold: mark “for review” if the % change deviates from revenue growth (20%) by more than 10 absolute points.4) Suggest 2-3 possible explanations HYPOTHESIS for each item marked; Also write down the EVIDENCE by which each will be verified. State that these are not findings.5) Do not make up any numbers or industry benchmarks that you cannot derive from the data.
This prompt is powerful because it gives context and expectation (20% growth), sets a clear deviation threshold, asks for explanations as hypotheses, and ties each hypothesis to evidence.
Common mistakes
- Mistaking a hypothesis for evidence. Writing conclusions without verifying the AI's "probably..." statement.
- Establishing circular expectations. Deriving the expectation from the data to be verified and producing false "fit".
- Not setting a threshold. Chasing every small fluctuation or overlooking the significant deviation.
- Analysis without context. Asking for an “anomaly” without promoting the business; receiving meaningless signs.
- Staying superficial in unexpected deviation. Covering a surprising twist with the first plausible explanation; whereas BDS requires deeper investigation into unexpected deviation.
Note: An expected deviation (e.g. increased depreciation due to a known investment) is different from an unexpected deviation. Unexpected deviation carries higher risk and requires stronger evidence. AI cannot distinguish between the two; The auditor makes this distinction.
In summary
Analytical procedures (BDS 520) is the art of seeing deviations by establishing expectations and comparing them with reality. AI is a powerful calculation and hypothesis engine in this field: it performs horizontal/vertical/ratio/trend analysis in seconds, flags deviations, suggests possible explanations. But the power of the procedure depends on the independence of expectation, and the explanation proposed by the AI is a hypothesis, not evidence. The auditor sets the deviation threshold, the auditor ensures the independence of expectation, and the auditor verifies each statement with evidence. An unexpected deviation always calls for deeper investigation.
Application task
Take a simple two-year income statement (hypothetical) and set a growth expectation (e.g. 20%). With the powerful prompt pattern above, have the AI perform horizontal and vertical analysis, apply a drift threshold, and prompt hypothesis + evidence matching for marked items. Then write in one paragraph how to set up a "non-circular independent expectation" for an item (what independent inputs would you use).
checklist
- [ ] I have established an expectation in advance (with independent input if possible) for the item under review.
- [ ] I confirmed the completeness of the data I compared.
- [ ] I received the horizontal/vertical/ratio analysis showing the calculation steps from the AI and verified it.
- [ ] I have set a deviation threshold consistent with materiality.
- [ ] I took the AI's explanations as hypotheses; I tested each one with proof.
- [ ] I distinguish between expected and unexpected deviation; I dug deeper into the unexpected.
- [ ] I documented expectation, reality, deviation, explanation, and corroborating evidence.