An auditor asks the client for "all sales transactions for the year" and gets back a spreadsheet of $42 million in revenue. The general ledger says $48 million. Six million dollars is missing, not because anyone hid it, but because the export query ran against one of three regional sales tables and nobody flagged that. Before you analyze a single transaction, you have to be sure the data on your screen is the data the client actually has.
HIGH-FREQUENCY: A data extraction request must specify four elements:
- Attribute structure: which fields (date, account, debit, credit, document number, user ID, approver)
- Format: file type (CSV, Excel, DB export), encoding, delimiter, date format
- Source: which system and which table or report (GL detail, subledger, enterprise resource planning (ERP) module, data warehouse)
- Time period and scope: date range, entities, currencies, transaction types
"Send us the sales data" gets you whatever the client decides to send. A precise request (fields, format, system, date range, with a tie-out to a control total) gets you data you can actually use.
Common mistakes
- Sending a vague extraction request. "The journal entries" without specified fields, source, or period produces an extract whose scope the auditor cannot verify. The request is part of the evidence, so write it precisely and keep a copy in the workpapers.
- Skipping the reconciliation to a control total. Auditor receives a file, runs analytics, finds nothing unusual. The file was missing 14% of transactions because the export had a stale date filter. Without reconciling to an independently confirmable total, missing data is invisible.
- Confusing measurement scales. Averaging a risk rating, treating an account number as a numeric ratio, or running regression on nominal categories produces output that looks valid but is mathematically meaningless. Match the technique to the scale before running it.
Bottom line
- A data extraction request must specify attribute structure, format, source, and time period.
- Relational databases link tables by primary and foreign keys; the schema tells you what to ask for, and normalization stores each fact in exactly one place.
- Measurement scales (nominal, ordinal, interval, ratio; continuous, discrete) determine which arithmetic is valid for a field.
- Completeness is verified by reconciling the extract to an independently confirmable control total (record counts, debit/credit totals, or a financial-statement line).
Exam shortcut
When a question describes an auditor running analytics on a client-provided file with no reconciliation step, the answer almost always involves completeness and accuracy of the data. Pick the choice that says the auditor must verify the data before analyzing it.
The full lesson (about 2,087 words, 14 min read) adds 1 worked example, all 5 common mistakes, a self-check, free in the app.
Learning objectives
- III.A1
Browse all free CPA AUD lessons or jump into free CPA AUD practice questions.