Spreadsheet analysis
Understand the rows before explaining the trend.
Analyse spreadsheets with clear definitions, reconciliation checks and decision-focused prompts rather than accepting a plausible chart summary.
Updated
Establish what one row represents
Before asking why revenue changed, establish whether a row represents an order, order line, invoice or payment. Summing an order total repeated on each line can produce a confident but incorrect result. Explain the identifiers, period and meaning of each important field.
Supply a data dictionary when you have one. When you do not, ask the assistant to propose definitions for you to confirm rather than silently guessing. A blank value is not necessarily zero, and a refund is not necessarily a negative sale in every export.
Inspect this workbook before analysing performance. List sheets, row-level units, identifiers, date fields, amounts and possible duplicate records. Separate confirmed definitions from hypotheses. Identify questions that could change the totals and wait for clarification on those before interpreting trends.
Reconcile before comparing
Use a known total from the source system as a checkpoint. Check whether filters, missing dates, different currencies or cancelled records explain any mismatch. A chart built from unreconciled figures can make the wrong calculation easier to believe.
Preserve the original data and document exclusions. For a material decision, review the calculations yourself or with an appropriately qualified colleague.
Calculate [metric] using these confirmed definitions. Show the included period, exclusions and calculation steps. Reconcile the result with [known source total]. If it differs, investigate the discrepancy before recommending an explanation. Preserve the original workbook.
Separate a change from its explanation
Illustrative scenario: average order value increases while order count falls. That does not by itself prove that customers are spending more willingly. A different product mix, fewer small orders or a changed discount policy could produce the same headline.
Ask for segment comparisons using available fields. Do not let a model present an unmeasured explanation as a finding.
Explain the observed change in [metric] by separating volume, mix and unit-value effects where the data allows. Show supporting calculations and unresolved alternatives. State which explanations cannot be tested with this workbook. Do not infer causality from a trend.
Make the analysis reproducible
A useful handoff records how the result was produced, not only what the chart appears to say. Request formulas or a transformation log, source ranges and a short list of checks. Verify boundary cases such as refunds, missing identifiers and dates on the period boundary.
Return a decision brief plus a reproducibility note: source sheets, included rows, definitions, transformations, formulas and checks. Label estimates separately from observed values. Identify the conclusion most sensitive to our assumptions and the next measurement that would help.
Common questions
Why does AI give different totals for the same spreadsheet?
Check row-level units, filters, dates, duplicates, currencies and treatment of blanks or refunds. Require an explicit calculation definition and reconciliation.
Should I upload the entire customer export?
Provide only the fields and records needed, and only material you are authorised to share. Remove unnecessary personal information.
Keep exploring
About this workflow
This is an original AI-assisted editorial workflow. Illustrative scenarios are not customer case studies, and the prompts have not been systematically evaluated on live models. Check consequential outputs against your sources and use qualified review where needed.