Use AI to analyze a spreadsheet: definitions, missing values and reproducible totals

A worked example: where the task becomes difficult

A fictional service company uploads monthly jobs with revenue, travel distance and completion status. Cancelled jobs have blank revenue; some distances are miles and others kilometers. An AI summary reporting average revenue per job will be wrong if it treats blanks as zero without an agreed definition. Start with a data dictionary and a few manually checked rows rather than asking for a dramatic business insight.

Decisions to make before implementation

Define the unit of analysis: row, completed job, customer or month. Choose treatment for duplicates, cancellations and missing values. Standardize units while preserving original fields for traceability. Calculate totals using formulas or a script that another analyst can rerun. Separate descriptive patterns from causal claims; higher revenue in one region does not prove that region causes better performance.

A practical sequence for the work

Use the sequence below as a task boundary, not as a claim that the example has been executed. Work with approved inputs and the project’s actual architecture. If a required integration or permission is unavailable, keep that stage visibly incomplete rather than generating a plausible substitute result.

• Create a column dictionary and identify confidential fields before sharing.
• Inspect duplicates, missing data, types and unit consistency.
• Compute independently checkable totals and a clearly defined comparison.
• Ask AI to explain patterns, alternative interpretations and follow-up questions.

A detailed brief you can adapt for your agent

Replace the illustrative context with your approved facts and controlled inputs. Keep the stated boundaries when adapting the brief. The expected deliverable matters more than a particular tool name: ask for an explanation grounded in the inspected material and evidence for the requested outcome.

Analyze this permitted job spreadsheet using the supplied column dictionary. Report row counts, duplicates, missing values and unit issues before insights. Calculate revenue per completed job with the agreed denominator and show a reproducible formula. Separate patterns from causal explanations. Do not expose identifiable small groups or invent unavailable fields.

Failure modes that an attractive preview can hide

A model may describe a chart it did not actually calculate or silently drop inconvenient rows. Retain the row count before and after cleaning and document each exclusion. Small groups can expose individuals even after names are removed. Do not publish breakdowns whose apparent precision exceeds the data coverage or whose privacy boundary has not been checked.

Technical references: NIST: AI Risk Management Framework

Acceptance checks and the evidence to retain

Keep the cleaned-data rules, calculation method, row coverage and limitations. The result should let another person reproduce the reported total without trusting the model’s narrative.

Specific answers

Common questions