As of 13 August 2026, AI can detect anomalies in your business data.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, chat-fluent skill, and roughly 1 hour until you can act on the result.

   How to actually do it

   1. Export the relevant business data as a CSV or spreadsheet, retaining dates, row identifiers, metric values, and any category or location fields needed to explain normal variation.
   2. Open the export and record what one row represents, the date range, the meaning and units of each column, known seasonal patterns, expected ranges, and any business events that may affect the figures.
   3. Remove unnecessary personal or confidential fields, then attach the cleaned file and paste the prompt with the bracketed sections replaced by your actual data description.
   4. Ask the model to produce the data-quality report and anomaly table, including the comparison baseline and calculation for every flagged item.
   5. Compare the anomaly table against the source spreadsheet by locating each cited row or period and checking the values, dates, filters, duplicates, missing entries, and calculations.
   6. Check each high-priority finding against operational records such as orders, invoices, staffing changes, outages, campaigns, or stock records before treating it as a business problem.
   7. Ask the model to rerun the analysis with any confirmed context and then send the verified findings to the colleague responsible for the affected process.

   Prompt

   Analyse the attached business data for anomalies. The data covers [DATE RANGE] and each row represents [WHAT ONE ROW REPRESENTS]. The important columns are [COLUMN NAMES AND MEANINGS]. Normal operating conditions are [KNOWN RANGES, SEASONAL PATTERNS, TARGETS OR BUSINESS RULES].
   
   Do not invent missing values, explanations, thresholds, or business context. First check the data types, duplicate rows, missing values, inconsistent labels, impossible values, and date coverage. State any data-quality problem that could make anomaly detection unreliable.
   
   Then identify unusual records, groups, periods, or changes using both the supplied business rules and suitable statistical comparisons. For every finding, show the relevant date or row identifier, metric, observed value, comparison baseline, size of the difference, and confidence or limitation. Separate data errors, unusual but potentially valid events, and patterns that need investigation. Do not call something an anomaly solely because it is large if the data shows a normal seasonal or one-off pattern.
   
   Return:
   1. A short data-quality report.
   2. A table of anomalies ranked by priority.
   3. A plain-English explanation of what each finding does and does not show.
   4. The exact checks or calculations used, in a form I can reproduce in a spreadsheet.
   5. Three practical follow-up checks for the highest-priority findings.
   6. A list of questions you need answered before treating any finding as a confirmed business problem.
   
   Use only the attached data and the context above. If the dataset is unsuitable for a reliable conclusion, say so clearly rather than filling the gaps.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.