As of 13 August 2026, AI can find outliers in your sales 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 30 minutes until you can act on the result.

   How to actually do it

   1. Export the relevant sales records from your sales or accounting system as a CSV or spreadsheet, retaining dates, transaction identifiers, sales values and the product, customer, region or channel fields needed for comparison.
   2. Open a copy of the export and remove passwords, unnecessary personal details and unrelated columns, while keeping the original file unchanged for checking.
   3. Write down what counts as unusual in this business, including seasonal periods, promotions, refunds, bulk orders, new products and known data-entry problems.
   4. Upload the copy to ChatGPT, Claude, Gemini or Julius AI and paste the supplied prompt with the column meanings and business context filled in.
   5. Ask the tool to return the ranked outlier table, the method used and the calculations rather than accepting an unexplained list of unusual rows.
   6. Compare every flagged row with the original sales system, order records, invoices, refunds and promotion calendar, and mark each one as a confirmed data issue, a genuine unusual event or unresolved.
   7. Send unresolved high-impact findings to the relevant sales, finance or data owner before changing reports, contacting a customer or taking disciplinary action.

   Prompt

   Analyse the sales data I provide and find potential outliers. Use only the data supplied and do not invent missing values, explanations or business context.
   
   Business context: [describe what is being sold, the usual sales cycle, relevant promotions, seasonal effects and any known data issues]
   Columns and meanings: [list each column and what it represents]
   What I want to detect: [for example unusually high or low order values, unusual daily totals, sudden changes by product or region, duplicate transactions, or data-entry errors]
   Relevant period: [date range]
   
   Do the following:
   1. State the number of rows, date range, important missing values and possible duplicate rows.
   2. Choose and clearly name suitable outlier methods for this data, such as comparisons with the relevant group, rolling averages, interquartile range or standard deviation. Do not use a method that ignores obvious seasonality or different product groups.
   3. Produce a ranked table of flagged rows or groups with the date or period, identifying fields, sales value, comparison baseline, size of deviation, method used and a plain-English reason for the flag.
   4. Separate data-quality anomalies from unusual but potentially genuine sales events.
   5. Explain which results are sensitive to the chosen method or incomplete data.
   6. Do not label anything as fraud, misconduct or an error unless the data proves it. End with specific checks I should perform against the original sales system before taking action.
   
   Show calculations or formulas where practical so I can reproduce the result in a spreadsheet.

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