PRACTICAL AI AT WORK · OPENAI · CHATGPT
A spreadsheet can contain thousands of rows and still leave a manager asking a simple question: what changed, and what should we look at next? ChatGPT can help explore uploaded data and produce tables or charts. It is most useful when the question and the checking method are clear.
OpenAI documents support for spreadsheet analysis, including code-backed calculations where available. A generated chart is not proof that the data was interpreted correctly. Treat the exercise as an analysis project with visible assumptions, not a request for a clever-looking dashboard.
Start with one business question
Choose a question such as “Which product groups explain the change in monthly sales?” or “Where are support cases waiting longest?” Agree what each measure means. Revenue might mean invoiced sales, paid invoices or recognised income; those are different datasets and different calculations.
Use an approved copy of the file and keep the original intact. Remove personal information that is not needed. Give the assistant a short explanation of each important column and specify the reporting period.
Inspect before calculating
Ask for a data-quality report first. Look for duplicate records, missing values, inconsistent categories and dates stored as text. Decide how cancellations, credits and incomplete periods should be handled. Do not let the assistant silently remove awkward rows to make a cleaner story.
Have it report the number of rows used and excluded, with reasons. Reconcile a known total before asking for a more ambitious analysis. If the starting figure is wrong, every attractive chart built on it is also suspect.
A PRACTICAL WORKFLOW
Ask for a calculation you can reproduce
Analyse this approved sales export for [period]. First describe the columns, missing values and duplicate identifiers. Do not alter the source file. Propose how to handle returns and incomplete records before calculating. Then show revenue by month and product group, reconcile the result to the overall total, and provide the calculation method or code. Distinguish observed changes from possible explanations.
Request the supporting table alongside the chart. Check a small sample independently in Excel, including an ordinary record and an exception. Where the result depends on a formula or transformation, have someone who understands it review that step.
Choose the chart for the question
Use a line chart for a trend over time and a bar chart for comparing categories. Label the units, period and treatment of missing data. Avoid decorative graphics that obscure the values. A chart should help a reader make a comparison, not require them to decode the designer’s intentions.
Be careful with explanations. A fall in sales during a month does not prove that a particular campaign caused it. Ask the assistant to list plausible explanations and the additional evidence needed to test them.
Build lasting spreadsheet skills
AI can accelerate exploration, but spreadsheet skills help you recognise an unreliable result. Tables, pivot tables, formulas and chart design remain valuable. For recurring reporting, move the agreed calculations into a maintained workbook or reporting model with a named owner.
Gemini in Sheets and Claude-based analysis can also be relevant, depending on the available features and approved setup. Compare the ability to inspect the method, reproduce the result and fit the output into your existing reporting process. Start with one question your team can verify confidently.
Sources and further reading
Product information checked 4 November 2026. Features and availability depend on your plan and administrator settings. Examples and suggested workflows are illustrative.