For Learning & Development Managers ·
What you'll accomplish
Your LMS exports a spreadsheet of assessment scores for a leadership program: cohort, region, pre-score, post-score. You know how to sort it. You do not have an hour to hunt for patterns across regions and cohorts. Uploading a de-identified copy to a regular ChatGPT chat gives you group comparisons and charts in minutes, and it can point to patterns you did not think to look for. You then recheck the numbers in your own spreadsheet before they go anywhere near an executive.
What you'll need
Check before you upload (read this first): Assessment scores tied to names are personnel data. Make a copy of the export and, in the copy, do all of the following before it goes near ChatGPT.
Then check your company's AI policy. It decides which assistant may see employee data at all. On a personal or free consumer plan, chat content may be used to train models unless you turn that setting off, so look at the data controls in your settings first.
What you should see: A sheet with codes, group labels and scores only. Troubleshooting: Read the columns from left to right and ask whether any of them alone could identify someone, such as a rare job title or a one-person team. If yes, generalize it.
ChatGPT's help pages are not reachable from every network, so these steps describe where the feature generally lives. Your screen may differ.
What you should see: The file name appears with your message. Troubleshooting: If you do not see an attach option, check your plan and your workspace settings, or ask your administrator.
Start with a description before analysis, so you can catch mistakes in how the file was read.
What you should see: A column list and a row count that match your sheet. Troubleshooting: If the counts differ, tell ChatGPT what the file actually contains, or re-export the file as CSV and attach it again.
Ask for one comparison at a time, and ask ChatGPT to explain how it calculated each figure.
What you should see: A short table and, if asked, a chart. Troubleshooting: If it averages the wrong column, name the exact column header in your next message.
Ask for the limits of the comparison: group sizes, missing values and what the data cannot show. This keeps you from reporting a difference that comes from a group of three.
This step is required. See the verification note in the next section.
Distribution check:
Show the spread of post-scores by cohort as a chart, and tell me whether any cohort looks different from the others. Note the group size of each cohort.
Outliers:
Which rows have unusually large score changes, up or down? List their codes so I can check them against the original file.
Missing data:
How many rows are missing a pre-score or a post-score, by group? Tell me how the missing rows would affect the averages you gave.
Summary for a report:
Write three plain-language sentences describing what this data shows and what it does not show. Include group sizes. Do not say the program caused the change.