A working set of prompts for hearing what people feel, finding what drives it, and making the case for what to do next. Built from my work with community banks and credit unions, and useful for any community-focused team that listens to the people it serves. Copy one, fill in the brackets, and save the version that works. Your prompt is your methodology.
Not at a bank or credit union? The version for everyone is in the Toolkit.
From my NEFMA Fall 2026 keynote: “Your customers are buying a feeling. Now, you can measure it.”
This is the step that keeps it affordable. With one approved tool, someone already on your team can do this work. Without it, you’re hiring a data scientist or paying a vendor’s margin. Sit down with compliance and IT and agree on five things.
Most people type “analyze the sentiment of these comments” and get mush back. These eight habits fix that.
Start with your call center notes. People say more on the phone than they’ll ever type into a survey. Put the comments in one spreadsheet (date, comment, location or program), then run these in order.
The master prompt. One row per comment, ready for a pivot table.
These are [member / customer / community] comments from a [community bank / credit union / organization], collected from [call center notes / surveys / reviews]. Sarcasm is common, so read for tone as well as words. Return one table row per comment with these columns: - Comment ID - Topics mentioned, chosen only from: [staff, digital, fees, wait times, loans, branch, other] - Sentiment toward each topic: positive, negative, or mixed - Main emotion (for example: grateful, relieved, frustrated, anxious, confused) - Effort: yes if they mention repeat contact or words like "again," "still," or "third time" - Leaving language: yes or no, with the quote - Any employee or location named - The exact words that support each tag Only tag what the comment actually says. Do not infer age, gender, race, or any other personal characteristic.
Run this on the tagged table. Some things only get mentioned when they break. Fix those. Some get praised unprompted. Feature those.
Using the tagged table, identify the top [5] drivers of negative sentiment. For each driver: - Describe the trigger: what happened right before the person felt this way - Give two supporting quotes - Classify it: a basic expectation (people only mention it when it breaks) or a delighter (people praise it without being asked) Then do the same for positive sentiment. Don't give me counts or percentages. I'll calculate those in the spreadsheet.
Closeness, counted. How often do people name a person or a place?
From these comments, list every employee, branch, or location mentioned by name. Return one table row per mention with: comment ID, the name, the sentiment toward them, and the exact quote.
A list your retention team can act on this week.
Flag every comment that mentions leaving, switching, closing an account, or moving money elsewhere. Also flag any that describe repeated effort: calling again, still waiting, a third attempt. Return the comment ID, the exact quote, and the topic it relates to. Only flag what is stated.
The tool will get some things wrong. These make the errors easy to find.
Read these by hand before you share anything.
Review the tags you just produced. List any comments where the tone could be sarcastic, where the sentiment is unclear, or where you were least confident. For each, explain why and give your best reading. I'll review these by hand.
Only works if you used the same categories both months. That’s why you save the prompt.
Here are two tagged tables using the same categories: [last month] and [this month]. Describe which topics and emotions shifted, and give quotes that show the change. Don't calculate percentages. I'll do that in the spreadsheet.
Close the loop, write to the feeling, and make the case upstairs.
“You asked. We fixed it.” The fastest way to show someone’s paying attention.
People told us [theme], in comments like these: [two or three quotes]. We changed [what you changed]. Draft a short, warm message for [email / app banner / sign] that tells them we heard them and what's different now. Plain language, no jargon, under 75 words.
Your happiest people already wrote your best copy.
Here are [10 to 20] comments from the people happiest with us. What words and feelings do they use most when they describe us? Suggest three headline directions that use their language, and quote the comments each one comes from.
Leadership funds retention, risk, and cost. Translate.
Here is a finding from our comments: [finding, with two or three quotes]. Write a one-paragraph summary for our [CEO / board] that frames it in terms of retention, risk, or cost. Mention that we found it with a tool we already have. Note what we don't know yet and what a 90-day test would prove.
The same habits work on the rest of the job: explaining, evaluating, and getting things approved.
Data journalism in one prompt.
Explain [metric] to a board member with no analytics background. Use one plain analogy, say what's driving it, and say what it does not tell us. Under 150 words.
Before the demo dazzles you.
Here is a vendor proposal: [paste it]. List the questions I should ask about where our data goes, who can see it, total cost after year one, what it would take to leave, and what work it still leaves to my team. Flag any claims I should verify independently.
Walk in with the answers before they ask.
We want to use [tool] for [purpose]. Write talking points for our compliance and IT teams covering: where data goes, what is stored, the vendor relationship, and licensing. Then list the objections they are likely to raise, with plain answers. Flag anything you're unsure about so I can verify it before the meeting.
One page. The human side no tool covers.
Draft a one-page data governance charter for a [organization type] with [number] employees. Cover: who owns which data, who approves new tools, how we handle personal data in AI tools, and how often we review the charter. Plain language, one page.
These are starting points, not endorsements of fit for your organization, and tools change. Never paste member, customer, or personal data into a public AI tool. Strip personal details first, use a tool your organization has approved, and loop in your own compliance, security, and vendor-management teams before you start.