Customer feedback analysis is more useful when topic, sentiment, and urgency remain separate. A calm message can describe a serious outage; an enthusiastic feature request may need no immediate action. Magic CSV lets you use Jev to ask one focused question per column while keeping the original response beside every answer.
What the result looks like
Illustrative example with synthetic data and authored answers, not a recorded model run. Your results will depend on the source data and instructions.
Scroll tables sideways to see every column.
| # | Feedback |
|---|---|
| 1 | Please correct the company name on my invoice. |
| 2 | The app crashes whenever I export. |
| 3 | Could you add a dark mode? |
| # | Feedback | Topic |
|---|---|---|
| 1 | Please correct the company name on my invoice. | Billing |
| 2 | The app crashes whenever I export. | Reliability |
| 3 | Could you add a dark mode? | Feature request |
Try it, step by step
Import the responses you want to analyze
Open a CSV or Excel export with one response per row and a Feedback column. Keep a response ID if you have one. Remove fields you do not need for the analysis, and start with a small selection containing different kinds of feedback.
Categorize the main topic
Choose AI actions → Categorize and name the column Topic. Enter Billing, Reliability, Feature request, and Other as Allowed answers, one per line. Select Feedback under Read these columns. In Extra instructions, say: “Choose the primary subject. If multiple topics appear, choose the issue blocking the customer. Use Other when no listed topic fits.” Run the column.
Flag the messages needing a reply
Choose AI actions → Flag. Name the column Needs a reply? and select Feedback as its source. Use Extra instructions: “Yes for a direct question, an unresolved problem, or a request for help. No for standalone praise or a completed conversation with no remaining request.” Run this as a separate decision.
Add urgency and review
Choose AI actions → Score and name the column Urgency. Set 0 means to “No current problem or time pressure” and 100 means to “Core workflow blocked now with no workaround.” Read Feedback, run the column, and inspect the results. Export CSV or Excel, then open the file in Excel or Google Sheets to sort or summarize the reviewed data.
Choose themes that answer your actual question
A product team might need Reliability, Usability, Feature request, and Other. A service team might need Booking, Delivery, Refund, and Other. Start with a short list that maps to how you will use the results. A taxonomy with many overlapping themes is harder to define and harder to check.
If a response spans several subjects, decide whether you want one primary theme or separate yes/no columns for specific mentions. “Mentions billing?” and “Mentions an export failure?” can both be Yes. A single Topic column selects one answer and should have a clear tie-breaking instruction.
Avoid treating sentiment as priority
“I love the app, but nobody on my team can log in today” combines positive language with a blocking problem. An urgency column should read the operational situation rather than the tone. If you also need sentiment, make it a separate category column with Positive, Neutral, Negative, and Mixed.
Use the scores to help organize a human review, rather than treating small score differences as objective measurements. Define the endpoints in terms your team understands, test them against real examples, and keep the same rubric when comparing batches.
Keep the path back to the customer
Keep the original response and its ID in the exported file. Click uncertain AI cells to inspect and correct their answers. Check several unflagged rows too, especially messages with sarcasm, missing context, or more than one request.
Magic CSV classifies the supplied text. It does not read a customer’s full support history, send replies, assign tickets, or update your help desk. Use the reviewed export in your existing support or product workflow. If you need a draft of new text, Compose a column is a separate AI action powered by OpenAI.
Common questions
- Can one feedback row have several labels?
- A Categories column selects one allowed answer. For independent labels, create separate Yes / No columns such as Mentions billing? and Mentions a bug?. That lets both be Yes for the same response.
- Can it summarize every customer comment?
- For a generated text summary, use Compose a column in the same workspace. Jev’s Categorize, Flag, and Score actions return constrained decisions; text generation uses OpenAI.
- Does an urgency score automatically trigger an action?
- No. It adds a score to the row. Review the answers and export them for your existing workflow; Magic CSV does not automatically contact customers or change support tickets.