As of 13 August 2026, AI can only partly identify customers at risk of leaving from their feedback.

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, power-user skill, and roughly 1 hour until you can act on the result.

   How to actually do it

   1. Open your customer relationship management or feedback system and export the written feedback, dates, ratings or NPS results, customer ID, product or segment, and any later cancellation or renewal outcome.
   2. Remove unnecessary names, email addresses, phone numbers and other personal data, then make sure the same customer has a consistent ID across the feedback and outcome files.
   3. Write down the business definition of leaving, the period being studied and the action your team might take for a customer flagged as higher concern.
   4. Paste the business context, the feedback data and the prompt into a chatbot, then save the returned risk categories, evidence and missing-information notes.
   5. Take a sample from each risk category and compare every category with the original feedback, checking that the cited evidence is exact and that low scores have not been treated as proof of leaving.
   6. If later churn or renewal outcomes are available, compare them with the model's categories in a spreadsheet and record missed leavers, wrongly flagged customers and records with no outcome.
   7. Ask a customer-service or analytics colleague to challenge the categories, agree a review threshold and approve a small human-led contact list before using the analysis more widely.

   Prompt

   You are helping analyse customer feedback for a UK business. Identify customers who may be at risk of leaving, but do not present this as a fact or a prediction with certainty.
   
   Business context:
   - Product or service: [DESCRIBE IT]
   - What counts as leaving: [CANCELLATION, NON-RENEWAL, LOST ACCOUNT, OR OTHER DEFINITION]
   - Relevant time window: [TIME WINDOW]
   - Available customer outcome data: [DESCRIBE WHETHER LATER CHURN OR RENEWAL OUTCOMES ARE INCLUDED]
   
   I will paste feedback data below. Each row may include a customer ID, date, channel, NPS or rating, and written feedback. Use customer IDs only for grouping and do not repeat unnecessary personal data.
   
   For each customer, return:
   - a risk category: higher concern, possible concern, or no clear leaving signal
   - the exact feedback evidence supporting that category
   - the likely issue or issues, such as service failure, price, product gap, unresolved complaint, or competitor mention
   - whether the evidence is direct, indirect, or ambiguous
   - a suggested next action that a human colleague could consider
   - missing information that prevents a reliable assessment
   
   Separate dissatisfaction from evidence of leaving intent. Do not treat a low score, an angry tone or a single complaint as proof that a customer will leave. Do not invent churn outcomes, customer context or probabilities. Flag duplicate, contradictory, very old and empty records. Explain which patterns are common across customers and which are isolated.
   
   If later churn or renewal outcomes are supplied, compare the risk categories with those outcomes, show false positives and missed leavers, and state the limitations of that comparison. Do not claim the analysis proves causation. Finish with a short list of the data and checks needed before anyone uses this to prioritise customer contact.
   
   Feedback data:
   [PASTE DATA HERE]

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