Transformez 500 avis clients en une feuille de route priorisée avec ce prompt IA
Why this prompt matters
Without a systematic way to analyze feedback, product teams default to building what the loudest customer asked for last, not what would actually move retention or NPS. Missprioritized sprints cost engineering weeks on features that address no real pain. A single poorly-read feedback batch can push a team to solve a symptom instead of the cause, compounding both technical debt and user-experience debt for the next two quarters.
What we use it for
Your team just closed a quarterly NPS survey and has 400 open-text responses sitting in a spreadsheet alongside 200 unread App Store reviews from the past month. Sprint planning is on Friday and the team is debating what to build next, but no one has had time to read through 600 feedback items. You need clear signal from that noise before the meeting starts.
Prompt
Role: Act as a senior product analyst with expertise in customer research and data synthesis. Your job is not to summarize feedback — it is to extract signal, identify patterns, and translate customer language into actionable product decisions that a team can execute this quarter. Context: I have collected [NUMBER] pieces of customer feedback from [SOURCES: App Store reviews / Support tickets / NPS surveys / User interviews / Social media / Other]. The product is [BRIEF PRODUCT DESCRIPTION]. Our current development focus is [AREA: onboarding / core feature / performance / reliability / pricing / other]. I want to understand what customers are actually saying versus what they are asking for, identify the most painful issues versus nice-to-have requests, and prioritize what to fix or build next. Task: 1. Cluster the feedback into themes. Identify 5 to 8 meaningful themes — not generic categories like "UI" or "bugs", but specific patterns like "users cannot find the export function" or "onboarding flow loses users after step 3". 2. For each theme, estimate frequency (how many feedback items relate to this theme), sentiment intensity (how frustrated or delighted do users sound), and inferred business impact (if unresolved, what does the company lose?). 3. Identify the top 3 Quick Wins — issues that appear frequently, cause real friction, and are likely low effort to fix. 4. Identify the top 3 Strategic Priorities — themes that represent deeper product gaps with high customer impact. 5. Flag any feedback that suggests a customer segment or use case the product is not currently serving well. 6. Write one paragraph that could serve as an internal brief: what are customers really saying, in plain language? Constraints: - Do not treat feature requests at face value. Look for the underlying job the customer is trying to do. - Separate correlation (many people mentioned X) from urgency (these customers sound ready to churn over X). - Do not invent themes that are not actually supported by the feedback. - Keep technical jargon out of the internal brief — write it so a non-technical founder or executive can read it. - If the feedback volume is too small to draw reliable conclusions, say so clearly rather than speculating. Output Format: 1. Theme clusters (name, description, estimated frequency, sentiment intensity: low/medium/high, inferred business impact) 2. Quick Wins (3 items with reasoning) 3. Strategic Priorities (3 items with reasoning) 4. Underserved segments or use cases (if any) 5. Internal brief paragraph (2-4 sentences, plain language) 6. Confidence note (how confident are you in these findings given the data quality and volume?) Feedback to analyze: [PASTE YOUR CUSTOMER FEEDBACK HERE — reviews, tickets, survey responses, or interview notes]
Result
THEME CLUSTERS
1. Export and data portability Description: Users are trying to get their data out of the product into spreadsheets, other tools, or for reporting to clients. The current export button exists but produces CSV files with inconsistent column naming, no date filters, and no per-project scope. Estimated frequency: 27% of feedback items Sentiment intensity: High — several users have directly stated they are evaluating alternatives because of this Inferred business impact: Significant churn risk in the B2B segment; enterprise buyers consistently flag data ownership as a procurement requirement
2. Onboarding drop-off after step 3 Description: First-time users consistently report confusion at the workspace setup step. The term "workspace" does not match the mental model users arrive with, and the required fields feel like bureaucracy before they have seen any value. Estimated frequency: 19% of feedback items Sentiment intensity: Medium — frustration rather than rage, but correlates with low day-7 retention in support ticket volume Inferred business impact: Estimated 30-40% of trial signups do not complete setup. Improving this step is the highest-leverage acquisition funnel fix available.
3. Notification overload Description: Users are receiving too many emails and in-app pings for low-priority events. Several long-term customers report that they have disabled all notifications, which means they are also missing important alerts. Estimated frequency: 14% of feedback items Sentiment intensity: Medium — annoyance that is eroding trust with power users Inferred business impact: Notification fatigue is a known precursor to disengagement; if power users are tuning out, they are also missing important product updates and re-engagement touchpoints
4. Mobile app stability on Android Description: Android users specifically report crashes on the reports screen when viewing projects with more than 50 items. This is a distinct pattern from general performance complaints. Estimated frequency: 11% of feedback items Sentiment intensity: High — these users are blocked, not inconvenienced Inferred business impact: Android accounts for an estimated 38% of mobile sessions; crashes on a core screen are a direct retention risk
5. Pricing and tier confusion Description: Users do not understand what the Pro plan includes versus Free. Several churned users mentioned they paid for Pro expecting a feature that is actually on Enterprise only. Estimated frequency: 9% of feedback items Sentiment intensity: High — these users feel misled Inferred business impact: Pricing page clarity could recover failed upgrades and reduce churn in the $29/month tier
QUICK WINS 1. Fix the export CSV column naming and add a date range filter. Low engineering effort, directly addresses the most urgent churn signal. 2. Rename "workspace" to "project" in the onboarding flow and remove the two required fields that users skip or fill with placeholder text. 3. Add a notification frequency control (daily digest vs. real-time) as a single toggle. Existing notification infrastructure supports this; it is primarily a UI addition.
STRATEGIC PRIORITIES 1. Export and data portability — build a full export center with scoped exports, date filters, and PDF report output for client-ready deliverables. 2. Onboarding redesign — the step-3 confusion is a symptom of a deeper mismatch between product language and user mental models. A short Jobs-To-Be-Done research sprint before the next redesign would pay for itself. 3. Mobile reliability — Android crash patterns suggest the reports screen has an unhandled edge case at scale. This needs a dedicated bug sprint before the next mobile release.
UNDERSERVED SEGMENTS Freelancers managing multiple clients appear in roughly 15% of feedback but the product currently treats all projects as belonging to a single workspace owner. This segment is hitting limits the product was not designed for, and several are asking for client-portal or sharing features that would require a different product model.
INTERNAL BRIEF Customers who are churning or at risk are not unhappy with the product's core concept — they are blocked by execution gaps: they cannot get their data out easily, they cannot set up the product without hitting confusing terminology, and on Android they cannot reliably use a core screen. The clearest near-term wins are operational fixes, not new features. The most important strategic gap is that a meaningful segment of freelancers is using the product in a way it was not designed for, and either the product adapts or that segment churns to a competitor built for them.
CONFIDENCE NOTE Moderate to high. The sample of 600 feedback items is sufficient to identify the top themes with reasonable confidence, but frequency estimates should be treated as directional, not precise. The Android crash pattern and export frustration are robust findings. The pricing confusion theme is based on a smaller subset and should be validated with a targeted survey before committing significant design work.
La plupart des équipes produit passent plus de temps à débattre des priorités qu'à lire les retours qui pourraient trancher le débat. Les données existent — avis App Store, réponses textuelles NPS, tickets de support, notes d'entretiens utilisateurs — mais traiter des centaines de commentaires bruts pour en tirer une décision de feuille de route défendable prend un temps que la plupart des équipes n'ont pas avant le sprint planning.
Ce prompt confie à l'IA un travail structuré sur ces retours. Au lieu de demander un résumé, il demande une extraction de signaux : regrouper les retours en thèmes pertinents, noter chacun par fréquence et urgence, distinguer la corrélation du risque de churn, et produire trois éléments que la plupart des analyses de retours oublient — les Quick Wins, les Priorités Stratégiques, et un signalement des segments d'utilisateurs mal desservis.
Ce qui rend ce prompt différent
Les prompts génériques du type « analyse mes retours » donnent des résumés génériques. Celui-ci force le modèle à distinguer ce que les clients disent de ce qu'ils veulent vraiment dire. Un utilisateur qui dit « ajoutez un mode sombre » peut en réalité être frustré par la fatigue oculaire lors de longues sessions — le besoin sous-jacent est le confort, pas l'esthétique. La contrainte de « chercher le job sous-jacent » pousse systématiquement les modèles au-delà des listes de fonctionnalités de surface.
L'exigence de note de confiance est tout aussi importante. Si vous collez 12 avis et demandez une analyse thématique, la plupart des modèles inventeront des schémas. Ce prompt demande au modèle de déclarer quand l'échantillon est trop petit pour tirer des conclusions fiables — ce qui s'avère étonnamment utile quand vos données sont maigres.
Comment l'utiliser
Remplissez quatre champs : le nombre d'éléments de feedback, leur provenance, une description du produit en une phrase, et votre domaine de développement actuel. Collez ensuite les retours bruts en bas. Le modèle s'occupe du reste — clustering, scoring, et rédaction du paragraphe de brief interne que vous pouvez déposer directement dans un message Slack ou un document de planification.
Fonctionne mieux avec Claude 4 Sonnet, GPT-5, ou Gemini 2.5 Pro. Pour de gros volumes de feedback (500+ éléments), collez par lots de 150 à 200 et demandez au modèle de fusionner les clusters thématiques lors d'un dernier passage.
À quoi ressemble le résultat
Le prompt produit six sections structurées : des clusters thématiques avec scores de fréquence et d'impact, des Quick Wins avec justification, des Priorités Stratégiques avec justification, les segments mal desservis (le cas échéant), un brief interne en langage clair, et une note de confiance sur la qualité des données. Le brief interne est conçu pour être lisible par un fondateur ou un dirigeant non technique sans édition.
L'exemple de résultat dans cet article est basé sur un produit SaaS fictif avec 600 éléments de feedback répartis sur cinq thèmes — de la friction à l'export aux crashes Android en passant par la confusion sur les prix. Il est représentatif de la profondeur et de la spécificité que le prompt produit de manière fiable.