AI Customer Feedback Classifier

2,000 Support Tickets Walk Into a Backlog

At Notion, the product team reads every single piece of customer feedback. Brian Freyburger, their Head of Product, has talked about this publicly — they built internal tooling specifically so PMs couldn't hide from user complaints. Most teams don't have that luxury. Most teams have a Slack channel called #feedback or a shared spreadsheet that someone updates "when they get a chance," which means never.

The gap between collecting feedback and acting on it is where product intuition goes to die.

Why Manual Classification Breaks Down

The problem isn't volume alone, though volume is a problem. A mid-stage SaaS product with 5,000 users generates roughly 200-400 pieces of feedback per month across support tickets, NPS responses, app reviews, and social mentions. At 50,000 users, you're looking at 2,000+.

The real problem is inconsistency. When three different PMs classify the same support ticket, they'll categorize it three different ways. One calls it a "UX issue," another a "feature request," a third files it under "onboarding." Productboard's 2024 survey found that 58% of product teams admit their feedback categorization is inconsistent across team members, and 41% say they've built features based on feedback that turned out to be from an unrepresentative sample.

This is underrated as a source of product failure. If your classification is noisy, your prioritization inherits that noise. You end up building for the loudest users instead of the most common pain points. Ken Norton, the former Google PM, wrote about this — the plural of anecdote is not data, but it sure feels like it when the anecdotes are in your inbox every morning.

How This Prompt Helps

This prompt processes raw feedback from any channel and applies consistent classification across four dimensions: category (bug, feature request, UX issue, performance, etc.), severity (business impact scoring), user segment, and frequency signal. It doesn't just sort — it scores each item for product actionability, so you can quickly see which feedback should influence your next sprint versus your next quarter.

The consistency is the real value. Run 500 tickets through this prompt and you get the same classification logic applied to every single one. No mood bias. No recency effect. No accidentally over-weighting that one angry enterprise customer who emails you directly.

When to Reach for This

• You have a backlog of unprocessed feedback across multiple channels and need a starting taxonomy
• You're preparing for quarterly planning and want data-backed evidence of the most common user pain points
• Your team argues about priorities and you need an objective classification layer before the debate
• You just launched a feature and want to rapidly categorize the early response signal
• You're building a case for a specific investment and need to quantify how much feedback supports it

What Good Looks Like

A strong output produces a categorized table with clear severity scores, a frequency analysis showing which issues come up most often, and a "product signal" section that highlights patterns the raw data makes hard to see — like "37% of churn-risk accounts mention the same onboarding step." The best outputs also flag classification edge cases where human judgment is needed.

Sources

• The State of Product Management 2024 — Productboard
• Bringing the Voice of the Customer into Product — Brian Freyburger
• How to Listen to Customers — Ken Norton