Product Review-to-Roadmap Extractor

Convert Amazon, Google Maps, Trustpilot, G2, app-store, survey, or support reviews into a ranked queue of defects, feature requests, and objections with traceable excerpts and roadmap-ready tickets. No AI key or external enrichment service is required.

What you get

The default dataset contains:

• review-analysis — one successfully analyzed unique actionable review with theme, type, secondary themes, severity, recency, confidence, source context, matched phrases, and excerpt. This is the primary pay-per-event unit.
• roadmap-ticket — an uncharged aggregate ranked using transparent frequency, severity, recency, and confidence components. Each ticket includes evidence, affected versions, a problem statement, recommended action, acceptance criteria, and validation metric.

The OUTPUT key-value-store record is an uncharged summary of invalid, duplicate, non-actionable, processed, budget-limited, and ticket-type counts.

Quick start

Run with no input for a useful built-in sample, or paste records:

{
  "reviews": [
    {
      "id": "r-1",
      "text": "Please add CSV export and scheduled reports.",
      "rating": 3,
      "productName": "Northstar Analytics",
      "date": "2026-08-01"
    },
    {
      "id": "r-2",
      "text": "The app crashes whenever I apply more than three filters.",
      "rating": 1,
      "productName": "Northstar Analytics",
      "version": "6.4.0"
    }
  ]
}

Choose an upstream dataset for production. Common camelCase, snake_case, nested author, timestamp, and rating formats are detected automatically; fieldMap supports explicit dot-separated paths.

Deterministic analysis

1. Normalize text, ID, rating, date, version, platform, product name, locale, author, URL, and helpful count.
2. Reject textless rows and remove exact normalized duplicates.
3. Detect actionable defects, requests, and objections using a versioned domain taxonomy. Praise without an actionable signal is skipped.
4. Preserve the matched phrases and a short source excerpt.
5. Rank per-product themes using frequency (40%), severity (30%), recency (20%), and confidence (10%).
6. Generate an evidence-linked triage ticket with validation-oriented acceptance criteria.

The same input and settings produce the same classification. The Actor does not call an LLM, transmit review text to an external model, or invent quotations.

Charging and budgets

The primary value event is actionable-review-analyzed. Invalid, exact-duplicate, non-actionable, and failed rows are not charged as analyzed reviews. Roadmap-ticket rows and the OUTPUT summary are uncharged. The Actor checks the maximum-charge budget before each paid batch and stops cleanly if no event capacity remains.

A one-time $0.0003 Actor-start event covers measured startup overhead, including invalid-input runs. Set an Apify maximum total charge for unfamiliar inputs.

Privacy and decision boundary

includeOriginalReview defaults to false; canonical fields are included by default. Review data may contain personal information—retain only the fields your purpose and policy permit.

Generated tickets are triage inputs, not customer commitments or proof of demand. Keyword methods can miss context, sarcasm, and unsupported languages. Validate roadmap choices with representative users, product strategy, effort, retention, support, and behavioral telemetry.

Development

npm ci
npm run typecheck
npm test
npm run build
npm run benchmark
npm run launch:verify

See launch/OPERATIONS.md for release checks, maintenance, privacy, and rollback.