Product feedback deduplication system with ARR-weighted prioritization
Build a Product Feedback Intelligence workflow that turns a stream of raw customer feedback messages into a de-duplicated, prioritized, and revenue-weighted dashboard for a product team. Purpose: product managers drop feedback into a team chat channel constantly, but it's noisy, repetitive, and impossible to prioritize. This workflow collects that feedback, groups semantically-identical requests together (counting how many distinct customers asked for each), sorts everything into themes, and shows it on a private internal dashboard ranked by demand — enriched with each request's total customer ARR so the team can weigh "how many asked" against "how much revenue is behind it." Triggers and flow: On an hourly schedule (cron), pull new messages from a feedback channel in Slack — or your team's chat tool. Use a checkpoint so each run only fetches new messages, and on an empty datastore, backfill a recent window (e.g. 60 days) so a fresh deployment self-populates. Pass new messages to an AI de-duplication step using Claude (or another LLM). It matches each message against existing feedback items (including reworded duplicates), either merging it into an existing item or creating a new categorized one. Count demand by distinct customer (email domain) so one chatty customer can't inflate a request. Process in chunks under a time budget and self-loop until the queue drains. Persist results cumulatively to shared storage. On a schedule (e.g. every few hours), sync each customer's deal size / ARR from your CRM (Salesforce, or your CRM) into a domain→ARR lookup cached in shared storage, so the dashboard can compute a "Touched ARR" figure per feedback item without hitting the CRM on every page load. Serve a private JSON API that reads the datastore and returns feedback paginated, filtered, and sorted server-side (keep the response small and fast), including live ingestion progress and per-item Touched ARR. Serve a private web dashboard (a polished single-page app) that reads that API: a ranked, sortable, paginated table of themed feedback with distinct-customer counts and Touched ARR, expandable rows showing who asked, theme filtering and search, plus a subscriptions view where users can opt into chat notifications for specific items or whole themes. When new feedback lands in a subscribed item or theme, post a notification to a chat channel (Slack, or your team's chat tool). Include lightweight admin actions (trigger a backfill, reset the datastore) behind private endpoints. Keep all dashboards and APIs private to your space. Store the datastore, work queue, subscriptions, and CRM cache in shared persistent storage, and make sure background writers that touch the same files are scheduled so they can't clobber each other. Tools used: Slack API, Anthropic Claude API, Salesforce API, 3B named volumes (persistent storage), 3B cron triggers, 3B HTTP routes
What this prompt builds
This workflow collects product feedback from Slack, deduplicates semantically identical requests using Claude, and groups them by distinct customer count and touched ARR pulled from Salesforce. It surfaces a ranked dashboard where product managers can filter by theme, subscribe to notifications on specific feedback items or categories, and prioritize features by both demand and revenue impact. The system runs on a scheduled cadence, self-populates historical messages on first deploy, and notifies subscribed users in Slack when relevant feedback arrives.
The problem
Product teams drown in unstructured customer feedback scattered across chat channels — duplicate requests pile up unnoticed, a vocal customer's single complaint can look like widespread demand, and PM's can't tell whether a feature idea represents ten customers or ten messages from one person. Without clear signal on who's asking and how much revenue those customers represent, teams waste cycles building the wrong things or miss high-value requests buried in noise.
Solution and impact
This workflow automatically de-duplicates feedback by semantic meaning (not exact wording), counts distinct customers behind each request, enriches every item with total ARR across requesters, and surfaces everything in a ranked dashboard sorted by demand and revenue impact. Product managers get a clean, prioritized view of what customers actually want — weighted by both frequency and deal size — and can subscribe to instant notifications when high-priority themes gather momentum, so the team ships features that move the business forward.