What it does
Feature Request Deduper and Scorer solves the everyday Product Ops problem of the same request arriving in fourteen different tickets, and the prioritisation debate that follows because nobody can say how many customers actually asked. You paste the incoming requests and, optionally, your existing backlog. The agent clusters near-duplicates together, RICE or ICE scores each merged cluster, sums the ARR of the requesting customers across unique accounts, and flags which clusters overlap something already on your backlog. What comes back is a scored priority list rather than a pile. The part that should decide whether you trust it is how it handles missing inputs. RICE needs reach, impact, confidence and effort, and real request data rarely has them. Rather than quietly defaulting and presenting a confident score, every estimated input is recorded as a visible assumption on its cluster, and `list_assumptions` returns them as their own answer. A score whose inputs you cannot inspect is a score you cannot defend in a roadmap meeting. `analyze` goes further and health-checks the batch before ranking anything: which RICE inputs are real versus defaulted, how much of the pile is repeat tickets from one account, zero-effort rows, and ARR that cannot be attributed. It is the honest first question — is this data even scorable — and most tools skip it. Every other step is callable alone too. `cluster_requests` does the dedupe. `score_clusters` does the scoring. `sum_customer_arr` answers how much revenue is behind each ask. `match_backlog` finds the overlaps. `run_full` does the lot. Clustering uses a deterministic offline token-similarity function, so the same input always produces the same clusters and nothing is sent anywhere. You can supply your own synonym groups — billing, invoicing, subscription charges — so the clustering understands your product vocabulary. The language model only narrates figures the analyzer already computed. No connectors and no credentials. Built for product managers and product ops teams.
Example prompts
- Here are this week's feature requests — dedupe and score them
- How many distinct customers actually asked for SSO, and how much ARR?
- Which of these requests are already on our backlog?
