KB Gap Finder

The problem this kills

Your support team answers the same questions over and over, but you can never quite prove which missing articles would actually move the needle. The knowledge base feels both bloated and incomplete at the same time. You skim tickets, you guess, you write a few articles on a hunch - and the deflection numbers barely budge.

The real signal is buried in your ticket history: which recurring questions have no good answer in the KB right now. Finding that by hand means reading hundreds of tickets and holding the whole KB in your head. Nobody has time for that, so the content backlog gets written by gut feel instead of by volume.

There's a second trap, too. Some questions do have an article - agents just can't find it. That's a search and tagging problem, not a content gap, and writing a duplicate article makes the KB worse. A good gap finder has to tell those two cases apart.

What you'll build

A small, private web app for knowledge managers and support leads. You feed it two things - a recent ticket export and your current KB article list - and it does the heavy reading for you:

• Clusters tickets into recurring themes (e.g. "how to change billing date", "reset 2FA on a lost phone").
• For each theme, checks whether a matching KB article already exists.
• Separates true gaps (no article) from findability problems (article exists, agents/customers aren't reaching it).
• Ranks the true gaps by volume (how many tickets) and estimated effort to write, so you get a prioritized "articles to write" list.
• Lets you review and approve which gaps become backlog items, assign an owner to each, and then export or email the backlog to your authors.

What's inside the Implementation Plan

A complete, paste-and-go runbook for Claude Code. Crucially, the plan opens by interviewing you about your business - your ticketing system, the exact columns in your export, how your KB is structured, your typical and peak ticket volumes, and your own rules for what counts as a "good" article. It reflects a short tailored spec back to you and waits for your thumbs-up before it builds anything. You get a tool shaped around your data and naming, not a generic template you have to bend to fit.

Inside you'll find:

• The discovery interview (8-12 targeted questions) and how the agent turns your answers into the data model.
• Step-by-step build instructions, each ending with a ready-to-copy prompt.
• The clustering + gap-detection logic, including the "gap vs. findability" split.
• The volume/effort ranking and the review-and-approve screen.
• Owner assignment, export, and the email-to-authors handoff.
• A "No API yet?" fallback so you can build the whole thing today from a CSV/Sheet, with a clean CSV export of the ranked backlog.

The governance it includes (this is the point)

This isn't a toy script - it's an internal tool you can trust with real data:

• Login so only your team can use it.
• Row-level security so people only ever see their own organization's data.
• A complete audit trail - who reviewed, approved, reassigned, or exported each gap, and when.
• A hard human-in-the-loop gate - the AI only drafts suggested gaps and titles. Nothing becomes a backlog item until the knowledge manager reviews it and approves.
• Duplicate guards - a dedupe key on the gap theme so the same gap can't be approved twice across runs.

Who it's for

Knowledge managers and support leads who plan content and want their backlog driven by evidence instead of gut feel. If you can download a ticket export and a list of your KB articles, you can build this - no coding background required.

You've got this - paste the first prompt and let the agent interview you.