Skip to content
AskVault Docs
Try Free →

Knowledge gaps detection

Written by , Founder, AskVault · Reviewed by Aashiq
Last updated: · 3 min read

What counts as a gap

Three patterns:

  1. Fallback fired. Bot replied "I don't have specific info on that".
  2. Low retrieval confidence. Top retrieved chunk's relevance score below threshold (default 0.7).
  3. Customer feedback. Thumbs-down on a bot reply with "this didn't answer my question".

Each logged as a knowledge gap.

Where to see gaps

Analytics > Unanswered Questions. Lists gaps with:

  • Original question.
  • Frequency (how many customers asked similar).
  • Channels affected.
  • Suggested action (add content, refine an existing source, train a Q&A pair).

About 60 to 80% of gaps are quick fixes; the rest need genuine new content.

Auto-clustering

Similar gap questions cluster:

  • "How do I cancel my plan?"
  • "I want to cancel."
  • "Stop my subscription please."

→ one cluster with 23 occurrences this week.

Click the cluster to see all variants. Address one Q&A pair; bot handles all 23.

Webhook on gap

Subscribe to knowledge.gap_detected:

{
  "event": "knowledge.gap_detected",
  "cluster_id": "gap_xxx",
  "question": "How do I cancel my plan?",
  "frequency": 23,
  "channels": ["widget", "whatsapp"],
  "first_seen": "2026-05-08T...",
  "suggested_action": "add_qa_pair"
}

Useful for routing gaps to your content team in Slack.

Fill workflow

  1. Pick the highest-frequency cluster.
  2. Decide: add a Q&A pair, edit an existing source, or crawl new content.
  3. Edit in Knowledge Hub.
  4. Mark gap as resolved.
  5. AskVault re-checks within 30 minutes.

Resolved gaps stop firing the webhook.

Priority signals

Gaps prioritized by:

  • Frequency. More asks = higher priority.
  • Channel impact. Widget on marketing site higher than internal Slack.
  • Visitor segment. Paid customers higher than anonymous.
  • Recency. New gaps surface above stale ones.

Limits

  • Gaps tracked per workspace. 100 at a time.
  • Cluster lifespan. 90 days before auto-archive.
  • Webhook frequency. Up to 1 per minute per cluster.

Sample impact

A team running for 90 days with 5,000 queries:

  • Day 1: 12% fallback rate.
  • Day 30: 8% (filled top 30 gaps).
  • Day 60: 5% (filled top 60 gaps).
  • Day 90: 4% (steady state).

Most gap-filling effort pays off in the first 60 days; long tail thereafter.

Common pitfalls

Filling the wrong gap. Focus on high-frequency clusters, not one-off questions.

Adding redundant content. Existing content may already cover; just retrieval is poor. Fix retrieval (add aliases, snippets).

Ignoring single-occurrence gaps. Sometimes a single high-stakes customer question deserves attention. Use visitor-segment filter.

FAQ

Does this work for voice channel?

Yes. Voice transcripts feed the same gap detector.

Can I export gaps to my project tracker?

Yes via webhook to Jira, Linear, Asana, etc.

Does the bot get smarter automatically?

No. Gaps surface; you add content. Future planned: auto-generation of draft Q&A pairs.

Was this page helpful?