Analytics & AI evals

Know what your knowledge base is missing

Kelu analytics show which questions succeed, which ones expose knowledge gaps, and how your AI answers score on automated groundedness evals — so your knowledge base keeps getting better with every interaction.

Six dimensions of knowledge health

From raw question volume to automated AI quality scoring

Top Questions

See what users ask most. Sort by volume, failure rate, or feedback score. Understand the exact language your users use to find things.

Unanswered Questions

Questions Kelu could not answer with confidence are flagged as knowledge gaps. These are your highest-priority knowledge base improvements.

Feedback Scores

Users can rate any answer with a thumbs up or down. Negative feedback is surfaced with the original question and the answer that failed.

Daily Trends

Track conversation volume, feedback ratio, and unanswered rate day over day. Spot the impact of a docs update or a new product release instantly.

Latency Metrics

P50, P95, and P99 response times for retrieval and generation. Broken down by knowledge base, source type, and time period.

Groundedness Evals

Automated LLM-as-judge evaluation that scores each answer for faithfulness to retrieved sources. Catch hallucinations before users do.

Citation Quality

Evaluates whether cited sources actually support the answer. Low citation quality scores indicate retrieval returning irrelevant chunks.

AI evals

Automated quality scoring on every answer

Kelu runs LLM-as-judge evals asynchronously after each response. Groundedness score measures whether the answer stays within the retrieved context. Citation quality score measures whether the cited sources actually support each claim. Both scores trend over time so you can see the impact of docs improvements.

  • Groundedness: answer vs. retrieved context faithfulness
  • Citation quality: cited sources vs. answer claims
  • Completeness: does the answer address the full question
  • Eval scores exportable via API for custom dashboards
ANALYTICS OVERVIEW — LAST 30 DAYS
4,821
Questions asked
92%
Answered
387
Gaps found
Groundedness score94%
Citation quality89%
User satisfaction87%
What teams say

Deployed in production, cited by the buyers who chose it

Deployed on our docs site in an afternoon. Every answer has citations, and the abstention gate means we've never had a customer complain about a made-up answer.
PN
Priya Nair
Head of Customer Support · Supabase
The knowledge base connected to our Slack, Confluence, and helpdesk in one setup. On-call teams get the same cited answer whether they ask in chat, in the widget, or from Cursor.
TR
Tom Richter
IT Operations Manager · Grafana Labs
The gap analytics turned into a real docs backlog. Deflection went up because we finally knew which pages were missing — the AI told us.
AC
Ana Castillo
VP of Customer Experience · Clerk

Frequently Asked Questions

Common questions about Kelu Analytics

Free plan retains 30 days of analytics. Pro retains 12 months. Enterprise has configurable retention up to unlimited.
Yes. All analytics are accessible via the Analytics API as JSON. You can pipe data into your own BI tools, Datadog, or Grafana dashboards.
Questions where Kelu returned a low-confidence answer or explicitly said it did not know. These are the questions your knowledge base fails to address — the most actionable signal for improving your content.
Evals use a separate, specialized judge LLM that is prompted to evaluate faithfulness to context rather than general accuracy. Scores are calibrated and normalized so you get a consistent 0–100 score per dimension.
Yes. Configure webhook alerts when the groundedness score drops below a threshold, or when the unanswered question rate exceeds a defined percentage. Alerts support Slack, email, and PagerDuty.

Start learning from your questions

Analytics are available on all plans from day one.

Get Started Free