Skip to content

11. Analytics & Insights

Analytics show whether Klariton is working: what customers ask, how mature each touchpoint is, how AI crawlers discover your content, where source material is missing, how much AI usage costs, and whether SafeGuard or Test Center require action.

Klariton uses explicit states instead of inventing numbers. A metric is either real, waiting for data, coming soon, or intentionally undefined.

AreaQuestion it answersStatus
OverviewWhat is live, healthy, and waiting for action?LIVE
MaturityHow complete is each touchpoint?LIVE
Question volumeWhich personas ask what, and where?LIVE
Agentic ReachAre AI crawlers reading the touchpoints?LIVE
Material gapsWhich questions lack grounded material?LIVE
LLM budgetWhat do generation, chat, and embeddings cost?LIVE
BIQ performanceWhich BIQs help, convert, or need review?PARTIAL
ConversionsProduct-card impressions, clicks, and funnel signalsPARTIAL
Test CenterStress tests for the real chatLIVE
SafeGuard signalsDrift and compliance findings on published BIQsLIVE

For detailed operation of each analytics subpage, see Analytics Deep Dive.

The overview combines activation progress with operating signals:

  • live BIQ count,
  • connected personas,
  • indexed material,
  • end-customer question volume,
  • SafeGuard health,
  • stale translations after content changes,
  • Test Center status when a chat touchpoint is involved.

When an organization has little traffic, the page shows a collection state instead of empty charts. This is expected during onboarding.

Maturity scores each touchpoint from 0 to 100. It helps you decide what to fix before relying on analytics.

Factors include:

FactorWhat it measures
Persona substrateWhether personas are defined well enough for targeting.
BIQ coverageWhether each persona has enough curated questions.
Confidence mixWhether answers are backed by material and sources.
Material inventoryWhether enough source material is linked and current.
Lead actionWhether the touchpoint has a conversion action.
Agentic decisionWhether AI crawler visibility was explicitly configured.

Each missing factor produces a concrete recommendation with a link back to the relevant Studio area.

Question volume shows which touchpoints and personas receive questions. Use it to answer:

  • Which topics are customers actually asking about?
  • Which personas are under-served?
  • Which touchpoints create the most unanswered or low-confidence questions?
  • Which questions should be promoted to BIQs?

Questions without a detected persona are grouped separately. Questions without a touchpoint context are shown as organization-wide.

Agentic Reach measures crawler reads from AI systems such as ChatGPT, Perplexity, Gemini, Claude, and similar agents. It includes:

  • total bot reads,
  • daily time series,
  • provider mix,
  • reach per touchpoint,
  • reach per product,
  • live log of recent crawler interactions,
  • freshness indicators for stale touchpoints.

Crawler traffic often appears days after publishing. A “waiting for crawler” state usually means the touchpoint is live but has not been discovered yet.

Material gaps are created when Klariton repeatedly answers from weak or external knowledge instead of your own material. The gap view clusters related questions so you can add the missing source once and improve several answers at the same time.

Best practice: Treat material gaps as the backlog for your knowledge base. Add material first, then regenerate or promote BIQs.

The LLM budget view tracks real token usage and cost estimates. It separates:

  • daily budget and current usage,
  • usage by provider,
  • usage by operation,
  • usage by touchpoint where attributable,
  • budget caps configured under Settings.

If no daily cap is configured, Klariton shows that explicitly instead of hiding the missing limit.

Conversion analytics are still growing. Current signals can include product-card impressions, card clicks, chat questions, product recommendations, and lead creation. Some funnel stages are proxies until full end-to-end attribution is available.

Use these metrics as directional signals:

  • repeated question plus low conversion means the answer may not be helpful enough,
  • product-card clicks show recommendation interest,
  • negative feedback should feed Test Center and BIQ review,
  • lead creation should be interpreted together with touchpoint and persona context.

The analytics section also exposes operating controls:

  • Test Center runs stress tests against the real chat and shows pass/fail results.
  • SafeGuard flags published BIQs that contradict material, lack support, or violate Brand Voice rules.

These are covered in detail in Test Center and SafeGuard.