What is session momentum in analytics?

Session momentum measures whether a visitor's engagement is accelerating or decelerating within a session. Increasing momentum — faster page transitions, deeper scrolling, more interactions — signals growing interest, while decreasing momentum signals disengagement.

What does click entropy measure?

Click entropy quantifies how random or focused a visitor's click behavior is. Low entropy indicates purposeful navigation toward a goal. High entropy indicates exploratory or confused behavior. It helps distinguish between visitors who know what they want and those who need guidance.

How is attention score calculated?

Attention score combines active engagement signals — scroll depth, mouse movement over content areas, time spent on visible content, and interaction with page elements — to measure genuine content consumption rather than simple page load or time-on-page metrics.

Session Momentum, Click Entropy, and Attention Score

What You'll See in the Dashboard

Open the Intelligence tab to find the Session Momentum, Click Entropy, and Attention Score cards. Momentum shows whether the visitor is accelerating (+) or decelerating (−) through your funnel. Click Entropy displays a 0–100 disorder score — low means purposeful clicking, high means erratic. Attention Score shows real-time focus intensity based on dwell patterns and interaction depth.

Business Actions: Set up a Rule to surface a contextual CTA when Session Momentum exceeds 75 (visitor is on a hot streak). Flag sessions with Click Entropy above 80 for UX review — those visitors are lost. Use Attention Score to identify your most-read content and double down on what works.

Model 17: Session Momentum

Session momentum measures the velocity and acceleration of a visitor's progression through your site. Unlike simple page-per-minute metrics, momentum captures the direction of movement: is the user moving purposefully toward a goal, or are they drifting aimlessly?

A high-momentum session typically follows a clear trajectory — landing page to category to product to cart. A low-momentum session meanders, backtracks, and stalls. By quantifying this in real time, you can identify the exact moment a session starts to lose steam and intervene.

The 7 Momentum Signals

Signal	Weight	Description
Funnel progression rate	0.25	Speed at which the visitor advances through defined funnel stages
Page-to-page transition speed	0.18	Average time between page loads (normalized for content length)
Forward navigation ratio	0.15	Ratio of forward clicks (deeper pages) to back-button usage
Engagement acceleration	0.14	Whether engagement score is increasing page-over-page
Search-to-click efficiency	0.10	How quickly the visitor finds and clicks what they searched for
Scroll velocity consistency	0.10	Steady scroll pace vs. erratic stop-start patterns
Session recency boost	0.08	Returning within 24 hours of a previous session gets a momentum bonus

Momentum Categories

The raw momentum score (0–100) maps to five categories that guide real-time action:

Score Range	Category	Pattern	Recommended Action
0–15	Stalled	No forward progression, idle or stuck	Proactive help widget, navigation suggestions
16–35	Drifting	Slow, aimless browsing with no clear direction	Content recommendations, guided pathways
36–55	Steady	Consistent pace, moderate funnel progress	Reinforce with social proof, related content
56–80	Accelerating	Rapid funnel advancement, purpose-driven	Clear the path, reduce distractions
81–100	Surging	Fast, decisive movement toward conversion	Minimize friction, show trust signals at checkout

Momentum Decay

Momentum decays exponentially when a visitor goes idle. After 30 seconds of inactivity, the score begins dropping at a rate of 5 points per 10 seconds. This prevents stale sessions from carrying artificially high momentum scores. When the visitor resumes activity, momentum recalculates from current behavioral signals rather than jumping back to pre-idle levels.

Under the Hood: Momentum Calculation

function calculateMomentum(session: SessionState): number {
  const funnelRate = session.currentFunnelStage / session.totalFunnelStages;
  const transitionSpeed = normalizeTransitionTime(session.avgPageTransitionMs);
  const forwardRatio = session.forwardNavigations /
    Math.max(session.forwardNavigations + session.backNavigations, 1);
  const engagementAccel = session.engagementSlope;
  const searchEfficiency = session.searchClickTime > 0
    ? Math.max(0, 1 - (session.searchClickTime / 30000)) : 0.5;
  const scrollConsistency = 1 - session.scrollVelocityVariance;
  const recencyBoost = session.hoursSinceLastVisit < 24
    ? 1 - (session.hoursSinceLastVisit / 24) : 0;

  let score = (funnelRate * 25) + (transitionSpeed * 18) +
    (forwardRatio * 15) + (engagementAccel * 14) +
    (searchEfficiency * 10) + (scrollConsistency * 10) +
    (recencyBoost * 8);

  // Apply idle decay
  const idleSeconds = (Date.now() - session.lastInteractionTs) / 1000;
  if (idleSeconds > 30) {
    score *= Math.exp(-0.05 * (idleSeconds - 30) / 10);
  }

  return Math.min(100, Math.max(0, Math.round(score)));
}

Model 18: Click Entropy

Click entropy borrows from information theory to measure the randomness or disorder of a visitor's click patterns. A visitor who clicks in a logical, predictable sequence (navigation → category → product → cart) produces low entropy. A visitor who clicks erratically across unrelated elements produces high entropy.

High click entropy is a strong signal of confusion, disorientation, or bot-like behavior. It complements the confusion model (Model 6) by focusing specifically on click-target distribution rather than broader behavioral signals.

How Entropy Is Calculated

ClickStream computes Shannon entropy over the distribution of click targets within a rolling window. Each unique element clicked is a category in the probability distribution. The formula produces a value between 0 (perfectly predictable — every click on the same element) and log₂(n) (maximum disorder — clicks uniformly distributed across n elements).

This raw entropy value is then normalized to a 0–100 scale relative to the expected entropy for the page type. A product listing page naturally has higher entropy than a checkout form, so the model adjusts for context.

The 6 Entropy Signals

Signal	Weight	Description
Click target diversity	0.30	Shannon entropy of clicked elements (normalized by page type)
Click sequence predictability	0.20	Markov chain transition probability between click targets
Spatial click dispersion	0.18	Standard deviation of click coordinates on the viewport
Temporal click regularity	0.12	Variance of inter-click intervals (bots produce very regular timing)
Click-to-content relevance	0.12	Whether clicked elements are semantically related to recent views
Dead-click ratio	0.08	Proportion of clicks on non-interactive elements

Entropy Interpretation

Score Range	Category	What It Means	Action
0–20	Laser-focused	Extremely predictable clicks, single-purpose visit	Clear conversion path
21–40	Purposeful	Logical click flow with occasional exploration	Standard experience
41–60	Exploratory	Browsing broadly, trying different areas	Content recommendations
61–80	Disoriented	Scattered clicks, likely struggling to find something	Search suggestions, help widget
81–100	Chaotic / Suspicious	Extremely random or bot-like click patterns	Bot check, UX investigation

Under the Hood: Shannon Entropy Calculation

function calculateClickEntropy(clicks: ClickEvent[]): number {
  const targetCounts = new Map<string, number>();
  for (const click of clicks) {
    const target = normalizeSelector(click.targetSelector);
    targetCounts.set(target, (targetCounts.get(target) || 0) + 1);
  }

  const total = clicks.length;
  let entropy = 0;
  for (const count of targetCounts.values()) {
    const p = count / total;
    entropy -= p * Math.log2(p);
  }

  // Normalize to 0-100 relative to max possible entropy
  const maxEntropy = Math.log2(targetCounts.size);
  const normalized = maxEntropy > 0 ? (entropy / maxEntropy) * 100 : 0;

  return Math.round(normalized);
}

Model 19: Attention Score

The attention score measures how deeply a visitor is concentrating on your content at any given moment. It goes beyond engagement (Model 3) by focusing on focus intensity rather than interaction breadth. A visitor can be highly engaged (clicking many things) but poorly attentive (skimming quickly). Attention captures whether they are actually absorbing what you are presenting.

The 8 Attention Signals

Signal	Weight	Description
Reading pace	0.22	Scroll speed calibrated to content density (words per viewport). Slow, steady scrolling = reading.
Pause frequency	0.18	Number and duration of scroll pauses on content sections (not ads, not navigation)
Tab focus duration	0.15	Continuous time with tab visible and active (no alt-tabs)
Mouse tracking content	0.12	Mouse position following text flow (left-to-right, top-to-bottom sweep)
Text selection events	0.10	Selecting text to copy, highlight, or re-read indicates deep attention
Viewport stability	0.08	Low scroll jitter — the viewport stays stable while the visitor reads
Return-to-section	0.08	Scrolling back up to re-read a previous section (high-signal attention)
Interaction delay after content	0.07	Time between finishing content and next action (longer = processing/thinking)

Attention vs. Engagement

These two scores are related but capture different dimensions of user behavior:

Dimension	Engagement (Model 3)	Attention (Model 19)
What it measures	Breadth and variety of interaction	Depth and focus of content consumption
High score means	Clicking, scrolling, navigating actively	Reading carefully, pausing to think, re-reading
Low score means	Passive or minimal interaction	Skimming, distracted, multi-tasking
Best use case	E-commerce, product exploration	Content sites, documentation, long-form articles

The Four Attention Archetypes

1. The Deep Reader

High Attention + High Engagement

Thoroughly consuming content and interacting with it. Your ideal audience for long-form content, documentation, and educational material. Action: serve more depth — related articles, downloadable guides, newsletter signup.

2. The Speed Scanner

Low Attention + High Engagement

Clicking around actively but not reading deeply. Looking for a specific answer or comparing options quickly. Action: improve scannability — better headings, summary boxes, table of contents.

3. The Passive Absorber

High Attention + Low Engagement

Reading carefully but not clicking or interacting. May be on mobile, may be a first-time visitor evaluating quality. Action: gentle engagement prompts — inline polls, expandable sections, subtle CTAs.

4. The Distracted Visitor

Low Attention + Low Engagement

Neither reading nor interacting meaningfully. Background tab, arrived accidentally, or lost interest. Action: re-engagement nudge or accept natural exit.

How Momentum, Entropy, and Attention Interact

These three models form a diagnostic triad that reveals session quality from complementary angles:

Combination	Interpretation	Action
High Momentum + Low Entropy + High Attention	Ideal session: focused, purposeful, absorbing content	Clear path to conversion
High Momentum + High Entropy + Low Attention	Bot-like: fast but random, not reading	Bot verification challenge
Low Momentum + Low Entropy + High Attention	Deep researcher: slow but focused, studying one area	Provide depth, comparison tools
Low Momentum + High Entropy + Low Attention	Completely lost: stuck, confused, clicking randomly	Proactive help, exit survey

Session momentum tells you the speed, click entropy tells you the order, and attention tells you the depth. Together, they give you a three-dimensional view of session quality that no single metric can provide.

Configuration & Tuning

All three models are configurable through the ClickStream dashboard:

Momentum decay rate: Adjust how quickly idle time erodes momentum (default: 5 points per 10 seconds after 30s idle).
Entropy normalization: Set expected entropy baselines per page template (listing vs. article vs. checkout).
Attention reading speed: Calibrate expected reading pace for your audience (default: 200–250 words per minute).
Cross-model alerts: Create compound rules like "Momentum > 70 AND Entropy < 30" to trigger personalization.