How does hover intent analysis work?

Hover intent analysis tracks mouse movement patterns over interactive elements like CTAs, product cards, and pricing sections. Prolonged hover time, approach-retreat patterns, and hover-to-click ratios reveal purchase consideration signals that traditional click analytics miss entirely.

What is scroll depth intelligence?

Scroll depth intelligence goes beyond basic scroll percentage by measuring scroll velocity, pause points, reverse scrolling, and content visibility duration. It identifies which page sections capture attention and which get skipped, enabling data-driven content optimization.

Can hover data predict conversions?

Yes. Visitors who hover over pricing elements, CTA buttons, and product details for extended periods convert at significantly higher rates. ClickStream's conversion probability model incorporates hover intent signals to score purchase readiness in real time.

Conversion Probability, Hover Intent, and Scroll Depth Intelligence

What You'll See in the Dashboard

Open the Intelligence tab to find the Conversion Probability, Hover Intent, and Scroll Depth Intelligence cards. Conversion Probability shows a real-time percentage with a confidence interval (e.g., "72% ± 8%"). Hover Intent highlights which products or CTAs the visitor is most interested in based on hover dwell time. Scroll Intelligence shows a visual heatmap of where this visitor paused, re-read, and skipped.

Business Actions: Create a Rule to trigger a limited-time offer when Conversion Probability exceeds 65%. Use Hover Intent data to personalize product recommendations in real time. Feed Scroll Intelligence into your content team to identify which sections lose readers.

Model 20: Conversion Probability

The conversion probability model produces a single number that answers the most important question in real-time analytics: What is the probability that this visitor will convert during this session?

Unlike the intent score (Model 1), which classifies motivation level, conversion probability is a calibrated probability estimate. An output of 0.72 means that, historically, 72% of visitors exhibiting this behavioral pattern completed a conversion. This calibration is critical for downstream decision-making — you can set thresholds based on actual expected conversion rates rather than arbitrary score ranges.

The 9 Conversion Signals

Signal	Weight	Description
Funnel stage reached	0.22	Deepest conversion funnel stage visited (browse → product → cart → checkout → payment)
Historical conversion rate	0.16	This visitor's personal conversion rate from prior sessions
Session intent score	0.14	Current intent model output (Model 1), providing behavioral context
Time in conversion zone	0.12	Active time spent on checkout, pricing, or payment pages
Cart value momentum	0.10	Whether the cart value is increasing (adding items) or decreasing (removing)
Form completion progress	0.08	Percentage of required checkout/signup form fields completed
Device & channel context	0.07	Conversion rates vary by device type, traffic source, and time of day
Frustration dampening	0.06	Current frustration score inversely affects conversion probability
Social proof exposure	0.05	Whether the visitor has seen reviews, testimonials, or trust badges

Calibration Method

Raw model output is calibrated using Platt scaling against historical conversion data. This ensures that when the model says "70% probability," approximately 70% of visitors with that score actually convert. ClickStream recalibrates weekly using a rolling 30-day window of conversion outcomes.

Probability Tiers

Probability	Tier	What It Means	Action
0–15%	Cold	Very unlikely to convert this session	Content nurturing, email capture
16–35%	Warming	Some conversion signals present	Social proof, case studies
36–55%	Considering	Active evaluation, could go either way	Urgency messaging, limited offers
56–80%	Likely	Strong conversion trajectory	Remove friction, streamline checkout
81–100%	Imminent	Conversion highly probable	Upsell/cross-sell, order bump

Under the Hood: Probability Calibration

function calibrateProbability(rawScore: number, params: PlattParams): number {
  // Platt scaling: P(y=1|f) = 1 / (1 + exp(A*f + B))
  const calibrated = 1 / (1 + Math.exp(params.A * rawScore + params.B));
  return Math.round(calibrated * 100);
}

// Recalibrate weekly using rolling 30-day conversion data
function fitPlattParams(predictions: number[], outcomes: boolean[]): PlattParams {
  // Gradient descent to minimize log-loss between
  // calibrated predictions and actual outcomes
  let A = -1, B = 0;
  const lr = 0.001;

  for (let epoch = 0; epoch < 1000; epoch++) {
    let gradA = 0, gradB = 0;
    for (let i = 0; i < predictions.length; i++) {
      const p = 1 / (1 + Math.exp(A * predictions[i] + B));
      const y = outcomes[i] ? 1 : 0;
      gradA += (p - y) * predictions[i];
      gradB += (p - y);
    }
    A -= lr * gradA / predictions.length;
    B -= lr * gradB / predictions.length;
  }

  return { A, B };
}

Model 21: Hover Intent

Hover intent analyzes mouse hover behavior to infer what a visitor is interested in before they click. Research shows that users hover over elements they are considering 300–800ms before deciding whether to click. By capturing this pre-click deliberation phase, ClickStream can identify interest even when no click occurs.

The 7 Hover Signals

Signal	Weight	Description
Hover dwell time	0.25	Time cursor rests on an interactive element (300ms+ = intentional)
Hover-to-click ratio	0.20	Proportion of hovered elements that are subsequently clicked
Hover trajectory	0.15	Direct approach vs. meandering path to the element
Repeated hovers	0.12	Returning to hover over the same element multiple times (deliberation)
Hover abandonment	0.10	Hovering over a CTA then moving away without clicking (hesitation)
Hover sequence	0.10	Order in which elements are hovered (comparison shopping pattern)
Hover velocity	0.08	Speed of cursor movement over elements (fast = scanning, slow = reading)

Hover Intent Categories

Pattern	Description	What It Reveals
Deliberate hover	Cursor moves directly to element, dwells 500ms+, then clicks	Strong interest, decided user
Comparison hover	Cursor alternates between 2–3 elements in sequence	Active comparison, evaluating options
Hesitation hover	Repeated hover-and-retreat on a CTA or "Buy" button	Interest but uncertainty — needs reassurance
Scanning hover	Fast cursor sweep across many elements without dwelling	Looking for something specific
Ghost hover	No hover activity on interactive elements (mobile or keyboard nav)	Different interaction mode, fall back to other signals

Mobile Considerations

Hover intent is primarily a desktop signal. On mobile devices, ClickStream substitutes with touch hesitation — the time between the finger approaching the screen (via touch-start proximity events on supported devices) and the actual tap. Long press events on product images also serve as a mobile analog to deliberate hovers.

Model 22: Scroll Depth Intelligence

Scroll depth intelligence goes far beyond "the user scrolled 73% of the page." It analyzes how the visitor scrolled — the rhythm, pauses, re-reads, and velocity changes — to extract rich behavioral signals about content engagement and interest.

The 8 Scroll Intelligence Signals

Signal	Weight	Description
Maximum depth reached	0.20	Deepest scroll position as a percentage of total page height
Content completion rate	0.18	Percentage of content sections the viewport has paused on (300ms+ per section)
Scroll velocity profile	0.15	Speed pattern: fast-skip vs. slow-read sections mapped to content zones
Pause-to-read ratio	0.14	Time spent paused vs. actively scrolling (high ratio = careful reading)
Re-scroll events	0.12	Scrolling back up to re-read a previously viewed section
Fold interaction	0.08	Whether the visitor scrolled past the initial viewport fold (and how quickly)
Scroll-to-action correlation	0.07	Whether scrolling to a CTA section correlates with interaction
Content zone dwell time	0.06	Time spent in each content zone (hero, body, pricing, footer)

Scroll Patterns and What They Reveal

Pattern	Scroll Behavior	Interpretation	Action
The Reader	Slow, steady scroll with regular pauses	Carefully consuming content top-to-bottom	Serve related content, newsletter prompt
The Scanner	Fast scroll with brief pauses at headings	Looking for specific information	Improve headings, add table of contents
The Bottom-Liner	Fast scroll to bottom, then slow scroll up	Checking conclusion first, then reading details	Put key info in summary/conclusion
The Bouncer	Scrolls to fold, stops, leaves	Above-the-fold content did not convince them to continue	Optimize hero section, value proposition
The Re-Reader	Multiple scroll-up events to specific sections	Complex content requiring re-reading, or comparison with later content	Add anchored navigation, expandable details

Under the Hood: Scroll Zone Analysis

interface ScrollZone {
  startPercent: number;
  endPercent: number;
  dwellMs: number;
  scrollVelocity: number;  // px/sec within zone
  revisits: number;
  contentType: 'hero' | 'body' | 'pricing' | 'cta' | 'footer';
}

function analyzeScrollDepth(zones: ScrollZone[]): ScrollIntelligence {
  const maxDepth = Math.max(...zones.map(z => z.endPercent));
  const totalDwell = zones.reduce((sum, z) => sum + z.dwellMs, 0);
  const totalScroll = zones.reduce((sum, z) =>
    sum + Math.abs(z.endPercent - z.startPercent), 0);

  const pauseRatio = totalDwell / (totalDwell + totalScroll);
  const reReadCount = zones.filter(z => z.revisits > 0).length;

  // Content completion: sections with 300ms+ dwell
  const completedSections = zones.filter(z => z.dwellMs >= 300).length;
  const completionRate = completedSections / zones.length;

  return {
    maxDepth,
    completionRate: Math.round(completionRate * 100),
    pauseRatio: Math.round(pauseRatio * 100),
    reReadSections: reReadCount,
    pattern: classifyScrollPattern(zones),
    score: calculateScrollScore(maxDepth, completionRate, pauseRatio, reReadCount)
  };
}

How Conversion, Hover, and Scroll Interact

These three models provide overlapping but distinct views of purchase readiness:

Combination	Interpretation	Action
High Conversion + Deliberate Hover + Deep Scroll	Ready to buy: has read everything, hovering on Buy button	Minimize checkout friction, show trust badges
High Conversion + Hesitation Hover + Shallow Scroll	Wants to buy but needs reassurance (skipped product details)	Surface reviews, guarantees, key product info
Low Conversion + Comparison Hover + Re-Read Scroll	Deep research phase: comparing options carefully	Comparison tools, side-by-side features
Low Conversion + No Hover + Bouncer Scroll	Not engaged: landing page failed to hook them	A/B test hero section, improve value proposition

Conversion probability tells you how likely the visitor is to buy. Hover intent tells you what they are considering. Scroll intelligence tells you how deeply they have evaluated your offering. Together, they enable truly intelligent real-time personalization.

Configuration & Tuning

All three models expose tunable parameters in the dashboard:

Conversion probability recalibration: Set the rolling window (default: 30 days) and minimum sample size for Platt scaling updates.
Hover dwell threshold: Adjust the minimum hover duration that counts as intentional (default: 300ms).
Scroll zone definitions: Customize content zone boundaries per page template for accurate scroll analysis.
Cross-model triggers: Create compound rules (e.g., "Conversion > 60% AND Hover Hesitation detected") for targeted interventions.

Conversion Probability, Hover Intent, and Scroll Intelligence