How does purchase timing prediction work?

Purchase timing prediction analyzes behavioral cadence — visit frequency patterns, time spent on product pages across sessions, comparison shopping intensity, and historical conversion timelines — to predict when a visitor is most likely to complete a purchase.

What causes cart abandonment and how can analytics help?

Cart abandonment is driven by unexpected costs, complex checkout flows, comparison shopping, and decision fatigue. Behavioral analytics detects abandonment intent before it happens by identifying hesitation patterns, enabling real-time interventions like simplified checkout or targeted offers.

Can you predict when a visitor will buy?

Yes. By analyzing behavioral patterns across multiple sessions — increasing product page engagement, narrowing comparison sets, and return visit acceleration — ClickStream's purchase timing model estimates conversion probability windows, enabling precisely timed marketing outreach.

Abandonment Detection and Purchase Timing

What You'll See in the Dashboard

Open the Intelligence tab to find the Abandonment Detection and Purchase Timing cards. Abandonment shows a real-time probability that the current visitor is about to leave without converting — watch it spike as exit signals accumulate. Purchase Timing shows how close each visitor is to converting, with an urgency classification to guide your outreach.

Business Actions: Set up a Rule to fire an exit-intent overlay when Abandonment Detection exceeds 80. Use Purchase Timing scores to time retargeting and lifecycle sends in your ESP. Feed abandonment signals into your cart-recovery automation.

Model 12: Abandonment Detection

The abandonment model detects when a user is about to leave without completing their intended action -- whether that is a cart checkout, a form submission, or simply engaging with content. Unlike post-hoc abandonment analysis, ClickStream detects abandonment signals in real time, giving you a window to intervene.

The 7 Abandonment Signals

The signals below illustrate what the model watches; how each is weighted is tunable per site.

Signal	Detection Method
Mouse drift to browser chrome	Cursor moving toward close/back buttons or address bar
Idle timeout	No interaction for 30+ seconds on an active page (cart, form)
Tab switch away	Page visibility hidden while in a conversion flow
Scroll to top on cart/checkout	Scrolling back to top of checkout page (looking to navigate away)
Form abandonment cues	Stopping mid-form after completing 30%+ of fields
Price shock indicators	Viewing total/shipping cost then pausing >10 seconds
Engagement collapse	Sharp drop in mouse movement and interaction rate

The 4 Abandonment Stages

The model emits a stage alongside the score. The stage names, score bands, and intervention windows below are our operating framework for reading that output:

Score Range	Stage	User State	Intervention Window
0–25	Engaged	Actively interacting. No abandonment signals.	N/A -- user is engaged.
26–50	Wavering	Slowing down, some hesitation signals.	15–30 seconds. Soft nudge opportunity.
51–75	Disengaging	Clear disengagement pattern. Preparing to leave.	5–15 seconds. Immediate intervention needed.
76–100	Exiting	Active exit behavior. Mouse at browser chrome.	1–5 seconds. Last-chance intervention only.

Exit Intent Formula

ClickStream's exit intent detection goes beyond simple "mouse near top of viewport" heuristics. Conceptually, it combines velocity, trajectory, and context into a composite score — shown here in simplified, illustrative form:

TypeScript (illustrative)

function calculateExitIntent(mouseEvents: MouseEvent[], context: PageContext): number {
  const recent = mouseEvents.slice(-10); // last 10 mouse positions
  if (recent.length < 3) return 0;

  // Calculate upward velocity (toward browser chrome)
  const yVelocity = (recent[0].y - recent[recent.length - 1].y)
    / (recent[recent.length - 1].timestamp - recent[0].timestamp);

  // Position factor: how close to viewport top
  const positionFactor = Math.max(0, 1 - recent[recent.length - 1].y / 100);

  // Trajectory: is cursor heading toward close/back buttons?
  const trajectory = calculateTrajectoryToTarget(recent, context.closeButtonPos);

  // Context factor: higher weight if user has items in cart
  const contextWeight = context.hasCartItems ? 1.5 : 1.0;

  const rawScore = (
    yVelocity * 0.35 +
    positionFactor * 0.30 +
    trajectory * 0.35
  ) * contextWeight;

  // rawScore is a 0–1 composite; scale to the 0–100 score range
  return Math.min(100, Math.max(0, rawScore * 100));
}

Timing Pipeline Latency

Two latency numbers govern this pipeline. The behavioral scoring pass inside the collector holds a CI-enforced p95 under 3 ms per event, and the documented Signals read round trip is roughly 50–150 ms. Because an exit decision plays out in a few hundred milliseconds, exit-intent interventions are driven client-side by the SDK using the most recent scores, with edge scores refreshing between interactions rather than inside the exit window.

Model 13: Purchase Timing

The purchase timing model predicts how close a user is to making a purchase decision. Unlike the intent score (which measures willingness), purchase timing measures proximity -- how many more interactions until the purchase event occurs.

The 8 Timing Signals

As with abandonment, the signals below illustrate what the model watches; how each is weighted is tunable per site.

Signal	Description
Funnel stage progression	How far through the conversion funnel (browse → product → cart → checkout)
Decision confidence velocity	Rate of change in decision confidence score (rising = closer)
Cart completeness	Percentage of typical pre-purchase actions completed
Comparison narrowing	Convergence of product consideration set
Payment readiness signals	Scrolling to payment section, clicking payment method selector
Time pressure indicators	Visiting sale pages, countdown timers, limited stock messages
Visit frequency acceleration	Increasing visit frequency = decision approaching
Social proof consumption	Reading reviews then stopping (final validation before purchase)

Proximity Score Ranges

The bands below are an operating framework for reading the model's conversion-proximity output:

Score Range	Proximity	Estimated Time to Purchase	Optimal Action
0–20	Distant	Multiple sessions away	Content marketing, brand building
21–40	Approaching	1–3 sessions away	Retargeting, email nurture
41–60	Near	This session, but more exploration needed	Social proof, comparison tools
61–80	Imminent	Within the next few minutes	Remove friction, ensure smooth checkout
81–100	Ready now	Within seconds	Clear CTA, one-click purchase if possible

Urgency Classification

The timing model outputs an urgency classification alongside proximity. The four urgency types below are our operating framework for matching messaging to that output:

Urgency Type	Detection Signals	Messaging Strategy
Time-limited	Sale page views, countdown timer interaction, promo code entry	Reinforce deadline, show remaining time
Need-driven	Urgent search terms, fast navigation, minimal comparison	Emphasize immediate availability and fast delivery
Research-complete	Long research phase now converging, reviews consumed	Confidence reinforcement, satisfaction guarantee
Externally triggered	Referrer from price-alert email, social recommendation	Validate the referral source, show deal details

Visit Frequency Acceleration

One of the strongest cross-session timing signals is increasing visit frequency. A user who visited once a week for three weeks, then twice in three days, is approaching a decision: