What are micro-conversions and why do they matter?

Micro-conversions are small engagement actions — email signups, PDF downloads, video plays, add-to-wishlist — that indicate progress toward a macro-conversion like a purchase. ClickStream tracks each of these steps as events and scores conversion readiness, so you can weight them into a ladder that reveals where visitors are in the buying journey and which touchpoints drive final conversion.

How does behavioral bot detection work?

Behavioral bot detection analyzes interaction patterns that distinguish humans from automated traffic — natural mouse movement curves, variable typing speeds, organic scroll patterns, and realistic session timing. Unlike IP-based blocking, it catches sophisticated bots that rotate IPs and spoof user agents.

Can ClickStream detect price-sensitive visitors?

Price sensitivity isn't one of ClickStream's 26 shipped models — it's a segmentation strategy you can build on top of shipped scores like intent, value, and campaign response, combined with page-level behavior such as sale-page views and coupon-field interaction. Segments built this way can drive targeted promotional offers in your own personalization code.

Loyalty, Bot Detection, and the Price & Micro-Conversion Playbooks

What You'll See in the Dashboard

Open the Intelligence tab to find the Loyalty Trajectory and Bot/Fraud Detection cards. Loyalty Trajectory shows whether a visitor's attachment to your site is growing or fading across sessions. Bot Detection shows a confidence score with a breakdown of suspicious signals. Price sensitivity and micro-conversion progress are not dashboard cards — they are segmentation playbooks you build on top of shipped scores and tracked events, covered below.

Check the Signals tab for a feed of bot-flagged sessions (real-time streaming on Scale and up) with drill-down into behavioral biometric evidence.

Business Actions: Flag high-loyalty visitors for VIP treatment and early access offers. Use your micro-conversion ladder to identify and fix drop-off points. Use Bot Detection scores to automatically exclude fraudulent traffic from your analytics and campaign audiences — and export the evidence to dispute bad ad spend.

Playbook: Segmenting Price-Sensitive Visitors

Price sensitivity isn't one of the 26 shipped models — it's a segmentation strategy you can build on top of shipped scores (intent, value, campaign response) and page-level behavior like sale-page views and coupon-field interaction. Some visitors exhibit clear price-driven behavior — they sort by price, compare costs across products, gravitate toward sale items, and hesitate at checkout when totals appear. Others show no price sensitivity at all, focusing on features, reviews, or brand.

Knowing where a visitor falls on this spectrum lets you personalize in real time: show discount messaging to price-sensitive visitors and value/quality messaging to everyone else.

8 Price-Sensitivity Signals Worth Weighting

ClickStream captures the raw behavior behind these signals as events; the weights below are a starting point for your own segmentation logic, not shipped model internals.

Signal	Weight	Description
Price sort behavior	0.20	Sorting products by "price: low to high" on listing pages
Discount page visits	0.18	Visiting sale, clearance, or coupon pages
Price comparison dwell	0.15	Time spent looking at price elements vs. feature/description elements
Coupon field interaction	0.12	Clicking into the coupon/promo code field at checkout
Cart price sensitivity	0.10	Removing items after seeing cart total, or switching to cheaper variants
Price-to-checkout delay	0.08	Hesitation time between viewing the total and clicking "Place Order"
Tab switch after pricing	0.10	Switching tabs immediately after viewing prices (comparison shopping)
Historical discount usage	0.07	Whether this visitor has previously converted using a discount code

Price Sensitivity Tiers

Score Range	Tier	Behavior Pattern	Recommended Strategy
0–20	Price-Insensitive	Focuses on features, brand, reviews — ignores pricing	Lead with value, quality, and exclusivity messaging
21–40	Price-Aware	Checks prices but does not optimize for lowest cost	Standard pricing display, highlight value-for-money
41–60	Price-Conscious	Compares prices actively, may seek deals	Show savings vs. alternatives, bundle discounts
61–80	Bargain-Seeker	Sorts by price, visits sale pages, hunts for coupons	Surface limited-time deals, free shipping thresholds
81–100	Price-Driven	Conversion depends almost entirely on price	Show best available offer immediately, avoid upsells

Example: A Price-Sensitivity Score You Could Build

function calculatePriceSensitivity(session: SessionState): number {
  let score = 0;

  // Price sort behavior (0.20)
  if (session.hasSortedByPrice) score += 20;

  // Discount page visits (0.18)
  const discountRatio = session.discountPageViews / Math.max(session.totalPageViews, 1);
  score += Math.min(discountRatio * 3, 1) * 18;

  // Price comparison dwell (0.15)
  const priceDwellRatio = session.priceDwellMs /
    Math.max(session.priceDwellMs + session.featureDwellMs, 1);
  score += priceDwellRatio * 15;

  // Coupon field interaction (0.12)
  if (session.couponFieldClicked) score += 12;

  // Cart modifications (0.10)
  if (session.removedItemsAfterTotal) score += 10;

  // Price-to-checkout delay (0.08)
  if (session.priceToCheckoutMs > 15000) score += 8;

  // Tab switch after pricing (0.10)
  if (session.tabSwitchAfterPricing) score += 10;

  // Historical discount usage (0.07)
  if (session.visitor.hasUsedDiscountBefore) score += 7;

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

The Loyalty Trajectory Model (Return Propensity)

The shipped loyalty-trajectory model predicts the probability that a visitor will return to your site. Unlike churn prediction (Model 10), which focuses on detecting imminent departure, loyalty scoring captures the positive dimension: how attached is this visitor to your product or brand?

Return propensity gives you both a tactical view (will this visitor be back next week?) and a strategic one (is their attachment growing or fading over time?).

The Loyalty Signals

The signal weights and tier names below are illustrative — the shipped model's internals are tunable per site — but these are the kinds of signals that drive the trajectory.

Signal	Weight	Description
Visit frequency trend	0.22	Whether sessions are becoming more or less frequent over time
Feature adoption breadth	0.16	Number of distinct product features or content areas explored across sessions
Session depth trend	0.14	Whether average pages-per-session is increasing or decreasing
Direct navigation ratio	0.12	Proportion of visits that start from direct URL/bookmark vs. search/referral
Engagement consistency	0.10	Variance in engagement scores across sessions (low variance = consistent loyalty)
Account actions	0.10	Saving preferences, creating wishlists, setting notifications
Content contribution	0.08	Writing reviews, asking questions, sharing content
Recovery after absence	0.08	Returning after a period of inactivity (indicates pull-back attraction)

Loyalty Tiers

Score Range	Tier	Pattern	Strategy
0–20	One-Timer	Single visit, no return signals	Email capture, retargeting ads
21–40	Occasional	Infrequent returns, task-driven visits	Re-engagement campaigns, new content alerts
41–60	Regular	Consistent visit pattern, moderate depth	Personalized recommendations, loyalty program invite
61–80	Loyal	Frequent visits, deep engagement, growing usage	VIP treatment, early access, referral program
81–100	Advocate	Daily/weekly visits, contributes content, direct navigation	Ambassador program, exclusive perks, feedback loop

Playbook: Building a Micro-Conversion Ladder

Most analytics tools only track macro-conversions — completed purchases, submitted forms, signed-up accounts. But the path to a macro-conversion is paved with dozens of micro-conversions: small behavioral steps that indicate forward progress.

A micro-conversion ladder isn't a shipped scoring model. ClickStream tracks every one of these steps as events and scores conversion readiness; the ladder below is a weighting you can apply to those events for a granular view of how close each visitor is to converting — and exactly where they stall.

The Micro-Conversion Ladder

Micro-Conversion	Points	Cumulative Example
First page scroll past fold	+2	2
Second page view (not bounce)	+3	5
Clicked a product/feature link	+5	10
Viewed pricing page	+8	18
Watched a demo video	+7	25
Downloaded a resource	+10	35
Signed up for newsletter	+12	47
Created an account	+15	62
Added item to cart	+12	74
Started checkout form	+10	84
Entered payment information	+8	92
Completed purchase	+8	100

How a Micro-Conversion Ladder Differs from Intent

Intent (Model 1) is a predictive score — it estimates likelihood based on behavioral patterns. A micro-conversion ladder is an observed score — it counts actual steps completed. A visitor can have high intent (looking eager) but a low ladder score (has not actually done anything yet). The combination is diagnostic:

	Low Micro-Conversion (0–40)	High Micro-Conversion (60–100)
Low Intent (0–40)	Casual browser. Normal early funnel.	Completed steps mechanically but lacks enthusiasm. Check for bot.
High Intent (60–100)	Eager but stuck. UX barrier blocking next step.	On track for conversion. Clear the path.

The Bot & Fraud Detection Model

The bot detection model is the security layer of the behavioral scoring pipeline. It analyzes how a visitor interacts with your site at the motor-control level — mouse dynamics, typing cadence, scroll physics, and timing patterns — to distinguish humans from bots, scrapers, and fraudulent actors.

Unlike CAPTCHA-based detection (which interrupts users), ClickStream's behavioral biometric approach runs silently in the background, scoring every session without any visible challenge.

The 10 Bot Detection Signals

The signals and weights below are an illustrative simplification of the behavioral layer. The shipped detector combines session-level behavioral scoring like this with Cloudflare Bot Management signals, a registry of 158 named bots (including 38 AI agents), datacenter-ASN checks, header-consistency analysis, and a dedicated stealth-browser lane.

Signal	Weight	Human Pattern	Bot Pattern
Mouse movement curvature	0.15	Curved, irregular paths with acceleration/deceleration	Straight lines between points, constant velocity
Click timing variance	0.12	Variable inter-click intervals (200–3000ms)	Uniform intervals (±10ms deviation)
Scroll physics	0.12	Inertial scrolling with natural deceleration	Instant jumps to exact pixel positions
Typing cadence	0.10	Variable keystroke intervals, common typo/correction patterns	Uniform typing speed, no corrections
Mouse idle micro-movements	0.10	Small jitter/drift during "idle" (hand tremor)	Perfectly stationary between actions
Viewport interaction coverage	0.08	Clustered around content zones with natural hotspots	Uniform distribution or exclusively on targets
Session timing pattern	0.08	Variable session duration, natural breaks	Repeated exact-duration sessions
Navigation pattern entropy	0.08	Semi-predictable but varied page sequences	Identical page sequences across sessions
JavaScript environment	0.10	Consistent browser APIs, natural fingerprint	Missing APIs, spoofed user-agent, headless browser signals
Request timing	0.07	Variable network timing, natural latency	Sub-millisecond consistency, impossible speeds

Detection Verdicts

Every visitor gets a single 0–100 Human score, and the dashboard collapses the underlying classification into a three-state verdict:

Verdict	What It Means	What Happens
Real Person	Strong human behavioral signals across the session	Counted normally in analytics, People, and audiences
Needs a Check	Mixed signals — could be a human with an unusual setup	Flagged for review; an operator can mark it human and the override sticks for 90 days
Automated (AI / Bot / Tool)	Classified into one of 11 bot categories, from search crawlers to stealth browsers	Excluded from analytics, People, and audiences; evidence available for ad-platform disputes

Ad Fraud Detection

Bot detection is particularly valuable for protecting ad spend. ClickStream can identify:

Click fraud: Flags likely click-fraud patterns in paid traffic segments — bots or click farms generating fake ad clicks that drain your budget.
Fake-click landings: Paid clicks that arrive on your site but never behave like humans once they land.
Attribution fraud: Suspicious click-ID patterns, helping you spot bot traffic that tries to claim credit for organic conversions.
Retargeting pollution: Bot visits creating fake audience segments that waste retargeting spend.

The Signals tab in the dashboard provides a feed of flagged sessions (real-time streaming on Scale and up) with drill-down capability. Each flagged session shows the specific behavioral biometric evidence that triggered the bot classification, so you can audit decisions and tune thresholds.

Under the Hood: Behavioral Biometric Scoring (Illustrative)

interface BiometricSignals {
  mouseCurvature: number;       // avg curvature of mouse paths (0 = straight, 1 = very curved)
  clickTimingCV: number;        // coefficient of variation of inter-click times
  scrollPhysics: number;        // inertia score (0 = instant jumps, 1 = natural deceleration)
  typingCadenceCV: number;      // coefficient of variation of keystroke intervals
  idleMicroMovements: number;   // count of <3px movements during "idle" periods
  viewportCoverage: number;     // entropy of click position distribution
  sessionTimingCV: number;      // variation across multiple sessions from same visitor
  navPatternEntropy: number;    // Shannon entropy of page transition sequences
  jsEnvironment: number;        // browser API consistency score (0 = suspicious, 1 = normal)
  requestTiming: number;        // variation in request-to-render timing
}

function calculateBotScore(signals: BiometricSignals): number {
  // Higher score = more likely to be a bot
  let botScore = 0;

  // Straight mouse paths are suspicious
  botScore += (1 - signals.mouseCurvature) * 15;

  // Too-regular click timing is suspicious
  botScore += signals.clickTimingCV < 0.1 ? 12 : 0;

  // No scroll inertia is suspicious
  botScore += (1 - signals.scrollPhysics) * 12;

  // Perfectly regular typing is suspicious
  botScore += signals.typingCadenceCV < 0.05 ? 10 : 0;

  // No idle micro-movements = not a human hand
  botScore += signals.idleMicroMovements < 2 ? 10 : 0;

  // Suspicious JS environment
  botScore += (1 - signals.jsEnvironment) * 10;

  // Other signals...
  botScore += (1 - Math.min(signals.viewportCoverage, 1)) * 8;
  botScore += signals.sessionTimingCV < 0.05 ? 8 : 0;
  botScore += signals.navPatternEntropy < 0.5 ? 8 : 0;
  botScore += signals.requestTiming < 0.01 ? 7 : 0;

  return Math.min(100, Math.round(botScore));
}

The Complete 26-Model Scoring Pipeline

With this final installment, the ClickStream behavioral intelligence pipeline is complete. Here is how all 26 models work together:

Category	Models	What They Answer
Foundation (1–3)	Intent, Frustration, Engagement	What does this visitor want, how are they feeling, and how deeply are they interacting?
Value & Safety (4–5)	Value, Anomaly Detection	How much is this visitor worth, and is their behavior normal?
Understanding (6–7)	Confusion, Emotional State	Is this visitor confused, and what is their emotional valence?
Decision (8–9)	Decision Confidence, Regret Probability	How confident is this visitor, and will they regret purchasing?
Retention (10–11)	Churn Prediction, LTV	Will this visitor come back, and what are they worth long-term?
Timing (12–13)	Abandonment, Purchase Timing	Is this visitor about to leave, and when are they most likely to buy?
Content (14–16)	Content Affinity, Form Friction, Next Best Action	What content resonates, what forms are broken, and what should you show next?
Session Quality (17–19)	Scroll Depth Intelligence, Session Quality, Conversion Readiness	How deeply are they reading, how healthy is the session, and how ready are they to convert?
Traffic & Navigation (20–22)	Bot Detection, Navigation Pattern, Return Visitor	Are they human, how do they move through the site, and are they coming back?
Campaign & Context (23–26)	Campaign Response, Device Engagement, Time-of-Day Affinity, Loyalty Trajectory	Which campaigns land, on which devices, at which hours — and is loyalty growing?

Session momentum and navigation entropy are computed alongside the models as inline session metrics.

Twenty-six models, all computed at the Cloudflare edge in real time, all visible in your ClickStream dashboard. No data warehouses. No batch processing. No third-party scripts. Just instant behavioral intelligence for every visitor, every session.