Intent signals

Behavior patterns Orwel reads from each session and the action each one suggests.

View as Markdown Open in Claude Open in ChatGPT

Orwel reads how a visitor behaves in a session (clicks, scroll, hovers, form inputs, pace) and turns it into intent signals: behavior patterns such as comparing options, ready to act or lost. They are statistical patterns of on-page behavior, not a psychological profile of the person, and they exist to trigger a useful action: show proof to someone who is verifying claims, offer help to someone who is lost, shorten the path for someone ready to act.

Signals are computed from on-site behavior only. They never use sensitive categories, and they are exposed to you through the app, the API and MCP; nothing is shown to the visitor unless you build it.

Signal to action

Signal	What the session looks like	A useful response
Decisive	Few pages, straight to pricing or signup	Keep the path short, put the CTA first
Skeptical	Rereads pricing, opens comparisons and FAQs	Show social proof, guarantees and docs
Lost	Back and forth, repeated searches, no progress	Offer chat or a guided tour
Curious	Many pages, long reads, no conversion yet	Offer a guide, newsletter or demo
Anxious	Starts forms and abandons them	Reduce fields, reassure about price and data
Engaged	Long session, deep scroll, returns often	Invite to trial, call or community

The 12 signals

Every session is scored against 12 behavior patterns, grouped below. A visitor usually shows a mix, with one or two dominant signals. The API names them personalities for backwards compatibility.

Action Architects

Users who are goal-oriented and structured.

Decisive

Quick to act, focused on results.

Impulsive

Acts on immediate urges, high energy.

Methodical

Systematic, detail-oriented.

Focused

Highly attentive to specific tasks.

Information Explorers

Users who seek knowledge and understanding.

Curious

Seeks new information, explores widely.

Skeptical

Cautious, verifies claims.

Lost

Disoriented or unsure.

Users driven by interaction and community.

Social

Seeks connection with others.

Reactive

Responds to external stimuli.

Emotional States

Transient states reflecting current mood.

Anxious

Worried or hesitant.

Playful

Fun-seeking, non-linear.

Engaged

Deeply involved and committed.

Calculation Logic

The engine uses a Bayesian Inference model combined with Rate-Based Feature Extraction.

Event Collection: The SDK captures granular events (clicks, scrolls, hovers, form inputs).
Feature Engineering: Events are normalized into rates (e.g., clicks_per_minute, interaction_ratio) to account for session duration.
Probabilistic Scoring: Each trait has a Gaussian probability model. The engine calculates the likelihood of the observed behavior matching each trait's profile.
Confidence Score: A confidence score (0-1) is assigned to each trait. Traits with high confidence (>0.6) are considered active.

Personality Combos (Synergies)

When a visitor exhibits two strong traits simultaneously, they form a Personality Combo. These combinations provide deeper insight into complex behaviors.

Complementary Synergies

Positive combinations that enhance user capability.

Executive Architect (Decisive + Methodical): Deep analysis followed by decisive action. The ideal B2B buyer.
Community Catalyst (Social + Playful): Drives engagement and viral growth.
Critical Researcher (Curious + Skeptical): Explores widely but evaluates rigorously. Hard to win over, but loyal once convinced.
Goal Achiever (Decisive + Engaged): Quick decision-making with long-term commitment.
Specialist Commitment (Focused + Engaged): Intense concentration with sustained interest.

Conflict Patterns

Opposing traits that indicate friction or uncertainty.

Tension Pattern (Impulsive + Methodical): Urgency vs. Deliberation. The user wants to act but feels the need to check details, leading to erratic behavior.
Stress Pattern (Anxious + Decisive): Caution vs. Action. Often results in "approach-avoidance" cycles (starting a form, stopping, starting again).
Navigation Conflict (Lost + Focused): Disorientation vs. Concentration. The user is trying hard to find something but failing.

Using Personalities

You can retrieve the inferred personalities for a visitor using the API:

http

GET /api/v1/identities/{id}/personalities

Use these insights to personalize the experience:

For Decisive users: Show "Buy Now" buttons and quick checkout.
For Skeptical users: Highlight social proof, guarantees, and detailed specs.
For Lost users: Trigger a proactive support chat or simplified navigation.