The ABSD Framework

Four layers from action to motivation.

Most AI systems operate at Layer 1. They track what users do. A few reach Layer 2. Almost none attempt Layer 3 or 4. YouTube's recommendation collapse — a temporary viewing episode permanently overriding your interest profile — is peer-reviewed evidence of what Layer 2 without Layer 3 produces (Haroon et al., 2023). This framework maps the inferential path from observable clicks to psychological drives, and identifies exactly what each transition requires.

Conceptual architecture

Each layer generates the one below it. AI infers upward.

Click any layer to expand its signals, theoretical grounding, and what it requires from the system. The data richness and inference confidence degrade as you move deeper. That is not a limitation. It mirrors psychological reality.

04
Motivational layer
Drive
The fundamental psychological needs and goal orientations that shape how users respond when states are challenged. Relatively stable over time.
Core drives
Competence Autonomy Relatedness Achievement goals
Grounded in: Self-Determination Theory. Deci and Ryan identified these three basic psychological needs across thousands of empirical studies spanning education, work, healthcare, and digital systems.
03
Cognitive-emotional layer
State
Momentary conditions: confusion, confidence, frustration, flow, anxiety, engagement. Not directly observable. Inferred from behavioural patterns using contextual weighting.
Detectable states
Confusion Flow Frustration Confidence Curiosity Boredom
Grounded in: Affective Computing (Picard, 1997). State detection from text achieves 70–79% accuracy in controlled settings. D'Mello and Graesser showed that transitions between states follow predictable dynamics.
02
Pattern layer
Behaviour
Temporal sequences across actions. Hesitation, thrashing, regression, acceleration, persistence, avoidance. Behaviours require sequence analysis, not individual event counting.
Seven signals
Hesitation Thrashing Regression Acceleration Persistence Avoidance Flow
Grounded in: Learning Analytics and JITAI research. Baker et al. demonstrated that gaming-the-system behaviours, detectable from interaction patterns, predict poor learning outcomes independent of performance scores.
01
Observable layer
Action
Discrete events: clicks, timestamps, submissions, navigation, keystrokes. What AI systems currently track with high fidelity. Consequences of underlying processes, not the processes themselves.
Captured today
Clicks Navigation paths Timestamps Submissions Session duration
Grounded in: Learning Analytics infrastructure. Siemens and Baker noted that the field captures behavioural traces but lacks frameworks for interpreting what those traces mean psychologically. This is where most AI sits today.
Layer transition mechanisms

Each transition requires a specific data transformation.

The four ABSD layers are not simply categories of user information. Each transition between layers is grounded in a specific research tradition that provides the methodology to accomplish it.

Action Behaviour
Bridge · Learning Analytics
Temporal aggregation
Raw events are discrete and isolated. The transition to Behaviour requires grouping events into temporal sequences and computing derived patterns across sessions. The transformation is aggregation: individual events become trajectories.
Not "user clicked help" but "user clicked help three times in ten minutes after failing the same task, and this pattern has recurred across four of the last five sessions."
Behaviour State
Bridge · Affective Computing
Contextually weighted probabilistic inference
Behavioural patterns are observable but ambiguous. The same pattern can indicate different states depending on context. The transition to State requires weighting behavioural evidence against contextual modifiers and producing a probability distribution.
Not "the user is confused" but "given this pattern, in this context, with this user's history, the most probable state is confusion (68%), with deliberation (22%) and distraction (10%) as alternatives."
State Drive
Bridge · Self-Determination Theory
Longitudinal state-response pattern analysis
A single state observation tells you what the user is experiencing now. Drive requires observing how the user characteristically responds when their states are challenged, across multiple interactions over time.
A user who enters a confusion state and persists through it shows a pattern consistent with strong competence drive. A user who disengages at the same state shows frustrated competence needs. Same state, different drive, different intervention required.
All layers Timing
Temporal calibration · JITAI
Vulnerability and receptivity windows
Just-in-Time Adaptive Intervention research does not map to a single layer transition. It provides the temporal logic that determines when each transition matters most. The same accurate state inference, acted upon at the wrong moment, can produce a negative outcome.
A user who is frustrated but still actively problem-solving may reject a simplification that they would have welcomed five minutes later. JITAI tells the system not just what to infer but when the inference is actionable.
Behavioural signal taxonomy

Seven signals. Each one a bridge between layers.

These signals appear across JITAI, affective computing, and learning analytics research. The contribution here is organising established signals into a coherent framework with explicit layer mappings. The table represents base associations, not rigid lookups.

Signal Observable pattern Inferred state Layer
Signal-to-state mappings are context-dependent. A 30-second pause before a complex financial trade may indicate deliberate prudence. The same pause before a simple checkout more plausibly indicates confusion. Contextual modifiers determine which interpretation applies. Explore the full signal library →
Contextual modifiers

Same signal. Different meaning.

The signal-to-state mappings represent base associations in the absence of countervailing context. In practice, three categories of modifier qualify how behavioural signals are interpreted at every layer.

Domain context
Industry · task type · interaction norms
A hesitation pattern in a complex enterprise onboarding carries different weight from the same pattern in a simple e-commerce checkout. The industry, task type, and typical completion time all modulate signal interpretation.
Example: 30-second pause. Complex SaaS configuration = deliberate. Simple form field = confusion.
User context
Baseline behaviour · history · expertise
A user who is consistently methodical should not have their deliberate pace interpreted as hesitation. Individual baseline behaviour, historical patterns, and expertise level all determine what counts as a deviation worth noting.
Example: Same session duration. New user = exploratory. Power user = something changed.
Temporal context
Time of day · journey stage · recent changes
A spike in navigation switching immediately after a UI redesign signals disorientation, not indecisiveness. Time of day, stage in the user journey, and recency of system changes all affect what behaviour patterns mean.
Example: Help-seeking spike. Post-update = interface confusion. Baseline period = genuine difficulty.
Inference confidence gradient

Data richness degrades as you move deeper.

This is not a limitation of the framework. It mirrors psychological reality. Surface actions are easier to observe than deep motivations. ABSD does not claim that one signal equals one state. It claims that behavioural signals, when interpreted with contextual modifiers, can support probabilistic state inference.

Action
96% fidelity
Behaviour
82% fidelity
State
70–79% accuracy
Drive
Longitudinal

State detection accuracy benchmarks from Calvo and D'Mello (2010) and Poria et al. (2017). Drive inference confidence reflects the longitudinal observation requirement described in Vansteenkiste and Ryan (2013).