Mobile view

Predicting
the Why

Scroll through the complete ABSD research in card format.

← swipe to navigate →
01 / 10
AI systems see everything. They understand almost nothing.
The data exists. The interpretation framework does not.
73%
Stuck at basic segmentation
40%
Revenue from good personalisation
500B
Events/day · Netflix · still no state inference
02 / 10
Four layers. Each one beneath the last.
Drive generates State. State produces Behaviour. Behaviour manifests as Action. AI must infer upward.
Drive
Why
State
What users feel
Behaviour
Patterns over time
Action
Observable · where AI sits today
03 / 10
Confidence degrades as you go deeper.
Action - 96% fidelity · standard event logging
Behaviour - 82% · temporal aggregation required
State - 70–79% · probabilistic, context-dependent
Drive - longitudinal · weeks to months of data
04 / 10
Four traditions. Never previously connected.
SDT - Drive layer · competence, autonomy, relatedness

Affective Computing - State layer · 70–79% accuracy from interaction data

JITAI Research - Behaviour layer · vulnerability and receptivity windows

Learning Analytics - Action layer · 78–84% dropout prediction
05 / 10
Seven signals. All in your data.
HesitationPause · uncertainty
ThrashingRapid switching · overwhelm
RegressionReturns · consolidation
AccelerationSpeed · boredom or overconfidence
PersistenceContinues after failure · high drive
AvoidanceEarly exit · anxiety
FlowSteady progress · optimal state
06 / 10
Same signal. Different meaning. Context determines which.
A 30-second pause before a complex financial trade is prudence. Before a simple form field it is confusion. Three modifiers: domain, user history, and temporal context. Apply all three before acting.
07 / 10
Netflix. 500B events. Layer 1 only.
When you pause a film, Netflix logs the event. It cannot tell whether the pause was anticipation, confusion, or discomfort. The signals for Layer 3 inference are in the data. The framework to read them is not.
08 / 10
Khan Academy. Mastery tracking. Not mastery understanding.
A student who answers after 45 seconds of hesitation gets the same credit as one who answers in 4 seconds. VanLehn (2011): this gap between AI and human tutoring persists. The AI sees the answer, not the state that produced it.
09 / 10
53 references. Live survey. Open research.
SP Jain EMBA · Applied Research Project. Survey and expert interviews with CX and EdTech practitioners. Full paper always readable - no login, no paywall.
50
References · 4 traditions
14+
Survey responses
5
Expert interviews
10 / 10
Does this resonate with your experience?
The survey takes 5 minutes. Your behavioural patterns as you answer are part of the research.
const pips = pipsEl.querySelectorAll('.mob-pip'); const io = new IntersectionObserver(entries => { entries.forEach(e => { if (e.isIntersecting) { const idx = parseInt(e.target.dataset.c) - 1; pips.forEach((p, i) => p.classList.toggle('done', i <= idx)); } }); }, { threshold: 0.5 }); cards.forEach(c => io.observe(c)); })();