Domain · Customer Experience

CX teams have the data.
They miss the interpretation.

73% of companies cannot use behavioural data beyond basic segmentation. The signals that would explain why customers leave, churn, or disengage are already in their logs. The framework to read them is not.

Most CX systems operate at Layer 1.

Customer experience platforms track events at massive scale. Clicks, session duration, purchase funnels, support ticket volumes, NPS scores. The data infrastructure is sophisticated. The interpretive framework is not. Most CX AI optimises for the next action - next product viewed, next message sent - without understanding the psychological state driving the current one.

The gap is structural. Forrester found 73% of companies struggle to use behavioural data beyond segmentation. The missing piece is not more data. It is a framework for what the data means.

Action
Strong
Behaviour
Partial
State
Minimal
Drive
Absent

Based on Forrester (2022), McKinsey (2021), Lemon and Verhoef (2016). Indicative of sector average, not any specific platform.

Seven signals. What each one means in a CX context.

The same signals that appear in learning contexts appear in CX - because they reflect fundamental patterns in how people process difficulty, choice, and uncertainty. The domain context changes their interpretation. Their presence in the data does not.

Hesitation
Long pause before confirming a purchase, upgrading a plan, or accepting a terms change.
What to do: Surface a one-line clarification relevant to the specific pause point - not a full help article. Reduce the information gap causing the hesitation without overwhelming.
Thrashing
Switching between pricing tiers, feature pages, or support channels without completing any action.
What to do: Trigger a contextual comparison or a proactive chat offer. The user is in decision paralysis - they need simplification, not more options.
Regression
Returning to an onboarding step, configuration screen, or product page completed days or weeks ago.
What to do: A targeted "need help with [specific step]?" prompt. The user has identified a gap in their earlier understanding - support the self-correction without making them restart.
Acceleration
A power user speeding through a new feature announcement without engaging.
What to do: Offer an expert-level summary or advanced mode. The issue is relevance, not pace. Slowing them down will not work.
Persistence
A user attempting to configure an integration three times, researching between each attempt.
What to do: Mostly nothing. A single contextual resource offered once if they have been trying for more than 20 minutes. Do not interrupt someone who is problem-solving.
Avoidance
A customer logging in daily but never opening the analytics section they feel they should understand.
What to do: A low-stakes entry point - one useful metric, not the full dashboard. Reduce the perceived barrier without revealing the system has detected the avoidance.
Flow
A customer completing their core workflow consistently, returning regularly, rarely contacting support.
What to do: Nothing. Do not send re-engagement emails to users in flow. Preserve the conditions that produced the state - do not interrupt them.

If CX AI reached Layer 3 - what changes.

Churn prediction shifts from when to why. Current churn models predict when a user is likely to leave based on engagement drop-off. A Layer 3 system can detect whether that drop-off is temporary distraction, product-fit frustration, or genuine disengagement - each requiring a different response.

Personalisation becomes relevant, not just targeted. Personalisation at Layer 1 delivers products the user has clicked before. At Layer 3, it delivers content and support calibrated to the user's current psychological state - what they need right now, not what they have done before.

Support becomes proactive without being intrusive. The single most common complaint about AI-driven support is that it intervenes at the wrong moment. Layer 3 detection distinguishes a user who is frustrated and disengaging from a user who is frustrated and persisting. The first needs intervention. The second does not.

The 40% revenue gap closes from the inside. McKinsey's finding - that companies excelling at personalisation generate 40% more revenue - is not about better targeting algorithms. It is about relevance. Layer 3 inference is the structural requirement for genuine relevance at scale.

The consequence of stopping at Layer 2

Recommendation drift is a CX failure mode — not just an entertainment problem.

YouTube's recommendation collapse — documented by Haroon et al. (2023) using 100,000 automated accounts — is the clearest peer-reviewed example of what Layer 2 pattern detection produces without Layer 3 state inference. A temporary viewing episode overrides an established interest profile because the algorithm treats all engagement as equivalent evidence of stable preference. The same failure mode appears in every CX system that personalises over time: a customer service interaction, a product recommendation, a support flow. One situational spike in behaviour permanently shifts the model. The customer gets recommendations calibrated to their worst moment, not their established needs. Layer 3 inference — detecting that a behaviour was situationally triggered rather than dispositionally driven — is the structural fix. This is not a YouTube-specific problem. It is the structural consequence of building personalisation on Layer 2 without Layer 3.

See it traced
Watch a CX signal travel from cart abandonment to psychological drive.
Trace a CX signal → Take the survey