Every decision a person makes leaves a trail. Not in the sense of surveillance — a trail in the sense that decisions do not arrive fully formed. They emerge through a process of incremental orientation: attention narrowing, alternatives closing off, behaviour shifting in ways that are visible before the decision is consciously made. A customer who is about to churn does not decide to churn on Tuesday morning. They have been drifting for weeks. The pattern was there. Most systems were not looking at the right resolution.
Intent AI is built around a single foundational claim: that human intent, at the individual level, is recognisable from behavioural signals before it is acted upon. Not predictable in the probabilistic sense — that a certain percentage of customers in a given segment will churn — but recognisable in the specific sense that this customer's current behavioural pattern matches the shape of a decision already forming.
The distinction matters because it changes what is possible. Probability-based prediction tells you where to direct resources across a population. Pattern recognition at individual resolution tells you which specific customer to reach, on which channel, with which message, and precisely when. The commercial difference between these two capabilities is not incremental. It is categorical.
Intent AI operates across three temporal horizons simultaneously, because human decisions do not exist at a single timescale. Micro-intentions are immediate: the specific search query, the product comparison behaviour, the transaction sequence that signals a decision forming right now. They are volatile — they can appear and dissipate within hours. Macro-intentions are mid-term: the stable motivations that aggregate from daily behaviour over weeks — a sustained interest in a product category, a pattern of financial behaviour consistent with a life event, a gradual shift in engagement that signals changing priorities. Long-term intentions are the enduring drivers: loyalty dispositions, lifestyle orientations, the deep preferences that shape how a customer relates to a brand over years.
Most personalisation systems work at one timescale and ignore the others. A campaign triggered by a micro-signal without awareness of the long-term intention can be precisely timed and completely wrong. A customer whose long-term intention is to consolidate their financial relationships with a single provider does not need to be sold a competing product. They need to be shown that you understand what they are building. Intent AI holds all three horizons in view simultaneously, because that is how human decision-making actually works.
The engine that makes this possible is the Customer DNA vector — a 1,024-dimensional behavioural embedding updated daily for each individual profile. This is not a segment assignment or a propensity score. It is a mathematical object that encodes the full complexity of a customer's behavioural state in a form that supports similarity queries, anomaly detection, and real-time inference. When a new customer arrives, their DNA vector can be compared against hundreds of millions of others in milliseconds to identify look-alike patterns without any manual rule construction. When an existing customer's vector shifts — a sudden change in the direction of the embedding — that shift is itself a signal worth examining.
The hardest problem Intent AI was built to solve is not individual pattern recognition. It is pattern recognition at the intersection of extreme data volume and individual resolution simultaneously. A telco processes billions of network events daily. A financial institution handles hundreds of millions of payment transactions. The intelligence that matters — the signal that reveals a specific customer's intent state — is embedded in that volume as a pattern that no single event makes visible. Intent AI was designed specifically for this environment: a system that can hold the full complexity of high-velocity, multidimensional data in view and resolve it into actionable individual-level intelligence.
The data that feeds the DNA vector comes from the full Deep Signal stack: Edge SDK signals, high-velocity network and payments data, clickstream, CRM enrichments. A telco subscriber's vector incorporates not just their app behaviour but the behavioural exhaust of their network usage, their payment history, their spatial patterns, their communication cadence. The richer the signal, the more accurate the intent detection — and this is where the data moat becomes most visible.
Explainability is not an afterthought in Intent AI's design. Every recommendation produced by the system comes with an append-only log of the reasoning behind it, expressed in plain language. A marketing manager can interrogate why a customer was selected for a specific intervention. A compliance officer can audit the decision pathway for any individual. A regulator can verify that the system is not using prohibited signal categories or producing discriminatory outputs. This transparency is possible precisely because Intent AI is not a black-box neural network trained end-to-end. It is a structured system that applies explicit models to explicit features, with interpretable intermediate states.
The shift from batch intelligence to continuous intelligence is, in practice, the shift from understanding customers in aggregate to understanding each one individually in real time. Intent AI makes the second thing possible at a scale where the first thing was previously the only option.