EP16: From Data to Intent: AI That Understands You, with Jonathan Lakin from IntentHQ

From Understanding Language to Understanding Behaviour

Intent HQ's approach to AI has been shaped by the belief that behaviour can reveal something fundamentally different from language. While today's large language models have become exceptionally good at understanding language and predicting what comes next, Jonathan argues that actions and behavioural sequences can provide a much deeper indication of what someone actually intends to do.

The distinction is important. Someone might have a long-term interest in rugby, for example, while separately developing an immediate intention to attend a particular game. Similarly, a series of behaviours can reveal where someone is within a much longer journey, such as moving house — from researching locations through to arranging a mortgage and eventually dealing with the Land Registry.

Intent HQ has developed its own suite of AI models to identify these patterns, ranging from long-term interests and affinities to near-term intent.

Privacy at the Source of the Data

Understanding behaviour often requires access to highly sensitive information, creating a fundamental tension between insight and privacy. Jonathan describes this as a balancing act: organisations want to generate as much useful insight as possible, while individuals should retain meaningful protection over their data.

Intent HQ's approach has therefore involved moving closer to the source of behavioural data. Mobile devices, for example, contain an enormous amount of information about what is happening around them. Rather than automatically moving that information elsewhere, Intent HQ has adopted an approach influenced by Apple's philosophy of keeping sensitive data on the device.

The challenge is to develop the techniques required to extract useful insights while maintaining privacy by default. For Jonathan, this is not simply a technical consideration. Privacy needs to be built into the architecture of AI systems from the beginning.

Combining Specialist AI with Large Language Models

Although Intent HQ has built its own AI models, Jonathan does not see the emergence of large language models as a replacement for earlier approaches. Instead, he sees an opportunity to combine different forms of AI according to their strengths.

Intent HQ has developed models using a range of techniques, including variational autoencoders, and is increasingly incorporating large language models into specific parts of its processes. LLMs can be particularly powerful where language and interaction are involved, while specialist behavioural models can provide capabilities that general-purpose language models were not designed to deliver.

The result is less about choosing one model over another and more about understanding how different AI systems can work together.

From Dashboards to Prescriptive AI

The way people interact with AI is changing too. Jonathan argues that users increasingly do not want another dashboard filled with information. They want systems that can tell them what has changed, identify the most important issues and recommend what they should do next.

This represents a shift from descriptive analytics towards a more prescriptive form of AI. Instead of simply presenting information to a user, AI can increasingly help translate that information into decisions and actions — ultimately making technology more useful without requiring people to navigate increasingly complicated interfaces.

Relevance Means Sending Less, Not More

One of the most striking examples Jonathan gives concerns customer communications. Organisations under pressure to hit targets can fall into a cycle of sending more and more messages. The result is a growing volume of irrelevant communication, eventually causing customers to disengage or switch off notifications altogether.

Intent HQ's causal AI is designed to understand the reasons behind these behaviours and identify which customers genuinely need to be contacted.The impact can be substantial. Jonathan cites work with O2 in which the number of messages sent per customer was reduced from around 460 a year to 65, while revenue simultaneously increased. The lesson is that better personalisation does not necessarily mean more communication. It can mean knowing when not to communicate at all.

AI Needs to Connect the Whole Value Chain

Jonathan's wider argument is that organisations need to rethink how they approach AI transformation. Rather than treating AI as a collection of individual tools or isolated use cases, he argues that businesses need to think horizontally — connecting data, prediction, decision-making and activation across the organisation.

A sophisticated AI model can still fail to create meaningful value if the organisation's processes prevent its output from being acted upon. Jonathan gives the example of an AI prediction that could generate significant commercial value but then becomes stuck while another department manually assesses the result in a spreadsheet. By the time the decision is made, much of the opportunity may have disappeared.

The solution is not necessarily a better prediction. It is connecting the prediction to the rest of the value chain. This leads Jonathan to a broader reframing: organisations increasingly need to think of themselves as data and compute companies, regardless of whether their traditional identity is as a bank, telecommunications company or another type of business.

Agentic AI and the Move Towards Autonomous Intent

For Jonathan, agentic AI is a natural extension of this horizontal approach. Rather than asking one AI system to perform every task, organisations can build fleets of specialised agents that work together. One might understand a company's brand and positioning, another its market, another customer behaviour and another how to optimise campaign performance.

Each agent can contribute a different form of expertise while working towards the same outcome. The potential is significant because humans rarely have the time or cognitive capacity to examine every dimension of a complex problem in this way. Jonathan describes the broader objective as creating "autonomous intent": taking an understanding of what someone wants and turning it into a process capable of acting on that understanding automatically.

Memory, Trust and the Architecture of Agentic Systems

As agents become more sophisticated, however, connecting them together introduces a new challenge: memory. Jonathan compares the architecture to the Circle of Willis, the network of blood vessels around the base of the brain. The idea is to create a system in which information can circulate between different agents, allowing them to understand what has already happened and respond accordingly.

Messaging between agents, connections to databases and external systems, and protocols such as MCP and A2A are making this increasingly possible, but technical capability alone is not enough; for organisations to adopt agentic AI, they need to trust it. That means being able to understand where information came from, validate outputs and determine when an AI system should — and should not — take action. Jonathan believes that transparency will therefore become an increasingly important part of AI design.

The Risk of Being Too Cautious

For organisations considering AI transformation, Jonathan identifies risk tolerance as one of the biggest barriers. He argues that many Western organisations have gradually reduced their willingness to accept risk to the point where it can prevent them from taking advantage of new opportunities.

His advice is to rethink the starting point. Rather than asking how AI can produce a small incremental improvement, organisations should ask how they could 10x what they are currently doing. That change in ambition can force teams to think beyond individual departments and consider how different parts of an organisation need to work together.

From Use Cases to Value Chains

Jonathan is critical of the industry's fixation on individual AI use cases. A use case may be useful, but it is only one component of a much larger process. To generate meaningful business value, organisations need to identify the entire value chain they want to transform and understand where AI can make an impact from the underlying data through to the final outcome.

This requires a different way of thinking about transformation — one that focuses less on deploying individual tools and more on redesigning how value flows through an organisation.

AI and the Future of the Global Workforce

Jonathan's perspective extends beyond individual businesses to the wider global economy. Following an extensive international trip focused on AI, he argues that emerging economies may ultimately adopt AI faster than many Western countries. Their greater willingness to take risks, combined with young and increasingly highly educated populations, could allow them to move quickly.

That could also change where work happens. AI may transform jobs rather than simply eliminate them. A role that currently involves directly handling customer calls, for example, could evolve into one focused on supervising or validating AI-generated decisions.

Jonathan expects significant disruption over the coming years, but believes the eventual outcome could be positive if societies find effective ways to manage the transition. He also sees an opportunity for emerging markets to leapfrog more established economies, particularly as access to open-source AI and increasing levels of investment lower the barriers to participation.

Building AI Transformation with Intent

For Jonathan, the next phase of AI is ultimately about more than increasingly powerful models. The real opportunity lies in connecting data, specialist AI, large language models and autonomous agents across the entire organisational value chain — while maintaining privacy, accountability and human oversight.

The organisations that benefit most may therefore not simply be those with the best individual AI tools. They will be the ones capable of rethinking how their businesses operate, setting more ambitious goals and creating the infrastructure required to turn AI-driven insight into action.

As Jonathan puts it, the opportunity is to move from understanding intent to making that intent increasingly autonomous, and that shift could fundamentally change how organisations — and the people who work within them — operate.