We’re deep into the AI hype cycle, but beneath the noise, something quieter and more fundamental is reshaping how customer experiences actually work. All hail the Rise of The Ontologies (no, it’s not the next Transformers movie).

Just what are we looking at here? In this article, I want to highlight what ontologies enable, specifically when it comes to digital customer experiences. There are lots of positive opportunities with AI, but also a healthy dose of skepticism. With that in mind, I’ve tried to pop in my own skeptical challenges to some of the points, and address my inner concerns.

What are ontologies?

The first time I heard this, I had to look it up and was pretty confused by the dictionary definition. It’s not the sexiest term, though don’t let the academic-sounding name fool you, it’s what makes AI genuinely useful. It’s how machines stop guessing and start understanding context, meaning, and relationships. Ontologies are the backbone that turn raw data into structured intelligence, and as the title here suggests, will be key to smarter digital experiences.

Ontologies in my mind look like a good old-fashion crime board straight out of a Netflix scandi-noir drama. The data is your evidence, raw, often messy, and the ontology is the string that connects it all together, making sense of the chaos. Unlike a static board, these connections are dynamic, and shift as new information comes in, making the system smarter over time.

It’s a big topic, and a weekly discussion point within the shared learning sessions we have as team. For those of a more technically minded disposition, my colleague Dom Selvon recently wrote a piece about the Ontology Revolution, well worth a read.

How could this look for digital experiences?

Personalisation that understands context

The traditional problem is that most personalisation engines rely on shallow pattern matching, “customers who bought X also bought Y.” This creates recommendations that feel robotic and can miss the mark altogether.

The Ontology Solution: Modern eCommerce platforms (should) now be using ontology-driven AI to build rich, meaningful profiles that understand the relationships between customer preferences, life circumstances, and purchase intent.

Banking Example… Consider a bank (or Building society if you’re mutually minded) that uses an ontology to help it understand its customers’ life stage transitions and financial goals and how they relate to each other. When someone starts making regular transfers to a savings account called “house fund” whilst researching mortgage rates, the system is able to understand the semantic relationship between saving behaviour, research patterns, and the home buying intent. Rather than simply offering generic mortgage products, it could provide contextually relevant guidance on credit-rating improvement (if needed), saving for your deposit, and local house market insights, all timed to the customer’s actual readiness stage, not just demographic assumptions.

Skeptic’s challenge: “Isn’t this just more sophisticated collaborative filtering? Traditional personalisation engines have been getting better at combining multiple data points for years.”

The critical difference here is that traditional systems find probabilistic patterns without understanding why they exist. Ontology driven systems infer meaning behind customers’ actions and importantly, intent. e.g. if a someone buys running shoes, traditional systems deterministically predict “athletic footwear purchase = likely to buy fitness products.” Ontological systems understand the relationships of athletic footwear with health, but also fashion, and are able to triangulate other elements that are linked to health, or fashion, to determine the most likely (probable) intent of the customer. This semantic understanding enables predictions that work even for edge cases and new customers where statistical models fail.

Assistants that understand urgency and emotion

We’ve all experienced the chatbot doom loop, where our messages seem willfully misunderstood, we just get a link to an FAQ, or best case scenario we’re passed to an actual human to help but left on hold. Ontology-driven assistants can understand the relationships between customer needs and the services being provided, infer sentiment based on what’s written, or asked. They won’t just match keywords to canned responses. It all comes back to interpreting intent and context, and understanding how the customer, and the organisation are supposed to work together.

Insurance claim example - Imagine you’ve got some dodgy plumbing in the kitchen that’s resulted in an escape of water. If you took the route of speaking with an assistant, an ontology driven system would be able to understand the relationship between the leak and the broader implications, and the actors or equipment involved in solving it (leaks cause water damage, water damage can cause damp and mould, plumbers fix leaks, dehumidifiers prevent mould). Using this understanding, it can ask questions; “Is your kitchen usable?”, “Is the floor covering soaked through?”, “Could your appliances be water damaged?” - all this builds a picture of how serious this is, and whether it just needs a plumber to shut down a minor leak, or broader interventions for a bigger escape of water; replacing flooring or units, dehumidifiers, even whether or not alternative accommodation is even needed.

Small moments that offer up more understanding, dare I say it, almost being a little more human, could help reduce stress levels, and even reassure by letting people know what’s covered, timelines and next steps - and providing expertise to help them give the right information.

From a business perspective, the value is there too. The ontology can flag potential fraudulent claims, through an understanding of normal patterns (kitchen floods typically involve adjacent areas, specific appliance types, seasonal pipe issues) versus suspicious ones. It could automatically pull in the right people to get things fixed. At the same time, it locks down compliance: capturing evidence, applying the right coverage rules, and flagging anything that needs a closer look. All tracked, all audit-ready, no regulatory loose ends.

Skeptic’s challenge: “Advanced chatbots already use decision trees and can handle complex flows. Machine learning has made conversational AI much more sophisticated, so what makes ontology-driven systems fundamentally different?”

The critical difference here is that traditional chatbots, follow pre-programmed, deterministic decision trees or use machine learning to predict the most likely next response based on training data, excelling at common scenarios, though they struggle with novel combos or edge cases.

If customer reports “water damage in my kitchen from a burst pipe, but I’m worried about my home office equipment in my study below,” the system understands the logical connections between these concerns… pipe bursts create structural water flow patterns, kitchen incidents often affect adjacent areas and lower levels, and electronic equipment has specific vulnerability and coverage requirements. It can then reason through any implications quickly and dynamically - explaining how water damage coverage applies to both direct damage and secondary effects.

This reasoning capability allows the system to handle genuinely complex, interconnected claims scenarios, and help you the customer understand what therefore is required, rather than just following standard incident categories that probably don’t exactly fit the need.

Of course, it’s not all rainbows and sunshine

There are a few examples of when this of course doesn’t work out well and the AI support ‘goes wrong’. Air Canada’s AI-powered chatbot giving a passenger wrong advice, with the airline arguing its chatbot is “responsible for its own actions”. Ultimately the airline had to pay money because it’s responsible for what its chatbot says. There are plenty of lessons to be learnt, which are nicely summed up here.

Just one more question

“You say all of this happens like magic, but how does it really understand?”

The magic happens when the simple components work together. Instead of matching keywords, the system understands context. Instead of following rigid scripts, it can reason through novel situations by understanding the relationships between concepts.

Sure, but what creates the semantic meaning and understanding? I’ve reached my technical limitation here so I’ve turned to Perplexity to help me understand and peer behind the curtain a little.

Think of semantic meaning as the difference between a dictionary and an encyclopedia. A dictionary tells you an apple is “a round fruit that grows on trees.” An encyclopedia explains that apples relate to orchards, nutrition, recipes, cultural traditions, and economics. Semantic understanding is about recognizing and connecting these broader relationships…

Summary

We’re at a point now where customer experiences can finally live up to their promise, and as experience designers, we have the opportunity to shape how this technology gets implemented.

The technology now allows us to actually deliver and go beyond the promise of personalisation. Ontologies will enable us to create experiences that I genuinely believe will feel more intelligent, helpful and tailored to us as individuals. They’re quietly becoming the secret weapon of companies that deliver experiences people actually want to use.

There are of course lots of practical considerations in terms of data quality in the setup, the role agents play, and also let’s not forget to ensure there are humans in the loop and controls in place. As people, we’re still crucial in providing oversight of the AI systems, and ensuring nuanced human, cultural or behavioural differences are accounted for .

These are all very much open cases for another day…