Why Customer Data Platforms Must Lead This Charge

In the previous post, we described a shift in how people will interact with products online — from passive browsing to dynamic, agent-driven interactions. The natural next question is: who is best positioned to build this?

The answer, we believe, is customer data platforms.

The iceberg beneath a simple search

Before we explain why, it’s worth understanding what actually happens when a user searches for something on a website today. It looks simple. A user types “blue exterior paint” and gets a list of results. But under the hood, even that basic interaction is an explosion of infrastructure.

The search query hits an indexing engine that has ingested and normalised an entire product catalogue — thousands or millions of SKUs, each with attributes, categories, pricing, inventory status, and media. The engine tokenises the query, applies relevance scoring, factors in business rules like boosting in-stock items or preferred brands, and returns ranked results in milliseconds.

That’s the baseline. Personalised search goes much further. The system needs to know who this user is — stitching together anonymous sessions with known profiles, reconciling data across devices, and resolving identity from fragments of behavioural signals. It then layers in that user’s history: what they’ve browsed, what they’ve purchased, what they’ve ignored. Machine learning models trained on this data re-rank results to surface what this specific person is most likely to want.

All of this happens in real-time, on every single query.

Now multiply that complexity for an agentic experience — where the system doesn’t just return results, but holds a multi-turn conversation, reasons about intent, and assembles a complete solution from across the catalogue. The data requirements don’t just increase. They compound.

CDPs already have the building blocks

This is exactly why customer data platforms are uniquely positioned. A CDP, by definition, already sits at the intersection of the three data layers required to power agentic product experiences:

User data. CDPs collect, unify, and resolve customer identity across every touchpoint. They know who the user is, what they’ve done, and what they care about. This isn’t a feature — it’s the core function of a CDP.

Product data. To power personalisation and recommendations, CDPs ingest and index product catalogues. They understand the relationships between products, categories, attributes, and inventory.

Behavioural context. CDPs capture the full stream of user events — page views, searches, clicks, purchases, abandonments. This is the real-time signal that makes personalisation work.

An agentic SDK needs all three of these layers working together, in real-time, to deliver a useful generative UI experience. It needs to know the user, know the products, and know the context.

Why general-purpose AI toolkits are missing context

There are excellent tools emerging for building AI-powered interfaces. Vercel’s AI SDK, for example, provides a strong developer framework for generative UI — the rendering layer for conversational experiences.

But a framework is not a data platform. These tools give you the scaffolding to build an agent. They don’t give you the customer profiles, the product index, the ML models, or the behavioural context to make that agent actually useful.

Without a CDP powering the backend, an agentic shopping experience is just a chatbot with good UI. It can talk, but it doesn’t know anything about the user or the products.

The rendering layer and the data layer are both necessary. But the data layer is where the value lives — and that’s where CDPs shine.

The infrastructure already exists

The remarkable thing is that CDPs don’t need to build this capability from scratch. The identity resolution, the event pipelines, the product indexing, the ML models — it’s all already there, built to serve today’s personalisation and analytics use cases.

What’s changed is that LLMs now make it possible to put a generative UI interface on top of this infrastructure. The hard part was never the conversation, it was building the data platform underneath it.

That platform already exists. It just needs a new interface.

This is the second post in a three-part series on the future of product experiences.