Data & AI Trends for 2026: Lessons from our Google Cloud Projects

Heading into 2026, the conversation around AI has changed. Organisations are looking to deploy AI systems they can trust, at scale and within European regulatory boundaries.

In 2025, Crystalloids partnered with organisations across retail, healthcare and e-commerce to design and run data-driven platforms at scale. Those projects highlighted both the direction of the market and the practical barriers companies must overcome in 2026.

Below, our CEO Richard Verhoef shares the seven trends that stood out most clearly from the work we delivered this year.

How Crystalloids approaches AI-ready customer data foundations

Throughout this article, we reference real projects where organisations prepared their data for agentic AI, real-time personalisation and European compliance. At Crystalloids, we support this by designing API-first, warehouse-native customer data architectures on BigQuery, built to be real-time, governed and ready for AI. 

1. Agentic Commerce & AI Shopping Agents

AI shopping assistants are becoming normal. The challenge now is to prepare your data to let them reason, decide, and act safely.

Richard’s perspective:
“The biggest thing we learned is that most product feeds are designed for keyword search, not for AI reasoning.

For an agent to reliably decide which jacket suits ‘windy but sunny weather,’ the product catalogue must contain attributes around material weight, breathability, and usage context. Most retailers simply aren’t ready for this. They’re still feeding agents keyword-optimised descriptions, not structured metadata.

The second lesson was governance. You can’t deploy a ‘black box’ agent and hope for the best. You need payment protocols, negotiation rules, and audit trails that align with the EU AI Act. And for anything high-stakes, you still need a human in the loop.

Finally, real-time data is non-negotiable. If an agent negotiates a purchase for a product that sold out ten minutes ago, trust collapses. Retailers must shift to warehouse-native, real-time architectures so agents reason over the latest truth, not yesterday’s snapshot.”

2. AI-Powered Discovery: Semantic Search, Recommendations & Generative Merchandising

Discovery is moving from keywords to meaning. AI understands intent, context and behaviour, but only when the data behind it is ready.

Richard’s perspective:
“The biggest breakthrough wasn’t a new algorithm, it was vector embeddings combined with structured product metadata.

AI-powered discovery finally works when you stop forcing the model to guess. When a customer searches for “gezellig tafelen” (cozy dining), an AI system should surface raclette sets or atmospheric lights even when the catalogue doesn’t contain that exact wording. That only happens when you enrich data with attributes that reflect real customer needs.

We also saw that understanding intent matters. A search for “schoencadeautje” (Sinterklaas shoe gift) is not just a query, it carries cultural expectations: small, inexpensive, fun. By combining behavioural signals and semantic data, AI learns the ‘why’ behind customer behaviour, not just the ‘what.’

Under the EU AI Act, transparency is becoming essential: AI systems must be able to explain why something was recommended. That has now become part of discovery, not an afterthought.”

3. The Warehouse-Native CDP Becomes the New Standard

European companies are moving identity, consent and audiences into BigQuery, not into external CDPs.

Richard’s perspective:

“What we saw in 2025 is that moving to the warehouse is no longer just a technical optimisation; in Europe, it has become a requirement for control and compliance. Data Sovereignty has become a real concern. 

Legal and privacy teams are increasingly pushing back on traditional CDPs because copying customer data into a third-party system creates GDPR and EU AI Act risks. With a Warehouse-Native (Zero-Copy) architecture, the data never leaves the organisation’s sovereign cloud region, giving them the lineage and auditability regulators expect.

We also noticed frustration with the ‘integration tax’ of packaged CDPs. European organisations are cost-conscious, and many realised they were paying twice to store the same data. Composable CDPs allow them to use the warehouse as the single source of truth, and to run segmentation or LTV modelling using their own business logic instead of a vendor’s rigid schema.

Finally, AI Readiness, the ‘Zero-Latency’ requirement, became impossible to ignore. Teams realised that Agentic AI and Predictive Personalisation break down when the underlying data is hours old. 

By keeping identity and decision logic in BigQuery and feeding it real-time signals, companies ensure that AI systems always reason over the most current information, preventing the embarrassing scenario where an agent recommends something that went out of stock minutes earlier.”

4. Real-Time Everything

Retailers now expect reactions in minutes, not overnight. This applies to stock, pricing, journeys and compliance.

Richard’s perspective:
“In 2025, real-time data wasn’t just useful, it became essential to prevent failure.

First, ghost inventory disappears when you shift to streaming data pipelines. Using Pub/Sub and Dataflow, retailers ensured that the inventory an AI agent sees is the inventory that actually exists. That alone prevented countless failed orders.

Second, dynamic pricing became real. With the spread of electronic shelf labels (ESL) in European supermarkets, pricing updates happen in minutes. Vertex AI models adjust prices for perishable goods to reduce waste, but this only works when the data backbone is real-time.

What retailers will expect in 2026 is not just ‘faster systems,’ but real-time governance. Under the EU AI Act, they will demand kill switches and monitoring that automatically halt an agent or algorithm when it behaves outside approved boundaries.”

5. AI-Native Data Clouds & Internal Data Agents

AI is moving inside the data warehouse. Internal agents will assist merchandisers, planners, marketers and analysts.

Richard’s perspective:
“Most organisations underestimate the semantic gap. An internal AI agent behaves like a smart intern, it has no tribal knowledge unless you give it definitions.

If Marketing defines revenue as gross including VAT and Finance defines it as net, the agent will give the wrong answer unless the semantic layer resolves the meaning. Teams that invested in consistent definitions succeeded; those who didn’t saw hallucinations and errors.

Access control was another blind spot. In a warehouse-native world, an AI agent can technically see everything unless restricted. GDPR requires strict row-level policies.

And with the EU AI Act, explainability becomes mandatory. Agents must log how they reached a conclusion, what data they used, and why a recommendation was made. This is no longer optional, it’s a regulatory reality.”

6. Predictive Personalisation & Customer Intelligence

Personalisation is becoming anticipatory: real-time, multilingual and deeply contextual.

Richard’s perspective:
“We learned that predictive personalisation is not about ‘better models.’ It’s about better architecture.

The teams that succeeded were the ones who consolidated all data, POS, app, returns, loyalty, into BigQuery. The warehouse-native approach ensures compliance and accuracy, eliminating the inconsistencies caused by copying data into multiple systems.

Real-time context also became crucial. Prediction requires understanding what a customer is doing now, not what they did last year. Real-time behavioural signals combined with Vertex AI allowed us to adjust recommendations on the fly.

Finally, multilingual empathy matters. Using Vertex AI and Gemini, we generated personalised content that captured nuance, not just translation. And explainable AI helped retailers justify personalised offers under the EU AI Act.”

7. Responsible AI, Privacy & Governance

The EU AI Act is reshaping how organisations build and deploy AI. Governance is no longer a blocker, it’s the foundation.

Richard’s perspective:
“The misconception I kept hearing is that governance slows you down. The opposite is true: governance enables scale.

In 2026, organisations must move from model governance to agent governance. As AI agents start negotiating prices and making autonomous decisions, companies need lineage, transparency and defined protocols.

Warehouse-native lineage in BigQuery ensures data never falls into black boxes. Agent-to-agent protocols, promoted by Google Cloud, ensure safe negotiation between agents. And the ‘open kitchen’ principle,  showing regulators how decisions are made, builds trust faster than any marketing campaign.

If you want to unleash autonomous agents safely, governance is your competitive advantage.”

Closing Thoughts

2025 showed us that the future of retail and customer experience isn’t just about more AI, it’s about smarter foundations.

Agentic commerce, real-time intelligence, warehouse-native CDPs, and predictive personalisation all depend on one thing: trustworthy, structured, explainable data.

In 2026, the organisations that win won’t be the ones deploying the most models.
They’ll be the ones building the clearest architecture, the one AI systems can understand, reason over, and justify.