Last week, I had the privilege of joining more than 15,000 data experts, technology vendors, and executives at Big Data London. This year, the event was bigger and better than ever, bringing together over 400 talks across dozens of tracks and introducing a new “Data Driven LDN” strand to dive deep on AI agents, governance, and data products.

I was lucky enough to attend many sessions as well as chat with numerous ex-colleagues, peers, prospects, and thought leaders. For me, three major themes stood out. Below, I reflect on my top takeaways and how they align with where ALTR is uniquely positioned to make an impact.

1. AI in Production is Maturing, but Caution Still Required

One of the clearest shifts at Big Data London was the move from experimentation toward putting AI to work. PoCs that had long lingered in pilot or sandbox phases are now entering production, but not without hard-won lessons and continued caution. The new Data Driven LDN track underscored this with deep dives into agentic AI, governance, and responsible scaling.

A recurring pattern: distributing workloads across multiple AI agents (or “AI workers”) is becoming more common. Rather than a single monolithic model handling all logic, architectures are emerging in which specialized agents (e.g., for ingestion, verification, orchestration, decisioning) collaborate. This modularization helps scale, isolates risk, and allows pieces to evolve independently. One valuable tip I heard was to clearly define the role and remit of each agent.

The strong consensus among speakers and practitioners was clear: scaling AI safely is not just about models. It demands trust, explainability, guardrails, and governance baked in from day one. A session in the Data & AI Governance Theatre stressed the need for “targeted controls” that evolve with fast-moving technology, letting innovation flourish without compromising safety.

2. Basics Matter More Than Ever: Avoiding Silos, Mapping Data, Deduplicating, Securing

It might sound obvious to say “get your basics right,” but the tone across many sessions and informal conversations in the halls was clear: working, valuable AI is built on rock-solid infrastructure and data hygiene. Without that foundation, the project will likely fail.

Avoid Data Silos & Achieve Data Visibility

One persistent challenge: many organizations’ data is fragmented, stuck across legacy systems, isolated team databases, or Excel spreadsheets in Slack or OneDrive. Several speakers returned to the point that you can’t govern or leverage data you can’t see or trust. The push toward “data products” (exposed, well-governed slices of data) implicitly assumes a unified data plane, not islands.

Understand What You Have and Where

Closely related: you must know what data you hold, where it lives (cloud, on-prem, data lake, warehouse, streaming tables), what schemas and versions exist, and, most importantly, what is sensitive in that data. I heard several organizations admit to surprises when an AI model accessed redundant, stale, or sensitive data simply because it wasn’t fully tracked.

Here, metadata and cataloging are no longer optional extras, they’re foundational.

Deduplication, Lineage & Data Hygiene

Even when data is consolidated, duplication, inconsistent formats, or conflicting versions remain eternal pain points. Several sessions on data engineering architecture emphasized streaming deduplication, canonical modeling, and lineage tracking as essential practices before you can confidently rely on AI outcomes.

Top Priority: Security & Protection

Perhaps the loudest, clearest message at Big Data London was that data security is non-negotiable. It’s not enough to encrypt at rest; you must control usage, track derivations, and enforce context-sensitive policies (e.g., time, location, sensitivity).

3. Metadata Management & Central Knowledge Graphs: From Nice-to-Have to Strategic Imperative

If there was a single buzzword that echoed in more conference rooms and exhibit halls than any other, it was metadata. The quality, structure, ownership, and accessibility of metadata are rapidly moving from peripheral concern to central leadership agenda.

The Rise of a Centralized Knowledge Graph

Multiple sessions called for a single, enterprise-wide knowledge graph (or semantic layer) that ties business concepts, data assets, and lineage together, enabling consistent context, automated discovery, and federated governance. Some talks described how this graph becomes the connective tissue across AI agents, analytics platforms, business applications, and even policy systems.

The purpose: rather than each team maintaining their own catalogs, you invest in a unified backbone so any consumer—analyst, engineer, AI agent can query “What does X field mean, where is it derived, and who owns it?”

Business Owners as Metadata Stewards

One of the more practical debates was around ownership: whose job is metadata? The consensus leaned heavily toward business-facing owners, not just IT or data teams. The idea is that domain experts who understand context, semantics, and business meaning should be accountable for metadata definitions, glossaries, and rules. Data teams act as facilitators and enforcers, but the semantic “truth” must come from the domain.

This aligns tightly with ALTR’s philosophy: security and policy are most powerful when aligned with business semantics, not as an afterthought bolted onto raw tables.

Automated Metadata Augmentation

In academic circles, researchers are already exploring how modern AI can assist with metadata tasks: generating context, detecting schema drift, auto-tagging, and facilitating discovery. A recent paper, “Impact and Influence of Modern AI in Metadata Management,” highlights how AI can help generate and maintain metadata in complex environments.

Wrapping Up

Big Data London 2025 made one thing abundantly clear: the organizations that will succeed with AI aren’t necessarily those with the most advanced models or the biggest budgets. They’re the ones that have invested in the fundamentals: unified data visibility, rigorous governance, and metadata strategies that tie it all together. As AI moves from experimentation to production at scale, these foundations aren’t just enablers of success; they’re prerequisites for it.

The conversations I had in the halls and sessions reinforced what we’re seeing at ALTR every day: security, governance, and business context must be woven into the fabric of your data architecture, not bolted on as an afterthought. The good news? Organizations that get these basics right are already seeing the payoff in faster, safer AI deployments.

If your team attended Big Data London or is wrestling with scaling AI across your organization, I’d love to connect and share lessons.

If your team attended Big Data London or is wrestling with scaling AI across your organization, I’d love to connect and share lessons.