It’s no secret that AI is disrupting industries up and down the economy. BI is no exception.
Traditional BI is slow, static, gated, and difficult to change. AI, by contrast, is the opposite. With conversational AI, anyone–regardless of technical ability–can ask any question about their data and get answers immediately, reducing a month or more of work down to minutes. No months-long data pipeline projects or fixed dashboards required.
It’s an exciting time. But to take full advantage of it, you need a data architecture that can deliver the data AI systems need to deliver accurate, timely responses.
This requires two things. The first is a solid data foundation. The second? Support for data context. These come together in what we call the context layer, or semantic layer, and this leads you directly to the data foundation underpinning AI.
Fortunately, the industry already has a best-in-class tool for supporting the context layer. In this article, we’ll look at why AI is sweeping the industry, how a data lakehouse provides the best foundation, and how to select the data lakehouse that’ll work best for you.
Why the future of BI is AI
It’s clear that traditional BI isn’t working for many businesses.
The numbers make this clear. The Luzmo 2025 State of Dashboards report found 41% of companies saying they spend four months building dashboards. And yet 72% of users say dashboards are so frustrating that they just download the data and work with it in spreadsheets.
Four months is a huge investment of time and money only for employees to end up in Excel.
Traditional BI is static, slow, and boring. It takes days, weeks, or months to produce data pipelines and dashboards. Most users don’t even end up using the dashboards they get, as they’re often no longer needed by the time they get them. The business has moved on. It’s also fragile, with pipelines and reports breaking randomly.
AI, by contrast, is fast, conversational, and dynamic. Users can ask questions of their data using natural language queries – complex, cross-cutting questions that would otherwise take intricate SQL queries and data engineering to produce.
AI doesn’t go out of date. Users can adjust their queries ad hoc. Large Language Models (LLMs) can detect breaking changes and adjust dynamically. Importantly, AI doesn’t just happen, though. It’s more than pointing an LLM at a CSV file and saying “go.” To deliver accurate results, AI requires context. And as we’ve written before, AI is only as good as the data that feeds it. The race for context will determine who wins and who doesn’t in the AI era.
The AI era raises the value of the data lakehouse
The challenge is creating a platform that enables discovering, deploying, and governing the context that AI needs to work from across the enterprise.
That’s where the data lakehouse comes in.
Many companies already use a data lakehouse. The combination of the versatility and flexibility of the data lake with the workhorse performance of the data warehouse has made them a go-to solution for analytics.
These attributes perfectly position the data lakehouse to be the machine that will replace traditional BI with AI. It provides the semantic backbone that AI workloads require to deliver accurate, timely results.
What to look for in an AI-ready data lakehouse platform
What specific features of a data lakehouse support the context needed for AI? Here are the table stakes capabilities you should verify before adopting any solution.
Federate and explore all your data
Context requires bringing in data from all across the company. That presents your first challenge.
To succeed in production, AI projects need to produce answers that are accurate, consistent, and auditable. This requires having both the right data and the right context. That requires the ability to access both data and business context, both unstructured data and structured data, both cloud data and on-premises data, wherever it lives in the enterprise.
The key to achieving this is data federation. Instead of pre-centralizing everything, you use data connectors to enable discovering data no matter where it lives. You can then selectively centralize the data you identify as your most important context for optimal performance.
In the AI world, the data lakehouse isn’t the place you dump all your data. It instead becomes your center of gravity. A data lakehouse that fits the bill should have:
- Strong data connector breadth for all the sources you support.
- Full support for on-premise and hybrid deployments, which become indispensable for regulated workloads in spaces such as financial services, healthcare, and government.
- Strong query performance on distributed data at scale.
Full Apache Iceberg support
How you store your most important context also matters. For most companies today, Apache Iceberg has quickly become the standard for both analytics and AI workloads.
Not all Iceberg implementations are created equal, though. Verify before you buy that your data lakehouse supports:
- The full feature set of Iceberg, including time travel, schema evolution, row-level deletes, and hidden partitioning.
- New features as they roll into the standard. As of this writing, that means support for Iceberg V3, which adds features such as row-level lineage and deletion vectors.
- Managed Iceberg, which automates the work involved in keeping tables running at peak performance through operations such as file compaction, snapshot management, and partition and sorting maintenance.
- An open query engine, such as Trino, as opposed to a proprietary engine that locks you into a vendor’s implementation.
Data products as the semantic layer
The access problem is fundamental. Without enterprise-wide access to your data, you don’t have access to the data and context needed for AI.
Once you’ve established that access, it’s time to think about the next layer on top of it. Data for AI requires rich metadata to be useful – i.e., a semantic layer that documents where the data came from, who owns it, and what business purpose it serves.
Data products govern the semantic layer, creating a grounding layer that makes data useful for AI. A data product is a curated, accessible wrapper built around high-quality datasets and bundled with relevant metadata. They take raw data and give it intent.
Without data products, the best you can say is that “the AI model, technically, has access.” With data products, you can assert, “the model has access to the right thing, with the right context – and we can prove it.”
A data lakehouse platform for AI should enable defining the basics of a data product, including:
- Documenting business meaning – i.e., the semantic layer without which AI won’t interpret data correctly.
- Data lineage, showing how data flows from source to transformation.
- Fine-grained access controls (via role-based access control) that regulate who can access the data product and its underlying datasets.
The data lakehouse you already needed – now ready for AI
AI is blowing traditional BI out of the water. Users are flocking to tools that enable them to have conversations with their data. But getting the right answers requires a context layer that enables both data access and rich semantics.
Fortunately, the data lakehouse is sitting right there, and the capabilities that made it work for analytics make it the perfect foundation for your context layer.
To learn more about the capabilities that make a data lakehouse work in the AI age, download our most recent data lakehouse buyer’s guide.
