AI Needs to Know What Your Data Means, and Whether to Trust It

Over the past year, nearly every data and analytics vendor has made the same argument: AI agents need business context. Give a model enterprise data without telling it how the business defines revenue, margin, or an active customer, and it will give confident answers to the wrong questions.

I agree. But definitions alone are not enough. Enterprise AI also needs reliable calculations behind its answers, evidence that the underlying data can be trusted, and clear limits on what it can access, change, and do. Basically the math still needs to be done.

None of that comes from connecting a model to a dataset. It comes from architecture decisions: where business logic lives, which engine performs the calculations, how quality and governance are applied, and how AI runs inside the systems a business already depends on. Those decisions shaped everything in our October Product Recap.

The centerpiece is Qlik Semantic Models, which bring shared business definitions (measures, dimensions, and relationships) into Qlik data products.

A business glossary helps people agree on terminology. A semantic model goes further, defining the measures, dimensions, and relationships that analytics can use to interpret and calculate results consistently. We could have built that model as a separate layer. We chose to put semantics inside governed data products, keeping business meaning close to the data it describes rather than creating another set of definitions to maintain.

Many Qlik customers have spent years encoding business logic in their analytics applications. That logic is hard-won knowledge about how their business runs, and they shouldn't have to recreate it. So alongside Semantic Models, we're introducing a way to extract existing semantic definitions from Qlik analytics applications into governed, discoverable data products. Extraction is planned to enter Public Preview on October 28, with general availability targeted for later in Q4.

Logic locked inside individual applications is difficult to govern and reuse. Moving it into data products gives every new application and AI experience the same starting point.

No enterprise runs on one technology, and the models and applications customers choose will keep changing. If every model, agent, and application holds its own interpretation of the same data, inconsistent meaning just moves somewhere new.

So interoperability has to cover more than access to data. It has to cover the definitions and relationships that determine what the data means. Qlik joined the Open Semantic Interchange initiative in January 2026, which has since become Apache OSSIE (Incubating), and we're developing support for it so semantic definitions can be used across more technologies.

Architecture also has to accommodate choice. Customers shouldn't have to rebuild their business context every time they adopt a new model or application. Our MCP Server is now available in Anthropic's Claude Directory, with more agentic marketplaces and directories to come, bringing Qlik's governed data and context into more of the AI tools customers already use.

Our data products already carry ownership, lineage, data quality, and the Qlik Trust Score. With semantics in the same place, meaning is managed alongside the information teams use to decide whether data is fit for use.

Take an agent asked why margin fell last quarter. The definition tells it how margin is calculated and which products and regions are in scope. It doesn't tell it whether the cost data refreshed, whether it passed its quality checks, or whether gaps in it could change the conclusion. An agent can calculate margin perfectly from data nobody should rely on.

The agents we're developing will need both: how a metric is defined, and the quality and provenance of the data behind it. Bringing meaning and trust together builds that foundation now, for the analytics customers run today and for agents as they arrive.

The first answer is also rarely the end. A drop in margin leads straight to the next questions: which regions, which products, compared with what, and what's missing. Qlik's Analytics Engine does the calculations and exposes associations, including values excluded by a selection that can reveal what an initial answer overlooks. Qlik Answers builds on that engine to bring more of this exploration into natural language, surfacing insights beyond the original question. We're also evolving its architecture and reasoning, with plans for customer-defined skills that reflect an organization's terminology and business rules.

AI shouldn't require customers to redesign their technology around it. The more we preserve existing applications, development practices, and governance models, the easier new capabilities are to adopt.

For data engineers, that means cutting repetitive work without losing control of how data is transformed. The Qlik Talend Studio AI Assistant, now in public preview, supports documentation, code generation, mappings, and troubleshooting. Eligible customers can enable the preview through their Qlik Cloud administrator. See the Qlik Community announcement for setup instructions.

For developers, it means working in familiar tools. Our modern script editor and VS Code extension bring Qlik development closer to the environments developers already use, and we're expanding Open Lakehouse support to Google Cloud, giving customers more choice over where they work with their data. Each of these has its own release timeline.

AI models have advanced quickly. The harder work now is connecting them to enterprise data without losing meaning, trust, or control.

Some of what I've described is available now, some is in preview, and some is still in development. The direction behind all of it is the same. An enterprise can act on an answer only when it knows what the answer means, where it came from, and whether it can be relied on. That's the standard we're building to.

Learn more: Explore Qlik's October Product Recap, including product availability, demonstrations, and documentation.