Data and analytics teams have spent years making data usable for people. Analysts have reconciled definitions, reporting teams have created consistent views of performance, and business intelligence layers have simplified complexity for decision-makers. 

AI changes the scale and nature of that requirement. Models and agents need data they can interpret, trust and act upon. That means more than access to larger volumes of information. It means richer business context, clearer ownership and stronger connections across data that has been historically divided between different functions, systems and external sources. 

Consider the source: harnessing unstructured data 

AI use cases often need a wider mix of structured and unstructured data. Market signals, customer and retailer information, product attributes, operational events, call transcripts, documents, knowledge bases, social signals and near-real-time feeds may all contribute to a single AI agent. Much of this data sits outside the traditional reporting estate. 

This takes data and analytics beyond clean, structured datasets into messier sources that still need to be made usable, governed and meaningful. AI also needs context: definitions, lineage, business rules, taxonomies, relationships and semantic meaning. 

One organization described its next step as ‘rewiring the data foundation for democratization, speed and the AI age’. Their focus was data built for both humans and machines, with the data foundation positioned at the heart of how intelligence would be activated across the business. 

For this company, abundance did not equal accessibility. They had real scale already, including AI-enabled pipelines and high availability. However, data duplication, functional silos and legacy platforms were still reducing agility and increasing cost.  

Building a connected data foundation for both humans and AI 

Going beyond simply consuming more data or building better reporting layers, data and analytics leaders must now focus on building the data connections that allows people, systems and agents to interpret the business consistently. Metadata, semantic layers, lineage, taxonomies, product attributes and external signals now matter as much as the dashboards that sit on top. 

Semantic layers deserve particular attention. Large language models can interpret language effectively, but they do not inherently understand company-specific terminology, metrics or governance rules. Semantic layers provide the business context that allows AI systems to interpret information consistently and retrieve data in a controlled manner. Our view is that Data & Analytics teams are best placed to build and maintain this layer. 

For data and analytics leaders, the implication is clear. The priority is not simply to consume more data or build better reporting layers. It is to create a connected data foundation that allows people, systems and agents to interpret the business consistently.

Many data and analytics governance models were built around central control. Data access was managed carefully. New reports were validated. Metrics were approved. Analytical work passed through specialist teams. 

That model made sense when data and analytical expertise sat in relatively few places. AI changes the scale of activity. More users can access AI-enabled tools. Agents can begin to act inside workflows. Central data and analytics teams cannot review every output, decision or data interaction without slowing the organization down or encouraging workarounds. 

Governance has to move towards guardrails. Data and analytics still has a critical role, but the work shifts towards defining the conditions within which others can move safely – to provide ‘freedom within a framework’. Those conditions include quality thresholds, common definitions, lineage, ownership, access rules, risk tiers, acceptable use, escalation paths and monitoring. While some responsibilities will sit with other teams (e.g. an AI CoE, technology foundations, cyber, legal or risk), data and analytics’ role is to make sure AI systems and agents act on data that is understood, traceable and fit for purpose.   

Data and analytics: part of the AI operating system 

This data governance is an essential part of the overall enterprise ‘AI operating system’, that also includes agent validation, monitoring and lifecycle control. Responsibility for developing, implementing and maintaining that AI operating system needs clear organizational ownership – which may well sit with the data and analytics team.    

Agentic AI makes this more concrete as agents are not simply another analytical output. They can use memory, reason over goals, execute actions, access tools and data, and collaborate with other agents.  

Data and analytics leaders need to know more than which datasets, dashboards and models exist. They also need to understand which agents use which data, which definitions they rely on, who owns them, what lifecycle stage they are in and which business processes they affect. 

Enabling data-led decision making 

Data and analytics teams should help define what ‘good enough’ means for data-backed decisions, including when an output needs human review and when an agent should be prevented from acting. Validation of AI-based data products typically requires a different, non-deterministic (statistical) testing approach which is quite different to the deterministic testing approaches of the past which led to a binary pass / fail outcome based on a specific and bounded set of test conditions.  

Human input also becomes part of the picture. In agentic workflows, people may provide information that agents learn from, store or act on. Poorly routed or low-quality human input can introduce misinformation, bias, delay and rework. Knowledge captured from people therefore needs to be traceable and treated with care, not accepted as unquestioned truth. 

Much of the AI skills debate focuses on engineers, model builders and technical specialists. While those skills matter, they are not the only constraint. 

As AI becomes easier to access, the harder work is often deciding which problems are worth solving, which data should be used, how outputs should be evaluated and when a recommendation is good enough to act on. Data and analytics teams need people who can work across business context, data understanding and technology. 

That requires stronger capability in problem framing, experimentation, output evaluation, decision assurance, business translation and value measurement. Knowing how to interact with AI tools and agents is a key first step but producing an answer is only part of the job. Someone needs to know whether the answer is actually valid and useful. 

The pattern is visible in several AI ambitions now being pursued by large organizations. Research and development analysts are expected to unlock large bodies of proprietary scientific knowledge. Procurement tools are expected to interpret contracts, forecasts and market signals. AI-powered sales execution tools are expected to recommend specific actions to field sales teams. Digital planners are expected to evaluate and execute approved forecast decisions. Each of these examples depends on business judgement as much as technical delivery.  

Using data and analytics to evaluate business opportunities 

We saw a similar pattern in the redesign of a commercial analytics function at one of our global consumer goods clients. The initial concern was that the function could not keep pace with business demand and was not making enough use of newer AI and machine learning techniques. But as it turned out, the more important issue was how to connect business need, proof of concept and enterprise delivery. The resulting approach placed greater emphasis on skills and roles that could evaluate opportunities, connect technology with business context and determine how successful experiments should scale.  

Data and analytics should play a central role in deciding which AI use cases deserve investment, how they should be implemented and whether they are creating better, faster or more reliable outcomes than traditional analytics. The point is not to scale more experiments. It is to scale the few capabilities that improve decisions, reduce cost, increase speed and create measurable business value. 

Many data and analytics functions still operate like an order-taking service. Requests come in, requirements are gathered, data products are built and dashboards are maintained. 

Some of that will remain and organizations still need regulatory reporting, management information and controlled performance packs. But AI-facing D&A capability needs a different rhythm. 

AI advantage will not come from isolated use cases. It will come from reusable capabilities: data products, shared definitions, semantic layers, knowledge bases, model services and decision engines that can be applied across multiple domains and improved over time. 

These capabilities need product ownership. They need roadmaps, feedback loops, adoption tracking, monitoring, value measurement and clear routes to decommission what is no longer useful. They are also part of the wider AI operating system, because they determine whether AI can scale safely and consistently beyond individual pilots. 

Reusable data foundations to repeatedly scale AI 

One global organization’s connected data platform provides a useful example. The platform powers multiple business products, including commercial, revenue growth management and performance management capabilities. It is framed as a foundation that allows data products to be built once and used across different domains and markets. Now, it’s being used successfully as a platform to scale AI products.  

The lesson is that AI rarely scales through standalone solutions. It scales when organizations build reusable foundations that make the next product faster, safer and cheaper to deploy than the last. 

The same pattern appears in organizations scaling AI more broadly. Once use cases have proven value, the challenge becomes turning successful capabilities into reusable assets that can be deployed elsewhere. Data and analytics leaders should judge success by whether capabilities can be trusted, reused, scaled and linked to business outcomes, not just whether a dashboard, model or agent has been delivered.