Enterprise artificial-intelligence spending is rising faster than many organisations can turn everyday business knowledge into something models and agents can use, leaving a gap between technically capable systems and the rules, definitions, and operating context that shape real decisions.

New research from Alteryx found that 80% of surveyed organisations expect AI spending to increase over the next two years and 69% already report moderate or significant returns. Yet 53% say they struggle to translate business context into AI systems and workflows, even though 77% regard that context as critical to producing accurate and relevant outputs.

The 2026 IT Leader Research surveyed 1,400 technology leaders globally and was released on 13 August. Its sample is weighted towards organisations already engaged with AI and enterprise data initiatives, while Alteryx has a commercial interest in the analytics and data-management market, so the findings are best read as evidence of implementation pressures among active adopters rather than a universal measure of business readiness.

The problem becomes more visible as AI moves from assistants into workflows. A model can retrieve documents and analyse records, but it will still struggle with organisation-specific definitions, approval thresholds, exceptions, and informal operating rules unless those are represented in the systems around it.

Data access is only half the problem

Enterprise AI programmes have often treated data availability as the main prerequisite for adoption. Alteryx’s figures show that access remains uneven — only 18% of organisations report fully self-service cloud-data access for business users — but making information technically reachable does not mean the system understands how the organisation uses it.

A revenue figure can be calculated differently by finance, sales, and regional teams; a customer may be considered active under one workflow and dormant under another; and an approval rule that staff understand instinctively may never have been written into a data model. Those inconsistencies are manageable when experienced employees can interpret them, but they become more dangerous when an automated system is expected to make or recommend decisions.

That is why business context is becoming a data-governance problem in its own right. Organisations need to capture definitions, rules, thresholds, ownership, and exceptions in forms that can be reused across analytics and AI rather than leaving them inside spreadsheets, documentation, or institutional memory.

The work is less visible than buying a new model, but it determines whether AI outputs can be compared, audited, and trusted across departments. A highly capable model connected to contradictory definitions will still produce unreliable results.

Agents increase the value of unwritten rules

Agentic AI raises the stakes because a system can move from generating an answer to initiating an action. Alteryx’s research found strong confidence that agents can produce measurable returns, with workflow automation among the areas technology leaders associate most closely with AI value.

Once an agent is allowed to update records, trigger processes, answer customers, or coordinate operational work, the difference between a documented policy and an unwritten convention becomes material. Human employees often recognise when an exception needs escalation because they understand organisational history and consequences that are not represented in the software.

An agent does not inherit that tacit knowledge automatically. It needs access to governed rules and contextual information, together with permissions, monitoring, and escalation routes that limit what happens when the situation falls outside expected conditions.

That makes the design of AI workflows an interdisciplinary task. Technical teams can build integrations and controls, while operational teams hold much of the knowledge needed to define what the system should do. Alteryx found that 71% of respondents associate successful AI initiatives with close collaboration between IT and business teams, which reflects the difficulty of separating technology delivery from process ownership.

AI ownership becomes an operating-model question

The research also exposes tension over where responsibility for AI should sit. Strategy and technical delivery often remain concentrated within IT, while business teams define requirements and own the processes being automated. That division can work for conventional software projects, but agents blur the boundary because they participate directly in operational decisions.

Governance therefore has to cover more than model selection. Organisations need clear ownership of business definitions, data quality, permissions, performance measures, and the consequences of automated actions, with enough auditability to identify why a system behaved as it did.

The gap between rising spend and weak contextualisation does not mean enterprise AI investment is failing; Alteryx’s own figures show that many respondents already report returns. It does suggest that the next constraint is moving away from access to models and towards the less glamorous work of making organisational knowledge explicit, consistent, and usable by machines.

As agents take on more operational work, businesses that have not resolved those definitions will find that automation exposes old ambiguities rather than removing them. AI can execute a process quickly, but it cannot supply the missing business rules on which a reliable process depends.