When investment outruns understanding

Enterprise spending is accelerating, but the harder challenge is making sure intelligent systems understand how the business itself works. New research suggests the organisations making the strongest progress are connecting technology with business context and measurable outcomes rather than simply investing more.

The first phase of enterprise adoption was dominated by experimentation. Organisations tested copilots, explored automation and tried to understand where new capabilities might fit without always knowing how success should be measured. That period is giving way to something less forgiving as boards expect returns and systems begin to influence real workflows rather than sit at the edge of them.

Alteryx’s 2026 IT Leader Research: The State of AI Ownership, Agents, and ROI, based on a Coleman Parkes survey of 1,400 technology leaders and automation specialists, suggests that the question has shifted from whether the technology works to whether it can be made useful inside a particular organisation. Most respondents already report moderate or significant returns, while 80 per cent expect spending to increase over the next two years. Greater investment is now colliding with weak business context and fragmented ownership, while the infrastructure underneath is becoming more complex.

The research is strongest where it exposes the gap between technical capability and operational understanding. More than half of respondents say their organisation struggles to translate business context into systems and workflows, while 77 per cent regard that context as critical to relevant and accurate outputs. The difficulty lies in turning the rules and exceptions that govern everyday decisions into something a system can reliably use.

Business knowledge is the missing layer

Much of what makes an organisation distinctive is difficult to find in a database. Pricing thresholds may sit in spreadsheets, process exceptions live in email threads and experienced employees know when a standard rule should not be applied. Even where the underlying data is accessible, the meaning around it can remain scattered across documents and individual judgement.

That distinction matters as systems move closer to action. A model can analyse a set of numbers without understanding which figure finance treats as authoritative or which exception invalidates an otherwise sensible recommendation. The failure may look like a model problem when the deeper issue is that the organisation has never made its own operating logic explicit enough to be reused.

Andy MacMillan, CEO of Alteryx, describes this as the widening gap between enthusiasm and execution. “Organizations have proven they’re willing to invest in AI, and many are already seeing returns,” he says. “But scaling AI requires more than better models. It requires making the business knowledge people use every day available to the systems making decisions.”

That is more demanding than simply connecting another source of data. Business context includes definitions and policies, but also the sequence in which decisions are made and the conditions under which a process changes direction. A forecast depends on assumptions that may never have been formally documented, while a supply chain decision can turn on constraints that are obvious to the people running the operation but invisible to a generic model.

The research therefore points towards a different maturity test. Organisations that can expose trusted business logic to intelligent systems are better placed to move from impressive outputs towards repeatable decisions. Those that cannot may continue adding capability without addressing why the technology struggles to fit the way work is done.

Ownership follows the knowledge

The same problem appears in the question of who should control deployment. The survey found broad support for collaboration between technology and business teams, with 71 per cent saying initiatives are most effective when the two work closely together. Yet strategy and delivery remain concentrated largely within IT, even though business units are more likely to define requirements and hold the domain knowledge needed to make the system useful.

Technical teams naturally retain responsibility for integration, security and much of the data access surrounding enterprise systems. The difficulty comes when that concentration of control leaves delivery too far removed from the people who understand how the process works. A system can be technically sound and still fail operationally if the assumptions, exceptions and decision criteria used by the business never make it into the workflow.

Limited self-service access to data reinforces the gap. Only 18 per cent of respondents say business users have fully self-service access to cloud data, while many still rely on technical teams. The employees with the strongest operational understanding can therefore be the same people least able to interrogate the information or shape the workflows that depend on it.

MacMillan argues that lasting value will depend on turning business logic into something that can be shared rather than leaving it trapped inside individual teams. “The organizations creating lasting value from AI will be the ones that operationalize their business logic, so it becomes visible, governed, repeatable, and ready for AI,” he says.

That does not mean every business unit should be free to deploy autonomous systems independently. It does suggest that ownership needs to move closer to the work without abandoning central safeguards. Domain experts need greater influence over how systems behave while technology teams establish the conditions under which that freedom is safe.

Agents make weak foundations harder to ignore

The urgency increases as organisations move from systems that generate outputs towards agents capable of executing parts of a process. The research finds considerable confidence in agentic systems, but the largest barrier to wider adoption is not model capability. It is the absence of clear processes for developing and managing agents.

Autonomy magnifies the consequences of weak context. A poor recommendation can be rejected by an employee who recognises that it does not fit the situation, while an agent acting on the same misunderstanding may carry the error into another system before anybody notices. The more independence an organisation grants, the more confident it must be that the agent is operating with the right information and within understood rules.

Richard Bovey, Chief for Data at AND Digital, warns that weak foundations become more dangerous as autonomy increases. “The organizations leading in AI are the ones investing in high-quality data foundations,” he says. “Without governance and reliable data platforms, AI workloads become brittle, costly, and difficult to audit. As agentic AI takes on more independent tasks, poor data becomes a systematic risk.”

The report does not suggest caution should halt adoption. Respondents expect agents to take on practical work in areas such as IT operations and customer service, but confidence is conditional on oversight. Agentic deployment therefore exposes unresolved problems that were easier to tolerate when systems merely generated text.

Poor context also undermines trust in subtler ways. Employees are less likely to rely on outputs that do not reflect their own understanding of the business, while leadership may respond to disappointing results by questioning the technology rather than the foundations beneath it. What appears to be resistance can be a rational response to systems that consistently miss nuances people know matter.

ROI becomes an operating discipline

The research suggests measurement is beginning to separate leaders from experimenters. More than two thirds of respondents report moderate or significant returns from existing investments, and workflow automation emerges as one of the clearest areas of value. Automation connects technical capability to an identifiable process where improvement can be measured rather than inferred from enthusiasm or adoption figures.

Organisations at the leading edge do not describe success simply in terms of greater spending. They point towards discipline around measuring business impact and creating governance that allows use to expand without losing control. Investment matters, but the ability to connect expenditure to a defined outcome matters more.

That should also change how projects are selected. A compelling demonstration may show what a model can do, but it says little about whether the organisation has the context or process maturity required to make the capability useful at scale. A stronger business case starts with the decision or workflow that needs to improve and works backwards into the technology required to change it.

Rising infrastructure costs add another layer of scrutiny. The report finds that expanding workloads are pushing organisations towards more complex hybrid and on-premise architectures while adding pressure to cloud spending. As investment grows, inefficient experimentation becomes harder to justify because every additional platform or workflow adds operating cost as well as capability.

The next stage of enterprise adoption will therefore be decided less by access to models than by the quality of the environment around them. Organisations need to understand the business rules that matter and ensure authoritative information can reach the systems using it. They also need a clear measure of success once those systems reach production, because none of the surrounding investment has much value if the outcome remains impossible to prove.

Enterprise value is still created through familiar disciplines: understanding the work and measuring whether outcomes improve. The organisations that do that well will not necessarily be those spending the most. They will be the ones that know enough about their own business to make the technology useful.