Summary
- Only 16% of surveyed UK AI leaders have full investment authority, and 11% control workforce strategy.
- Some 51% report agents executing actions across workflows, while 77% encounter governance-related delays.
- Just 32% say their organisations act quickly when AI performance evidence identifies opportunities.
Only 16% of UK artificial intelligence leaders surveyed by IBM have full authority over AI investment decisions, despite growing responsibility for the technology’s performance, governance and contribution to business results.
Research published on 17 November by the IBM Institute for Business Value also found that just 11% of respondents controlled workforce strategy, while fewer than one-third said their organisations responded quickly when AI performance information revealed opportunities to improve outcomes.
The findings suggest that companies are giving senior AI executives responsibility for substantial organisational changes without consistently granting them the authority to redirect spending, alter working practices or influence decisions about employees.
IBM surveyed 1,667 executives across 33 geographies and 20 industries, including 100 respondents in the United Kingdom. The research covered senior figures responsible for AI strategy, implementation, governance and value creation, rather than exclusively executives holding the formal title of chief AI officer.
AI leadership encounters established decision structures
Among UK respondents, 34% said AI was reshaping their organisation’s business model, although the proportion reporting direct control over the resources required for such changes was substantially smaller.
Investment authority influences which applications receive funding, whether computing infrastructure is purchased and when projects that fail to meet expectations are revised or discontinued. Workforce decisions affect recruitment, training, departmental responsibilities and the allocation of employees to redesigned processes.
When an AI leader can identify an opportunity but cannot authorise the resources needed to pursue it, implementation depends on decisions made by other executives. That can be appropriate where significant investment or employment changes require financial and board oversight, but it introduces difficulties if responsibilities and decision rights are not clearly established.
IBM found that 41% of surveyed UK executives said opportunities identified through AI performance information were discussed but action remained slow. Only 32% reported that their organisations could quickly adjust decisions and resources in response to those findings.
Such delays may reflect competing investment priorities, uncertainty about anticipated returns or the practical challenges of changing established business processes. The survey does not identify a single cause for slow decisions, nor does it establish that giving AI executives unrestricted authority would necessarily improve outcomes.
However, organisations need mechanisms for determining when additional spending is justified and when evidence indicates that an application should be changed or withdrawn. Formal responsibility for AI without an effective route into these decisions can weaken accountability.
Autonomous systems introduce different governance demands
The research also examined the use of AI agents, which can retrieve information, interact with software and carry out sequences of actions rather than simply producing responses for employees.
Some 51% of UK respondents said agents already initiated and executed actions across workflows. At the same time, 77% reported that governance, risk or compliance requirements delayed AI-driven activity.
These findings do not necessarily indicate that governance controls are excessive. An agent preparing an internal report presents different risks from one changing customer records, approving payments or granting access to confidential information. The appropriate level of supervision depends on the activity and the consequences of an incorrect action.
While 79% of UK respondents said their organisations had defined activities where human oversight was essential, only 58% reported that AI decision rights were clearly specified and consistently applied.
The distinction between defining a policy and enforcing it in operational software becomes increasingly important as agents acquire more capabilities. A requirement for human approval offers limited protection if the systems involved cannot reliably identify when an approval is needed or prevent an action from proceeding without it.
Businesses also need to monitor how permissions change. An agent initially authorised to retrieve information may subsequently gain the ability to update records or interact with external applications, creating risks that were not present during its original assessment.
Identity management, activity logging and restrictions on access therefore need to operate alongside business governance arrangements. These controls help establish what an agent was permitted to do, which actions it performed and who remains accountable for the outcome.
Better performance requires responsive investment
IBM identified a group of higher-impact AI executives, representing approximately one-fifth of its global sample, whose organisations demonstrated stronger practices around governance, decision-making and value management.
These executives were nearly four times as likely as their peers to report being able to pause, adjust or redirect AI systems quickly without disrupting operations. Among the higher-impact group, 94% said business outcomes informed decisions about which initiatives received additional funding and employees.
The figures describe global findings rather than results specific to the United Kingdom, and they identify associations rather than proving that any single management practice caused superior performance.
Nevertheless, the comparison illustrates the importance of connecting technical performance with decisions about investment and operating priorities.
AI programmes can produce impressive measures of task completion, processing speed or usage without establishing that the organisation is financially better off. Faster document preparation may save employee time, for example, while additional supervision, integration and maintenance introduce costs elsewhere.
A useful assessment must account for the entire process, including work that employees still perform and any corrections needed when automated systems produce inaccurate results.
Recent research into enterprise AI financial accountability identified similar difficulties connecting deployments with commercial outcomes. IBM’s findings add the question of whether the executives monitoring performance have sufficient influence over the decisions that determine its value.
Competitive pressure adds to organisational strain
Nearly three-quarters of surveyed UK AI leaders, 73%, believed their organisations risked falling behind competitors in AI transformation, while 29% said necessary preparations had not yet been completed.
Competitive pressure can encourage businesses to introduce applications before establishing consistent governance, particularly when individual departments are able to purchase software independently. Central technology teams may subsequently inherit responsibility for tools they did not select and processes they did not design.
Addressing that fragmentation does not require every decision to be concentrated in one executive position. Finance, technology, security, legal and operational leaders may retain separate responsibilities, provided their authority is clear and decisions can be escalated or revised when circumstances change.
IBM’s results are based on executives’ reported experiences rather than independent measurements of their organisations’ financial performance. The UK sample of 100 also means that percentages should not be treated as precise estimates for every British business.
Even with those limitations, the findings identify a substantial organisational issue. Companies increasingly expect AI leaders to manage performance, accountability and adoption, while established investment and workforce decision structures can limit their ability to respond. As autonomous systems become embedded in operational processes, that division of responsibility will require clearer rules and more effective coordination.












