SAP study warns of AI sprawl across big businesses

SAP has released research describing AI sprawl across business functions. The study surveyed 2,600 business leaders in 13 countries.

The findings point to uneven adoption of artificial intelligence inside large organisations. Finance, legal, sales and marketing, human resources, and procurement each face different barriers to getting value from AI spending.

Finance teams are among the most advanced AI users, but they often deploy it within their own function rather than across the wider business. Finance respondents had invested more than any other function, with 65 per cent saying they were scaling or leading in AI automation and 52 per cent saying the same for generative AI.

Even so, only 16 per cent of finance AI is deployed across functions. Just 34 per cent of finance respondents said they were data-ready for AI, while integration and interoperability were cited as the main barriers to agentic AI.

Five fault lines

Legal teams showed a different pattern. Legal respondents said they apply AI to more tasks than any other function, at 33 per cent today, a figure expected to rise to 50 per cent within two years.

However, 49 per cent said AI was not delivering its full potential, the highest share of any function in the study. While 82 per cent reported a defined AI strategy at leadership level, governance readiness for AI processes and frameworks stood at 26 per cent, the lowest among the functions examined.

Sales and marketing teams were among the fastest movers on new AI uses. Those respondents ranked second for data readiness at 63 per cent and highest for piloting agentic use cases at 68 per cent.

That pace has also raised governance concerns. Eighty per cent of sales and marketing respondents viewed shadow AI as their most material AI risk, while 73 per cent reported using shadow AI at least occasionally.

Human resources was presented as a function with relatively strong generative AI maturity despite lower average spending. HR respondents reported one of the lowest average AI spending levels, at USD $25.1 million, yet 24 per cent said they were leading in generative AI implementations.

Data quality remains a weak point in HR, where only 37 per cent of respondents said they were data-ready for AI. The research argued that this gap is significant because HR handles sensitive workforce decisions and could play a central role in broader workplace change.

Procurement also stood out for adopting newer forms of AI despite more limited budgets than some other functions. Procurement respondents were ahead of most peers in agentic AI maturity and cross-functional AI deployments.

At the same time, 78 per cent of procurement respondents agreed that delivering return on investment from AI depends on data readiness, integration, and use-case readiness. Yet 86 per cent reported incomplete or inconsistent data, the highest level of any function covered in the study.

Integration challenge

The wider message from the research is that AI investment alone is no longer the main dividing line between success and failure. Instead, value now depends more on whether companies can link projects across departments, improve data quality, and set common governance standards.

"Organisations aren't missing out on value from AI because they lack ambition or investment. Instead, the dollars are falling through the cracks between functions," said Rachel Hunter, Head of AI, SAP Australia and New Zealand.

Hunter said the best-performing companies will be those that connect the approaches taken by different functions rather than letting each team work in isolation.

"The organisations that make the most of AI will be those who remove AI sprawl and connect the discipline of finance, the oversight of legal, the pace of sales and marketing, procurement's innovation, and the workforce focus of HR around trusted data and shared outcomes," said Hunter.

The study reflects a broader shift in how large companies are assessing AI programmes. Early efforts often focused on individual tools or department-led experiments, but the findings suggest the main problem for many businesses is now fragmentation rather than lack of interest.

That fragmentation can take several forms, including separate data systems, unclear rules on use, and uneven readiness between departments that depend on one another. In practice, a function that moves quickly can still fail to generate wider returns if the rest of the organisation cannot connect to its systems or trust its outputs.

For finance, the issue is extending tools beyond a mature core. For legal, it is matching rising deployment with proper oversight. For customer-facing teams, the concern is bringing unofficial AI use into approved systems. For HR and procurement, the challenge is improving data foundations before wider rollout.

"The organisations pulling ahead will treat AI as an enterprise operating model, not a series of departmental technology projects," said Hunter. "That means shared data, connected processes, clear governance, and a workforce equipped to use AI with confidence. Investment matters, but integration is what turns it into value."