There are increasing fears over the risks that come with what is being termed DIY AI.

It comes as a new report found finance leaders are increasingly worried about the security risks associated with building AI agents from scratch, with 67% saying ‘DIY AI’ could increase risk.

The global study, conducted by research house FT Longitude in partnership with invoice tech firm Basware said despite the security fears, six in ten (61%) finance leaders said they had already rolled out custom-developed AI agents ‘largely as an experiment’ to appease demands from their boss. This is despite one in four saying they have no idea what an AI agent actually does.

The study AI to ROI: Unlocking Value with AI Agents surveyed 200 CFOs globally, also revealed that 43% felt AI would cost them their job.

It added: “Finance leaders are no longer willing to assign budget to artificial intelligence (AI) experiments. Instead, they want to know where they can apply AI to get tangible, measurable outcomes.

“New Basware research highlights that agentic AI is the way forward. While overall AI return on investment (ROI) rose from 35% to 67% in the last year, agentic AI far outperforms the rest – with an average ROI of 80%. But these results are not universal, and not everyone is feeling AI’s benefits.

“Research by MIT has found that only 5% of AI pilot schemes achieve rapid revenue acceleration, meaning 95% don’t. This suggests the desire to use AI is not always matched by its returns – yet.

“Basware’s research suggests there is a way to get real value out of AI. The survey of 200 finance leaders finds that targeted, advanced AI agents can execute a specific job in the finance department to a high standard, converting hype into a tangible, operational ROI.”

Commenting on the research, Jason Kurtz, CEO, Basware said that businesses had ‘reached a tipping point’ around AI experimentation and that demand for tangible results was now a top priority.

“Finance teams that focus on areas where AI can have immediate impact, such as automating accounts payable, improving compliance, reducing errors, and detecting fraud,” Kurtz said.

The report also found that overall AI return on investment (ROI) rose from 35% to 67% in the last year, survey data shows agentic AI far – and companies using third-party solutions already embedded with AI agents – outperformed all categories with an average ROI of 80%. Four in 10 leaders also said they faced ‘pressure’ from their superiors to ‘do something with AI’.

Two-thirds (66%) of respondents to the survey say there is more hype around agentic AI than any previous technology shift, yet three-quarters are still figuring out the best way to leverage it. And the C-Suite is losing patience.

In all 72% of respondents said they see accounts payable (AP) – often the most manual and data-heavy part of the finance function – as the most obvious starting point for agentic AI.

“So, it’s no surprise that 44% of finance leaders are feeling pressure from leadership to ‘do something with AI.’ But what if they haven’t had any clear guidance about what that means? About 6 in 10 (61%) are rolling out agents just to see what they can do. Without clear objectives or an understanding of the problems AI can solve, it’s unsurprising that 95% of AI pilots fail to achieve revenue acceleration,” The report cautioned.