Agentic AI's Promise vs. Reality: Why CFOs Should Be Skeptical

The vendor pitch arrives with clockwork predictability: agentic AI is the next revolution in enterprise finance, a technology so transformative that CFOs risk obsolescence if they don't act quickly. Yet buried in the same breath comes a rather awkward admission from Gartner—more than 40% of agentic AI projects will likely be abandoned by 2027. For CFOs already navigating compressed margins and heightened scrutiny, the question isn't whether agentic AI represents genuine capability but whether the current push represents thoughtful deployment or vendor-driven hype.

What agentic AI actually means

The distinction between generative AI and agentic AI matters more than the buzzword evolution suggests. Generative AI, the technology that captured attention in 2022, remains fundamentally assistive. It synthesises information, drafts responses, surfaces insights. A finance professional still needs to review, approve, and execute.

Agentic AI operates differently. These systems don't merely suggest—they decide and act. According to Gartner's forecast, by 2028 a third of enterprise software will incorporate agentic capabilities, with 15% of daily work decisions made autonomously. In finance terms, that means systems that don't just flag overdue invoices but autonomously adjust collection strategies, modify payment terms, or escalate issues based on real-time assessment of customer risk profiles.

Enjoying this article?

Get stories like this in your inbox every week.

Join the Community

The use cases being promoted—autonomous invoice reconciliation, adaptive procurement workflows that reroute approvals when supply chain disruptions occur—sound compelling. What remains less clear is how many of these capabilities reflect actual deployment versus theoretical potential dressed up in vendor roadmaps.

When an AI agent autonomously approves a payment or adjusts credit terms, who carries liability if that decision proves flawed? The technology may be advancing faster than the compliance infrastructure needed to support it.

The governance implications deserve particular attention. Finance operates under regulatory frameworks where human accountability and audit trails aren't optional niceties but legal requirements. When an AI agent autonomously approves a payment or adjusts credit terms, who carries liability if that decision proves flawed? The technology may be advancing faster than the compliance infrastructure needed to support it.

The failure rate nobody wants to discuss

Gartner's 40% failure prediction should feature prominently in any honest assessment, not as a footnote. That figure suggests the technology remains immature, the implementation challenges substantial, or both.

Three factors appear to drive project abandonment. Costs escalate as organisations discover that effective agentic AI requires clean data architecture, integrated systems, and ongoing oversight—not simply licensing software and watching automation unfold. Business value proves elusive when agents operate in silos, unable to access the cross-functional data needed to make genuinely informed decisions. Risk controls lag behind capability, leaving finance leaders exposed to autonomous actions they can't adequately monitor or reverse.

Esker, the automation platform vendor whose content sparked this analysis, acknowledges this reality even whilst promoting agentic capabilities. Their argument—that standalone AI tools fail without unified digital infrastructure—happens to be correct. But it also reveals how much foundational work most organisations still need before agentic AI becomes viable.

Most finance departments haven't yet built the integrated data platforms and end-to-end process automation that would allow agentic AI to function effectively. They're being sold Formula One technology whilst still driving on gravel roads.

What finance leaders should actually consider

The comparison to "the advent of the internet" that occasionally surfaces in agentic AI marketing requires substantial qualification. The internet created entirely new categories of activity and commerce. Agentic AI, at least in its current enterprise finance applications, represents incremental automation of existing processes—meaningful, potentially valuable, but hardly civilisation-shifting.

CFOs facing genuine margin pressure might reasonably ask whether their capital is better deployed perfecting existing automation, cleaning data infrastructure, and training staff on generative AI tools they already own. The most sophisticated technology delivers negligible value if the organisation lacks capacity to implement it properly.

What's interesting here is the speed at which vendors have pivoted positioning. Generative AI emerged less than three years ago. Many finance functions are still determining optimal use cases for assistive AI. Yet the message has already shifted to autonomous execution as the "real" frontier, with generative capabilities implicitly downgraded to table stakes.

That acceleration suggests vendor anxiety as much as technological maturity. The enterprise software market demands continuous innovation narratives. Whether enterprises actually need those innovations on the proposed timeline is a separate question.

The organisations most likely to extract value from agentic AI in the near term are those already operating with robust automation, clean master data, and integrated financial systems—precisely the enterprises that need it least urgently. For the majority still wrestling with basic process standardisation and data quality, agentic AI risks becoming an expensive distraction from more fundamental work.

Watch whether early adopters can demonstrate sustained ROI beyond pilot programmes, and whether regulatory frameworks evolve to accommodate autonomous financial decision-making. Until both those conditions materialise, finance leaders might serve shareholders better by treating agentic AI as emerging capability rather than urgent imperative, regardless of how many vendors insist otherwise.

More articles by Ross Williams