Basware: Why are 61% of CFOs deploying AI blind?

Finance leaders are deploying artificial intelligence agents at pace, but most lack a clear understanding of what they are implementing or how to measure success, according to new research from Basware.

The company's AI to ROI report, produced by Financial Times Longitude and surveying 200 finance leaders globally, finds that 44% of CFOs face pressure from leadership to adopt agentic AI. 

In response, 61% are rolling out AI agents experimentally without understanding ROI or outcomes.

One in four finance leaders admit they do not understand what an AI agent looks like in practice. Three quarters are still working out how to leverage the technology effectively.

The uncertainty comes despite strong early returns. Finance leaders using AI agents report an average ROI of 80%, outpacing general AI investments which deliver 67%—up from just 35% in 2024. 

Some 72% of those surveyed report successful AI agent deployments, with 58% citing unexpected benefits beyond initial objectives.

AP leads as testing ground

Accounts payable has emerged as the primary deployment area, with 72% of finance leaders identifying it as the ideal starting point for AI agents. 

The process combines structured rules, repeatable workflows, and measurable outcomes that allow AI to operate at scale with clear accountability.

Current deployments concentrate on invoice capture and data entry automation (30% using daily), cash flow management (24%), and scenario modeling (23%). Finance teams are also applying agents to real-time market analysis, financial reporting, compliance checks, and fraud detection.

Kevin Kamau, Director of Product Management for Data and AI at Basware, explains the appeal: "AI is grounded in data, and AP is essentially a data-cleansing engine. Invoices go through the same processes every time. It's very rules based. This is the ideal playground for AI to automate."

Build or buy remains unresolved

Finance teams remain split on whether to build AI agents in-house or embed them in existing software, with equal proportions expressing concerns about each approach (34% and 35% respectively).

In accounts payable, 32% are implementing embedded agents versus 20% building custom solutions. For financial planning and analysis, the pattern reverses: 35% build in-house compared with 29% using embedded tools.

Kevin frames the decision: "If AI improves a process shared across many organisations, embedding it in software usually makes sense. But if AI creates an advantage unique to your business, that's where building becomes justified. Leaders should buy to accelerate, build to differentiate."

Performance gap widens between leaders and followers

The research reveals a sharp divide. Organisations achieving ROI above 101% – representing 23% of respondents – use AI agents far more extensively than lower performers.

Half of high performers use AI agents for compliance checks versus 6% of low performers. For cash flow management and financial reporting, the split is 46% versus 12% and 8% respectively.

Low performers display distinct characteristics: 71% feel pressure to "do something with AI" compared with 13% of leaders. Two-thirds worry about building agents from scratch versus 4% of leaders. More than half (57%) remain uncertain about what AI agents do, compared with 9% of leaders.

The pattern suggests ROI correlates with strategic deployment rather than experimentation driven by executive pressure.

Governance and workforce concerns persist

Nearly half of finance leaders (46%) will not deploy AI agents without established governance frameworks. 

Anssi Ruokonen, Head of Data and AI at Basware, advises a phased approach: "Choose one process where impact is easy to measure, define a clear governance model, and set decision thresholds so everyone understands when the AI can act and when a human needs to step in."

Job displacement remains a concern. One-third of finance leaders report AI is already displacing department roles, though the research suggests automation shifts workers toward higher-value strategic activities rather than eliminating positions.

The research points to a clear pattern: targeted AI agent deployment in well-defined processes delivers returns. Success depends on closed-loop execution that embeds decisions into operational workflows, not on model sophistication.

Finance leaders navigating AI adoption face a choice between experimental deployment under executive pressure and strategic implementation with clear governance and measurable outcomes. 

The ROI gap between these approaches is widening, and early evidence suggests those who move strategically now are building advantages that will be difficult for followers to close.