Finance teams are moving beyond AI pilots and proofs of concept. Their focus now is simple: measurable return on investment.
New research from Basware and FT Longitude shows that while general AI initiatives delivered an average ROI of 67 percent last year, agentic AI systems achieved around 80 percent by autonomously executing complex workflows. The difference is prompting CIOs and CFOs to rethink where they place their automation budgets.
Unlike generative AI tools that summarize reports or draft content, agentic AI systems operate within defined rules and approval thresholds to carry out tasks independently. In finance, that means moving from insight to action without pausing for manual intervention.
Boardroom Pressure Shifts AI Strategy
Nearly half of CFOs say they face pressure from boards and executive teams to implement AI across operations. Yet 61 percent acknowledge their organizations launched custom-built AI agents mainly as experiments to explore technical capability rather than solve specific business problems. That experimentation is losing support.
Jason Kurtz, CEO of Basware, says patience is wearing thin.
“We’ve reached a tipping point where boards and CEOs are done with AI experiments and expecting real results,” he notes. “AI for AI’s sake is a waste.”
Traditional AI models often generate predictions that still require human interpretation. Agentic systems close that gap by embedding decision-making directly into workflows. The result is faster execution and clearer accountability.
Why Accounts Payable Has Become the Proving Ground
Within finance departments, accounts payable (AP) has emerged as the primary use case for agentic AI. According to the report, 72 percent of finance leaders see AP as the logical starting point.
The reason is practical. AP processes are structured and repeatable. Invoices arrive, data is extracted and verified, compliance checks are performed, and payments are scheduled. These are high-volume, rules-based tasks that suit autonomous systems.
One in five finance leaders already use agents to automate invoice capture and data entry. Other applications include duplicate invoice detection, fraud identification, and reducing overpayments. These are operational tasks where systems can act with high autonomy when properly configured.
Data quality plays a critical role. Basware’s AI models are trained on more than two billion processed invoices, allowing them to distinguish between legitimate anomalies and genuine errors without constant human review.
Kevin Kamau, Director of Product Management for Data and AI at Basware, describes accounts payable as a “proving ground” because it combines scale, oversight, and measurable outcomes in a way few other finance functions do.
Build or Buy? A Strategic Decision
As adoption grows, finance and technology leaders face another question: should they build agentic AI systems in-house or embed them within existing software?
The answer varies by function.
In accounts payable, 32 percent of finance leaders prefer agentic AI embedded within vendor platforms, compared with 20 percent who build solutions internally. In financial planning and analysis (FP&A), however, 35 percent favor self-built systems, while 29 percent opt for embedded tools.
The pattern suggests a clear rule of thumb. For standardized processes common across many organizations, such as AP, buying embedded solutions can accelerate deployment and reduce risk. For functions that create competitive advantage, building internally may offer greater differentiation.
Governance Enables Scale, Not Delay
Autonomy raises understandable concerns. Nearly half of finance leaders say they would not deploy agentic AI without strong governance frameworks in place.
In regulated environments, caution is warranted. Autonomous systems must operate within strict guardrails.
However, the most successful organizations treat governance as an enabler rather than a barrier. Leaders confident in their oversight frameworks are far more likely to use agents for complex compliance tasks. About 50 percent of high-confidence organizations deploy agents in this way, compared with just 6 percent among their more hesitant peers.
Anssi Ruokonen, Head of Data and AI at Basware, recommends treating AI agents like junior employees. They require testing, oversight, and gradual increases in responsibility. A human remains accountable, particularly for significant decisions.
Rethinking Work, Not Eliminating It
The rise of digital workers has sparked concerns about job displacement. Around one-third of finance leaders believe workforce changes are already underway.
Supporters argue the shift is less about elimination and more about evolution. Automating repetitive tasks such as extracting information from PDFs frees finance professionals to focus on higher-value activities like liquidity management, faster financial closes, and strategic planning.
Organizations that deploy agentic AI consistently, rather than as isolated experiments, report stronger returns. Data from the study shows that 71 percent of teams with weak ROI acted under pressure without a clear plan. By contrast, only 13 percent of high-performing teams fell into that category.
The takeaway is straightforward. Agentic AI produces measurable impact when it is embedded into workflows with clear objectives and disciplined governance.
As finance leaders refine their approach, the message from early adopters is clear: purpose-driven automation, not experimentation alone, is what turns AI investment into sustained financial return.
