AI hype cycles have accelerated, from initial automation to copilots, agentic AI and the ‘AI eats software’ debate. The struggle that persists among CFOs is where to start and how to drive real value, against the backdrop of mounting leadership frustration towards AI.
Research from Basware found that 61% of global finance leaders say their organization has rolled out customer-developed AI agents largely as an experiment, simply to see what the technology can do. While a quarter admit they still don’t fully understand what an AI agent looks like in practice.
But the experimentation era is closing, with boards and the C-suite losing patience, demanding return on investment.
The question isn’t whether to adopt AI, it’s about how to deliver ROI quickly, forcing a decision on when businesses should build their own AI systems versus when they should buy pre-made, specialized AI solutions.
Why finance can deliver faster AI ROI
The office of the CFO is a natural starting point for AI adoption. Finance processes deal with vast volumes of data from diverse sources — supplier invoices in varying formats, unstructured documents, and cross-system exceptions — that require interpretation, pattern recognition, and contextual judgment at a scale no manual team can sustain. This creates an environment where AI’s impact is both significant and measurable, whether it’s hours saved for staff members or metrics such as cashflow.
Almost three-quarters of finance leaders see accounts payable (AP) as the most obvious starting point for agentic AI, in part due to being the most manual and data-heavy part of the finance function. AI agents here can automate back-office tasks within these processes, facilitating a touchless invoice process which only flags issues when they arise and demands human expertise. Beyond automation, AI also enables ad-hoc analysis and problem-solving capabilities that weren’t previously feasible at scale. And importantly, the AP process doesn’t sit in isolation — it touches procurement, business units, and suppliers alike, meaning AI improvements here ripple across a far wider audience than the finance department alone.
For CFOs, AI impact can be tied to measurable cost savings, reduced error cycles, improved compliance, and improved working capital visibility, all of which contribute to improving overall cashflow for the business.
Finance already measures ROI closely, making AI’s impact easier to prove, moving it from speculative innovation to operational optimization that the entire business feels.
The next step is how to go about implementing AI and whether to build in-house or buy a ready-made solution.
The case for building
The biggest indicator for when to build AI solutions is when the target process is central to competitive advantage, directly impacting revenue, margin, or risk. If off-the-shelf tools don’t align closely with operational reality and internal data is a key strategic asset, then the potential for AI ROI is there.
Strengthening the brand moat is essential as AI increasingly becomes a differentiating ingredient inside software.
When building, businesses gain full control over data, the ability to tailor models to unique outputs, and full traceability to support with audits.
Proprietary risk models are a strong candidate for in-house AI solutions, for example, as outputs link directly to an area of leadership scrutiny, and the ability of custom models to improve over time as internal data compounds. Here, bespoke AI becomes embedded into decision-making, underpinned by continuous learning and improvement.
However, it comes at the expense of a higher upfront cost, a need for AI governance, and ongoing maintenance for the AI solution, but if the competitive advantage is there and the AI model is implemented successfully, then the cost is more than worth it.
Importantly, not every finance workflow needs bespoke AI. Many shouldn’t. By understanding that, and working on the specific area of competitive advantage, finance teams can drive AI ROI through the hype.
The case for buying
Buying ready-made AI solutions from specialists isn’t less strategic, it’s a smart allocation of resources when not tied to the core value proposition.
Standardized, efficiency-focused functions such as invoice processing are perfect for pre-built AI solutions. That workflow is core to another organization’s value proposition, so they have already gone through the process of building a bespoke system with their industry expertise. Basware, for example, has managed over $10 trillion in spend, having processed invoices for over 40 years, already building the industry’s largest data set and market trust.
Buying facilitates faster deployment, reduced implementation risk, access to vendor expertise, and predictable ROI timelines as a result. If it’s a case of automating an inefficient process, buying can mitigate the stress and potential pitfalls.
The added benefit of buying is allowing teams to focus on the strategic AI building in relation to the core value proposition. It’s about buying the infrastructure when the process is standardized and high volume, and building when it’s tied to competitive differentiation.
Scoring easy wins with leadership
The ultimate goal when implementing AI in finance is disassociating with experimentation, which leadership is sick of, and linking projects directly to ROI.
CFOs should therefore prioritize low complexity use cases, often involving high volumes of data, that have clear measurable outputs.
After AP, the top three agentic AI deployments being considered by finance leaders are automating invoice capture and data entry, cash flow management, and scenario modeling and forecasting. These use cases each generate tangible results that are easy for CFOs to quantify and communicate.
When presenting results to leadership, this could be framed as or processing time reduction or recovered revenue from fraud prevention, for example, tied to risk reduction and operational improvement rather than simply an innovation theatre.
These visible wins shift the narrative from experimentation to outcomes, building internal trust and, in many cases, unlocking future budget for AI projects.
