With his background as an early blockchain adopter and extensive experience advising FinTech startups, Nicholas Braiden has a unique perspective on the technologies reshaping digital finance. He advocates for the transformative potential of financial technology to redefine everything from payments to lending. Today, we delve into the practical application of agentic AI in the enterprise, exploring how autonomous systems are moving beyond experimentation to deliver significant returns. Our discussion will cover the impressive ROI these agents are generating, particularly in accounts payable, the strategic decision between building versus buying these capabilities, and how to implement a governance framework that accelerates, rather than hinders, adoption. We will also touch on how this technology is fundamentally shifting the nature of work for finance professionals.
Autonomous agents are delivering an 80% average ROI, significantly outperforming general AI projects. What specific capabilities drive this performance gap, and how should a CIO rethink their automation budget allocation to capitalize on this? Please share some metrics you’ve observed.
The performance gap is really about one thing: action. While general AI projects, which delivered a respectable 67% ROI, are fantastic at summarizing data or drafting text, they typically stop short of execution. They generate an insight, but a human still has to interpret it and then take the next step. Autonomous agents are different. They close that gap between insight and action. They don’t just identify a potential fraudulent invoice; they can quarantine it, flag it, and route it for a specific type of review, all within predefined rules and approval thresholds. That 80% average ROI comes from agents handling complex, end-to-end processes without constant human intervention. For a CIO, this means automation budgets can’t just be about analytics anymore; they must be reallocated toward systems that embed decision-making directly into the workflow, turning a cost center like AP into a hub of autonomous efficiency.
Many finance teams have rolled out AI agents as experiments rather than to solve specific business problems. Why is patience for this experimental phase running out, and what steps can leaders take to pivot from testing capabilities to achieving tangible returns on their AI investments?
Patience is wearing thin because the boardroom pressure is immense. We’ve reached a tipping point where, as Basware CEO Jason Kurtz puts it, CEOs are done with AI experiments and are demanding real results. The data is telling: a staggering 61% of finance leaders admit their initial AI agent rollouts were primarily to test capabilities, not to solve a concrete business problem. This approach of “AI for AI’s sake” is a surefire way to waste resources. To pivot, leaders must first shift their mindset from a technology-first to a problem-first approach. Instead of asking “What can this AI do?”, they should ask “What is our most painful, high-volume, rules-based process?” The data shows that 71% of teams achieving weak returns acted under pressure without clear direction. The first step is to anchor any AI initiative to a specific, measurable business outcome, like reducing duplicate payments or accelerating invoice processing time, and then deploy the agent with that singular purpose.
Accounts payable is often cited as the primary proving ground for agentic AI in finance. Could you walk us through a specific AP workflow, like duplicate invoice detection, explaining how an autonomous agent handles it differently from both a human employee and a traditional AI model?
Absolutely. Let’s take duplicate invoice detection, a classic AP headache. A human employee relies on memory and manual checks—sifting through records, comparing invoice numbers, amounts, and vendor names. It’s tedious, slow, and prone to error, especially at scale. A traditional AI model is a step up; it might use pattern recognition to flag a list of potential duplicates, but it still creates a work queue for a human to investigate and resolve. The agentic AI system transforms this entirely. Because it’s trained on a massive dataset—Basware’s, for instance, uses over two billion processed invoices—it understands context. It doesn’t just match numbers; it can differentiate between a true duplicate and a legitimate recurring monthly invoice from the same vendor. More importantly, it then acts. It can automatically reject the duplicate, notify the vendor of the error, and close the loop, all without a human ever touching it. It’s the difference between being a helpful analyst and being a diligent, autonomous worker.
We’re seeing a trend where leaders prefer buying embedded AI for shared processes like AP but building it for unique functions like FP&A. What key factors should a CTO consider in this build-versus-buy decision, and what are the trade-offs between vendor dependency and competitive advantage?
This split reflects a very pragmatic approach. The core question a CTO must ask is: “Will this AI capability make us better at a standard industry process, or will it create a unique competitive advantage?” For a process like accounts payable, which is fundamentally similar across most organizations, buying an embedded solution from a vendor makes perfect sense. About 32% of leaders prefer this route for AP. The vendor has the massive dataset, the pre-built workflows, and the expertise, allowing a company to accelerate a standard process without reinventing the wheel. The trade-off is some vendor dependency, but the benefit is speed and reliability. For something like financial planning and analysis (FP&A), however, the models and data might be core to a company’s unique strategy. Here, 35% of leaders opt to build in-house because the AI itself becomes a competitive differentiator. The trade-off is a higher upfront investment and longer development time, but the reward is a proprietary system that competitors can’t replicate. The rule of thumb is simple: buy to accelerate, build to differentiate.
Nearly half of finance leaders are hesitant to deploy agents without clear governance, yet the most successful firms use governance to scale faster. How does an organization build a governance framework that enables speed and trust, rather than creating bottlenecks? Can you outline the first three steps?
This is a critical paradox, but it makes perfect sense. Weak governance creates fear and hesitation, while strong governance builds the confidence needed to scale. It’s not about restriction; it’s about creating safe, reliable guardrails. The first three steps to building an enabling framework are clear. First, define the operational boundaries. This means setting explicit approval thresholds—for example, an agent can autonomously process any invoice under $1,000, but anything over that requires human sign-off. Second, establish a clear “human-in-the-loop” protocol. Determine who is responsible for overseeing the agent’s decisions and at what specific checkpoints their intervention is required. Third, create a transparent audit trail. Every action the agent takes must be logged and easily reviewable, ensuring full accountability. This disciplined approach is why leaders with strong governance are far more likely to trust agents with complex tasks like compliance checks—50% of them do, compared to just 6% of their less confident peers.
The idea of treating AI agents like “junior colleagues” is compelling. How does this analogy translate into practice regarding training, setting approval thresholds, and gradually increasing autonomy within a finance department? Could you provide a real-world example of this process?
This analogy, which Anssi Ruokonen at Basware articulated so well, is the perfect mental model. You wouldn’t hire a new junior analyst and immediately give them the authority to approve a million-dollar payment. You build trust over time. In practice, this means starting an agent on a very tight leash. For example, in an invoice processing workflow, you might initially have the agent only perform data extraction and initial coding. A human then reviews 100% of its work. Once the agent consistently demonstrates high accuracy, you graduate it. You might allow it to fully process invoices from a trusted, high-volume vendor up to a small threshold, say $500, with only spot-checks from a human. As it continues to perform flawlessly, you gradually raise that approval threshold and expand its responsibilities to more complex vendors or invoice types. It’s a crawl-walk-run approach that mirrors how we onboard and develop human talent, building organizational trust and confidence with each successful, controlled step.
A third of finance leaders believe agentic AI is already causing job displacement. In your view, how does this technology truly shift the nature of a finance professional’s work? What new skills will be most critical for finance teams to develop over the next few years?
While the fear of displacement is understandable, I believe the reality is more of a fundamental shift in the nature of the work. The goal here isn’t just task efficiency; it’s about achieving greater operating leverage. By automating the manual, repetitive tasks like extracting data from PDFs or chasing down invoice approvals, you free up your highly skilled finance professionals to focus on far more valuable activities. Instead of just processing transactions, they can now analyze spending trends, optimize working capital, and provide strategic advice to the business. The skills that will become most critical are analytical and strategic. Finance professionals will need to be adept at interpreting the data the AI systems serve up, managing the AI agents themselves, and collaborating with the business to make smarter liquidity and forecasting decisions. The future finance professional is less of a bookkeeper and more of a strategic business partner.
What is your forecast for agentic AI in finance over the next three to five years?
Over the next three to five years, I forecast that agentic AI will become as standard in the finance tech stack as ERP systems are today. The “experimental” phase will be a distant memory, and the conversation will shift from “if” we should deploy agents to “how” we can best manage a hybrid workforce of human and digital employees. We’ll see agents move beyond AP into more complex areas like treasury management and compliance monitoring, performing sophisticated, multi-step tasks. The most successful finance organizations will be those that have mastered the governance and human-machine collaboration required to get the most out of their digital colleagues. This won’t just be about closing the books faster; it will be about creating a finance function that is truly predictive, strategic, and a core driver of the enterprise’s competitive advantage.