Originally aired March 16, 2026 on Extropy On AIr.
Most AI pilots in finance stall because they stop at insight. The teams seeing real ROI are the ones letting AI agents execute work, like routing, holding, escalating and posting invoices, inside clear guardrails. In this episode, Kevin Zirkle of Extropy Advisors and Adam Lustig of Basware unpack why accounts payable is where agentic AI is proving itself first, and what separates the leaders from everyone still experimenting.
Listen: Extropy On AIr podcast page | Spotify
Key takeaways
- MIT research found 95% of AI pilots fail to accelerate revenue. The problem usually isn’t the technology, it’s stopping at insight instead of execution.
- In Basware’s “AI to ROI” research, AI ROI expectations climbed from 35% to 67%, and agentic AI averages around 80%.
- 72% of finance leaders say AP is the most obvious starting point for AI agents.
- Buy to accelerate, build to differentiate. Standard processes like AP rarely justify a custom build.
- Governance is infrastructure, not friction. Clear guardrails are what make autonomy possible at scale.
Why do most AI pilots in finance fail?
Kevin opened with a stat that surprised him: according to MIT research, 95% of AI pilots fail to achieve revenue acceleration. Adam’s take is that most pilots are built to produce insight. They surface a dashboard or a recommendation, and then a person still has to do the work. Finance doesn’t reward interesting dashboards. It rewards throughput, accuracy and compliance.
Agentic AI changes that. Instead of telling someone an invoice needs attention, an agent routes it, holds it, escalates it or posts it, all within guardrails the customer defines. That shift from informing work to performing work is where the return shows up.
Why is accounts payable the best place to start with AI agents?
Adam describes AP as structured repetition at scale. Invoices arrive in volume, data gets extracted, validated, matched, approved, posted and eventually audited. It’s rules driven but exception heavy, which is exactly where agents do well.
His analogy is a factory floor. You don’t have an engineer inspecting every nut and bolt on every car. You automate the repeatable work and let experts handle what actually requires judgment. AP works the same way, and the research backs it up: 72% of finance leaders point to AP as the most obvious starting point, and the majority of teams that have deployed agents there report success.
What separates AI leaders from followers in finance?
The Basware report shows a wide spread in results. 23% of organizations reported ROI above 101%, while 42% saw 50% or less. Adam boiled the difference down to three questions every finance leader should be able to answer:
- Can you name a workflow where success is defined by a real number today?
- Do you have rules that define when an agent acts versus when a human steps in?
- Is AI on your quarterly operating agenda, or is it still an annual strategy topic discussed at the board level?
The teams answering yes tend to see the highest ROI and feel the least pressure, because usage builds confidence and confidence expands adoption.
Should you buy or build AI agents for finance?
Adam’s rule is simple: buy to accelerate, build to differentiate. If AI improves a standard process like accounts payable, an embedded solution almost always wins on time to value. If AI creates something genuinely unique to your business model, building can make sense.
He shared the example of a Basware customer, a financial services firm that spent 18 months building its own invoice automation agent. It worked well for one country, but the firm operates in 14 markets, each with different tax rules and invoice formats. Meanwhile, a competitor deployed an embedded AP solution trained on billions of invoices and hit high automation rates in months. Owning the code doesn’t create ROI. Execution does.
Is AI governance slowing finance teams down?
46% of finance leaders say they won’t deploy an AI agent without proper governance in place. Some read that as caution or friction. Adam sees it as infrastructure. When you hire someone into AP, you don’t tell them to post whatever they think is right. You define approval levels, escalation paths and oversight. Agents need the same thing, and those guardrails are what make autonomy possible at scale.
Will AI agents replace AP jobs?
Roughly a third of respondents in the research say agents are already displacing some work. Adam’s point is to separate tasks from roles. Agents take on the repetitive pieces, like data capture, invoice matching, coding and compliance checks. People move toward the work that needs judgment and relationships.
One global consumer goods company used that shift to refocus its AP team on supplier relationship management and early-pay programs. In the first year, they recovered hundreds of thousands in early-pay discounts they had been missing.
What does it mean that agentic AI is an operating model decision?
The line Kevin pulled from the report as the most important: agentic AI is becoming less of a technology choice and more of an operating model decision. In five years, the finance teams that feel great about their AI journey won’t be the ones that ran the most pilots. They’ll be the ones that picked one workflow, made it work, built confidence and scaled from there.
In their words
Finance doesn’t reward interesting dashboards. It rewards throughput, accuracy, and compliance.
Adam Lustig, Basware
People hear the word governance and think friction, where I think of it more as infrastructure.
Adam Lustig, Basware
About the speakers
Kevin Zirkle is Director of Business Development at Extropy Advisors, where he works across Extropy’s partner ecosystem, matching organizations with the right technology and the right people to move transformation from the whiteboard into production.
Adam Lustig is VP of Solution Value Consulting at Basware, a leader in AP automation and e-invoicing.
FAQ
What is agentic AI in accounts payable?
Agentic AI goes beyond flagging problems or producing insights. AI agents actually do the work, such as routing invoices, placing holds, escalating exceptions and posting transactions, within guardrails the business defines. In AP, that shift from informing work to performing work is where the ROI shows up.
Why is accounts payable a good starting point for AI agents?
AP is structured repetition at scale. Invoices arrive in volume and follow a clear path of capture, validation, matching, approval, posting and audit. The process is rules driven but exception heavy, which is exactly where agents perform well. In Basware research, 72% of finance leaders called AP the most obvious place to start.
Should finance teams buy or build AI agents?
A simple rule from the episode: buy to accelerate, build to differentiate. If AI improves a standard process like AP, an embedded solution almost always wins on time to value. Build only when AI creates something unique to your business model.
Keep exploring
- More episodes of Extropy On AIr
- Previous in the Basware series: Unlocking Intelligent Invoice Lifecycle Management with Basware
- Previous episode with show notes: Entropy vs. Extropy: Why Digital Transformations Fail
- Next in the Basware series: AP Is Not an Afterthought, with Basware CEO Jason Kurtz
- Next episode: AI in Finance Transformation: What SSON Revealed and the Dexcom Win
- Why Extropy
Ready to move from AI pilots to real AP results? Book a discovery call with our team.
