Finance teams aren’t just thinking about the role AI plays vs. people.
As AI agents in finance take on more work, they’re deciding where AI can accelerate—and most importantly, where human judgment, governance, and accountability need to remain.
In practice, agentic AI for accounts payable can complete defined tasks across the AP workflow—from capturing and coding invoices to matching purchase orders and routing approvals. But the more work AI performs, the more important visibility and human oversight become.
Recent Tipalti-sponsored research underscores that need. Nearly half of finance professionals surveyed—48%—are concerned about the risks of using AI.
And when evaluating AI products, 55% said the ability to see and review the actions AI takes was extremely important, while 50% said the same about retaining control of decisions and processes.
At our latest virtual event, Elevate 2026, I spoke with Moaaz Iqbal, group financial controller at Vivino, and Ho Yoon, VP of finance at DreamHost, about AI’s impact on AP, what trustworthy AI looks like in this space—and how people and AI can work together.
Key Takeaways
- Trustworthy agentic AI for accounts payable should make its work explainable, controllable, and auditable.
- Finance teams need new skills to evaluate AI output and should build confidence by focusing on clearly defined, lower-risk use cases.
- AI can complete routine work, but people must retain oversight, judgment, and accountability for the outcome.
1. Make agentic AI in AP explainable, controllable, and auditable
I opened our discussion by challenging the “false choice between human and AI.” To me, AI isn’t a direct substitute for a finance team, but it can be a valuable partner.
Still, for AI to be effectively deployed in AP, it needs to be trustworthy. Whether they’re evaluating individual AI agents or a broader AI-powered AP automation solution, finance leaders need to know how the technology reaches decisions and what controls remain in their hands.
Errors in AP can disrupt cash flow, expose sensitive data, and create compliance risks. That makes trust essential—but how should finance teams evaluate it?
Moaaz Iqbal drew on his experience at Vivino, where he’s working to integrate AI and automation into existing workflows.
“If something needs to be trusted, you have to go down, all the way down to the roots. It should be explainable. If it’s explainable, it should be controllable. If it’s controllable, it should be auditable,” he said. “That’s the chain that we normally follow as auditors to make sure that we understand something down to the roots from the top.”
“This is the same lens that auditors use to implement AI,” he continued. “They should know which part of the process within AI they can trust and expand upon before rolling it out to their clients.”
That framework—explainable, controllable, and auditable—offers finance teams a useful way to evaluate where and how they can trust AI.
If it’s explainable, it should be controllable. If it’s controllable, it should be auditable.
Moaaz Iqbal
Group Financial Controller, Vivino
2. Build the skills to evaluate AI output
Trust and transparency are cornerstones of successful AI integration. However, even trustworthy AI isn’t infallible.
Throughout our conversation, we agreed that AI output should be subject to the same scrutiny as work produced by human teams. But there’s one significant problem: scrutinizing an AI’s output is a little different. Without transparency, it can be difficult to review how an AI system reached its conclusion.
With people, it’s easy enough to ask for an explanation and redo calculations. The same can’t be said for AI. AI can sometimes seem like a black box: humans can’t always understand how it reached a conclusion, and the AI itself can’t always be counted on to explain it.
“The ability to actually review AI outputs is a skill in itself. You have to train the team on how to review those outputs,” said Ho Yoon. “It’s emerging as a new competency in the finance world.”
That point particularly resonated with me. As AI becomes more embedded in finance workflows, knowing how to question, test, and validate its output will become an essential skill—not just for managers, but for everyone on the team working with AI agents.
Finance leaders will need to prepare their teams to work with AI and treat reviewing its output as a core competency.
The ability to actually review AI outputs is a skill in itself. You have to train the team on how to review those outputs.
Ho Yoon
VP of Finance, DreamHost
3. Start small when adopting AI agents in finance
The pressure to adopt AI quickly is understandable. As AI agents in finance become more widely available, teams may feel pressure to implement them across multiple workflows at once. But in something as high-stakes as AP, speed without the right controls can introduce unnecessary risk. Finance teams are already raising valid concerns about data privacy, accuracy, and explainability. Addressing those concerns is essential.
In my work with Tipalti customers, I’ve seen that the organizations making the most meaningful progress aren’t necessarily the ones moving the fastest. They’re the ones moving confidently from one well-defined use case to the next. Incremental gains might not be headline news, but this measured approach builds organizational confidence while reducing the risk of serious errors.
This approach doesn’t benefit the business alone, however. It’s easy to forget that for many teams, being asked to suddenly adopt a new, rapidly changing technology can be overwhelming. Taking small but deliberate steps forward makes transformation seem achievable.
“For most of us, the small wins matter a lot,” Iqbal said. “That’s how I do it myself. I just divide it into small tasks and take those small wins, and then it normalizes. Before you know it, you’re just accepting the new AI world, which you thought was going to be a long journey ahead. Now you’re just a part of it.”
4. Choose the right AP work for AI agents
A true human/AI partnership has the potential to transform AP. But how do you determine what work goes to AI and what stays with human team members?
In many workflows, team members offload repetitive or boring tasks to AI. From the outside, it might sound like every AI-enhanced team has a clean division of labor, but the reality isn’t quite that simple.
Humans can’t just hand an AI agent a complex task with no context and expect a reliable result. Before AI can be of any use, teams need to understand the underlying process, establish clear parameters, and give the system the context it needs.
That’s especially important when introducing AI into invoice management and other interconnected AP workflows, where actions across invoice capture, coding, matching, approvals, payments, and reconciliation are interconnected.
“If I’m automating using AI, I first make sure that I understand how my AP process works and build an algorithm around it,” said Iqbal. “For example, if I want to pull out overaged payables, I need to make sure that I train AI on how I’ve used these filters within Tipalti or my ERP.
Once my AI engine understands how my filtering pattern works, it will adopt it, and then it’s just a click of a button away to get my desired reports.”
Preparing AI for business-specific tasks takes care and context. Teams also need a clear understanding of the technology’s current capabilities and limitations.
“I feel that with the training process, you have to be mindful and thorough with it. You also have to understand how your agentic AI is working and its current capabilities,” he added. “Your expectations and approach have to be aligned.”
5. Keep people accountable for the outcome
At least in some respects, introducing an AI agent into an accounts payable workflow isn’t unlike training a new coworker. And AI still needs clearly defined controls, appropriate review, and human accountability.
“The ownership of the work output is the thing that absolutely cannot be delegated,” said Yoon. “You can hand off anything that’s routine or repetitive to AI, and then you layer on top of that the human review of those outputs. It’s a great place to start, but I’ve also found value in reversing it, where you turn AI into the reviewer of human work as well.”
For me, this was one of the most important takeaways from our conversation: AI in AP is here to assist people, not to replace their judgment or accountability.
“It’s not only for the efficiency,” Yoon added. “This is how we empower. We need to empower our team members and equip them with the skill sets they’ll need for the rest of their careers. At the end of the day, each team member is responsible for his or her output, including anything generated using AI tools.”
Finance leaders aren’t working toward an all-robot workplace. They’re working toward teams that know how to use AI thoughtfully, review its output, and apply their own expertise where it matters most. That will take patience, experimentation, and a willingness to learn—but the payoff can extend beyond efficiency to stronger skills across the finance organization.
We really have to view this as a partnership between our people and their AI tools if we want those efficiency and empowerment gains.
Ho Yoon
VP of Finance, DreamHost
Building a high-trust human-AI partnership
It’s easy to try to pit humans and machines against one another. But my conversation with Moaaz and Ho reinforced that human/AI collaboration will be critical to the future of AP. The industry has already taken meaningful steps toward integration, but in the ever-changing AI landscape, uncertainty still remains.
The finance teams that succeed won’t necessarily be the ones that automate the most work. They’ll be the ones who understand what to automate, how to evaluate the results, and where human judgment needs to stay put. That’s the foundation of trustworthy, AI-powered AP automation: AI agents completing well-defined work while people retain visibility, control, and accountability.
If you want to hear more from my conversation with Moaaz and Ho, watch the full session from Tipalti’s Elevate online customer event.
