3 min
Everyone in AP Knows They Need AI. Almost Nobody Knows What to Do Next.
By: Daniel Shore
A few years ago, I walked into a budget meeting. I was working at a mid-sized company. The CFO had one item he kept circling back to: AI.
Where were we on it? What was the plan? When would we see results?
The question was fair - but it was the wrong room to be asking it.
An AI tool that your team does not trust and cannot audit is not an asset. It's a liability with a good pitch deck.
The AP team sitting across from him had heard the same thing from their last two managers. They had read the articles. They had tried the free tools. A couple of them had typed a question into ChatGPT once, gotten a confident wrong answer, and quietly closed the tab.
They knew AI was coming and yet they had no idea how to get there.
That gap – between knowing you need to do something and knowing how to actually do it – is where most AP teams are stuck right now. That's not going to change as fast as the vendors would like you to believe, at least in my opinion.
The mandate without the roadmap
AI in AP is not a question of awareness. Every AP leader I talk to knows it's a priority. Leadership has made that clear. In some cases, it's in their KPIs and performance goals.
The problem is the how. Nobody handed them a roadmap. Nobody trained them on what to trust, what to audit, or where the risk is. They're being asked to modernize a function that cannot afford mistakes, using a tool that still gets things wrong in ways that aren't always obvious.
That's not resistance. That's caution and prudence. And in AP, caution is usually the right instinct.
The firm had been paid
I've heard this come up more than once: a major professional services firm, one of the largest in the world, submitted a document full of AI-generated research citations. Most of them didn't exist. The firm had paid a significant amount for work that turned out to be, in large part, fabricated.
That story travels. AP professionals hear it and think: if a firm that sophisticated can get burned, what does that mean for us?
And they're right to think that. AP is not a function where you can publish a correction and move on. If an invoice gets processed incorrectly because a model hallucinated a vendor name or misread a PO number, the downstream consequences are real. Suppliers get shorted. Audits get complicated.
Relationships get damaged.
The fear is not irrational. The tool is still genuinely new, still in its infancy. And the margin for error in AP is extremely low. AP professionals like myself sometimes get called boring – but you know what boring is? Boring is cautious, boring is calculating, boring gets important things right.
Where companies are actually making progress
The teams I've seen move forward on AI are not the ones who decided to figure it out on their own. They're the ones who found a budget line that wasn't the normal AP budget.
At one company I worked with, AI adoption happened because their IT team had a dedicated AI initiative with its own funding. AP didn't have to compete for resources against headcount and software renewals. They got a seat at a different table and moved faster because of it.
That's the practical lesson: if you're waiting for AI to fit into your existing AP budget conversation, you may be waiting a long time. The companies making progress are finding creative paths to funding and partnering with IT or operations, not going it alone.
What actually comes next
The honest prediction is that AI in AP will not look like a single transformation moment. It will look like incremental, boring, high-value progress in specific areas.
Invoice matching. Exception flagging. Duplicate detection. These are not glamorous use cases, but they are where AI earns its keep in an AP context because they are high-volume, rule-bound, and verifiable. You can check the output. You can build confidence before expanding scope.
The teams that will be ahead in three years are not the ones who made the boldest AI bet today. They're the ones who picked a narrow problem, implemented something they could audit, built trust in the tool, and then expanded from there.
The worst thing you can do right now is nothing, because someone upstream is going to make this decision for you eventually. The second worst thing is to do too much too fast in a function where errors have real consequences.
Start small. Verify everything. Then go wider.
The vendor noise problem
One more thing worth saying: the AI marketing in this space is loud, and a lot of it sounds the same. Every solution will tell you it's intelligent, automated, and transformative. Very few of them will tell you what happens when it gets something wrong, how you catch it, and who is accountable.
Those are the questions worth asking before you buy anything. Not because AI is not useful, it is, but because the implementation matters as much as the technology.
An AI tool that your team does not trust and cannot audit is not an asset: It's a liability with a good pitch deck.
AP Trends & Predictions is part of an ongoing series from the AP Institute. See all of our content here.