86% of CFOs think building their personal AI literacy is important. Only about a third have actually deployed agentic AI at any meaningful scale.
That gap is the whole story.
And it doesn’t close with more enthusiasm. It closes with a foundation people trust enough to act on. The reason most teams stall between wanting AI and deploying it at scale isn’t the model; it’s that they can’t yet stand behind the data underneath it.
That’s the problem we work on every day at apexanalytix: building AI on top of complex transaction data, at scale, without adding new risk in the process.
I spent three days at Gartner Finance Symposium/Xpo in National Harbor last week. apexanalytix sponsored the coffee station, which felt appropriate. You want people alert for this stuff.
Two sessions in particular have been sitting with me since. Not because they were surprising. But because they named things that are genuinely hard to say out loud in a room full of finance leaders.
AI Is Already Here – Scale Is Not
Mike Helsel, a Gartner analyst who covers finance technology, opened with a visual of 47 technologies finance teams are navigating right now. The actual AI-specific technologies on that list: nine. But that wasn’t the point.
The point was that AI is embedded in almost everything else on that list, including tools finance teams have been running for years.
- Robotic process automation (RPA) tools
- Analytics and business intelligence platforms
- Enterprise resource planning (ERP) systems
The AI didn’t announce itself, it just showed up inside the products you already own.
The most useful thing Mike said: your AI strategy has two lanes.
One runs through your existing vendors as they ship new capabilities. The other is purpose-built AI investment you make separately. Most organizations are only thinking about the second lane and ignoring the first, which is already moving.
The other observation worth carrying home: adoption and investment aren’t the same thing. The Gartner Bullseye research Mike walked through tracks 63 technologies across adoption, perceived value, and future investment. The AI quadrant lit up green, but most of those technologies are still in the “exploring” ring, not the center. Governance, notably, was sitting in the outer ring.
Finance teams are investing in AI faster than they’re governing it. That’s not a data team problem.
Three Traps CFOs Keep Walking Into
Valeria Di Maso’s session was a sharp edge. She covers data and analytics at Gartner, and she spent 30 minutes doing something rare: telling a room full of senior finance executives exactly where they are making it worse.
The three traps she outlined are worth repeating.
1. Waiting for Perfect Data Is a Losing Strategy
The first is waiting for perfect data before starting.
Valeria called it the perfection trap, and she was not gentle about it. Finance teams that are still “cleaning data” two years after an AI investment aren’t facing a data problem. They’re facing a decision problem.
AI doesn’t need perfect data. It needs fit-for-purpose data, meaning data that is complete enough, owned clearly enough, and refreshed quickly enough to support a specific decision. That’s a CFO call, not a data team deliverable.
Here’s where I’d push the point a step further than the session did. You don’t get fit-for-purpose data by cleaning your own records in isolation. The fastest path is to validate what you already have against authoritative third-party sources, government databases, regulatory lists, and other trusted entities, so gaps and errors surface against an external benchmark instead of your own assumptions.
It’s a lesson we’ve learned the hard way in the procure-to-pay world: a third of organizations have no regular process for cleaning their data at all, and the ones that do aren’t doing it often enough. The internal record is rarely the most current version of the truth.
Treating trusted external data as the reference point is what turns “good enough” data into data you can actually put a decision on.
2. Garbage In, Gospel Out
The second trap is assuming AI will fix your data problems on the way through… It won’t.
AI amplifies whatever you give it, good or bad. The “garbage in, gospel out” version of this is when AI returns confident, well-packaged analysis built on bad inputs, and the finance team treats it as authoritative because it looks authoritative. The machine doesn’t know it’s wrong. It just knows how to sound right. The CFO is still accountable for that decision. AI doesn’t absorb ownership.
The practical defense is to fix the inputs before the model ever touches them. We see what bad inputs cost in the clearest possible terms: nearly 30% of duplicate payments trace back to vendor master errors, things as ordinary as a duplicate record or an inconsistent supplier name.
No amount of AI sophistication downstream corrects for that; it just packages the error more convincingly. Scoring and validating the data going in, against trusted external sources, is what keeps confident analysis from being confidently wrong.
3. The Risk No Dashboard Will Show You
The third trap is setting AI live and walking away.
Model drift is quiet. It doesn’t fail immediately. The forecast still updates, the dashboard still shows numbers, the alerts still fire.
What changes slowly is whether those outputs still reflect reality. Governance is the only thing that catches this, and governance requires knowing who owns the output, who checks it, and at what moments a human has to reenter the loop.
The control loop framing she used was the most useful thing I heard all week:
- Use AI to identify patterns and for detection
- Let your team apply context and judgment
- Keep accountability for the final decision with a personn
Not because AI is wrong. Because ownership is not transferable.
This is the part we’ve had to build for real. Catching drift isn’t a quarterly review; it’s continuous monitoring of the things that change underneath a decision, sanctions status, bank account changes, financial health, so the system flags movement the moment it happens rather than at the next audit.
And when AI acts on its own, the loop only holds if you can see inside it: every action leaving an audit trail of why the agent acted, what it changed, and how to reverse it. That’s what makes “keep a human in the loop” an operating model instead of a slogan.
You can only own a decision you can see and undo.
What AI Governance Looks Like in Procure-to-Pay
Both sessions were, explicitly, about finance technology strategy. But the underlying question was the same one that shows up the moment AI meets the procure-to-pay cycle:
- The data is messy
- Transaction volumes are massive
- A single mistake can move real money
- AI can automate more of the process
- But where should human judgment remain in the loop?
The answer these analysts gave is pretty close to the one we’ve landed on.
Start with the data you can trust, define the decisions it needs to support, and keep humans in the loop on anything that has downstream consequences.
For a platform that monitors supplier risk and protects against overpayments across trillions in spend, that’s not a philosophy. It’s an engineering requirement.
If you want a concrete place to start: pick one AI-assisted decision you already rely on, and ask where its data comes from, who owns it, and how you’d know if it drifted.
If you can’t answer all three, that’s your first project, not your next model. And if you want to compare notes on where your organization sits on the adoption-versus-governance curve Valeria described, it’s worth a conversation.
The Most Valuable Insight Came From Customers
Honestly, the highlight for me was our customers seeking us out.
Familiar faces from Chick-fil-A, Howard Hughes Medical Institute, and Albertsons stopped by the booth, not because we paged them, but because they wanted to. Many of them were friendly faces I’d just seen at our Icon Conference back in April, and it was so good to see them again.
These are people who use the platform every day, and those conversations were genuinely energizing, the kind that remind you why this work is worth doing.
You can sit through every session on the agenda and learn a lot, but nothing beats hearing directly from the teams who trust you with their supplier data and their spend, and for whom you recover millions of dollars. That kind of relationship is what tells you the work matters.
It’s also, quietly, what sits behind the recognition we picked up this year: apexanalytix was named a Leader in the 2026 Gartner® Magic Quadrant™ for Supplier Risk Management Solutions, recognized for both Completeness of Vision and Ability to Execute.
That placement is gratifying but the customers who came to find us at the booth are the proof behind it.
Good sessions. Great coffee (we made sure of that). And the best customers in the business.
