Finance leaders are hearing a lot about agents right now: AI systems that don’t just answer questions but take action inside a workflow. The pitch is compelling — point an agent at your reconciliation, your close, your variance analysis, and let it run. Fewer hours lost to manual work, faster answers, less dependency on whoever happens to know the process.
Here’s the question that determines whether any of that works: what is the agent really acting on?
Most finance math has one right answer
Most of what finance does isn’t ambiguous. A royalty calculation has one right answer. A tax allocation has one right answer. Reconciling a delivery-partner payout against a point-of-sale record has one right answer, and it’s usually sitting in a spreadsheet formula somewhere, not something an LLM should be inferring fresh each time. This is deterministic work, governed by rules finance already knows rather than judgment calls a model is suited to make. The value an agent adds isn’t figuring out the math but acting fast on math that’s already correct, and being able to show its work when someone asks why.
That “why” is where most AI-in-finance efforts run into trouble because nobody can trace the answer back to the rule that produced it.
The logic isn’t missing, it’s just ungoverned
None of this is a new problem. Finance has encoded royalty formulas, allocation rules, and close procedures for as long as spreadsheets have existed, usually in a file only one or two people fully understand. That used to be a manageable risk: an analyst leaves, someone rebuilds the spreadsheet, work continues.
Handing that same process to an agent raises the stakes, because the rules now need to live somewhere everyone, including the AI, can actually see. That’s what a business logic layer does: it gives the logic finance already owns a place to live where it’s visible, repeatable, and auditable, instead of trapped in a file only one person understands.
What that looked like at Papa Johns
Papa Johns runs more than 6,000 stores worldwide through a franchise model, and every order that comes in through a delivery partner like UberEats or DoorDash has to be reconciled correctly, because the result determines what a franchisee actually gets paid. A reconciliation error doesn’t stay contained — it ripples into franchisee payments and the numbers that flow into the close.
The team had already invested in Google Cloud and BigQuery, but the reconciliation logic itself lived where finance logic usually lives: spread across systems, spreadsheets, and dashboards, reassembled by hand every close cycle. Michael Wyant, VP of Enterprise Data and Corporate Solutions at Papa Johns, named the gap “The missing piece has been the finance-specific business logic.”
So that’s where the team started. Not by deploying an agent, but by making the reconciliation logic itself visible and governed inside Alteryx One, running natively against BigQuery. The result: manual finance work cut by up to 400 hours a year, with SOX controls documented automatically instead of drifting out of date in a separate file. Only once that foundation was in place did the Gemini-powered reconciliation agent become something the business could actually trust with the answer.
What agents get once the foundation exists
Papa Johns didn’t get an AI agent by asking for one. It got one by making the underlying logic something an agent, an auditor, or a new hire could all see and rely on in the same way. The emerging connection between platforms like Alteryx and Google Cloud through the Model Context Protocol points toward more of this: agents that can act on governed data with the right context and permissions already built in, rather than raw access to whatever they happen to find.
None of that replaces the judgment finance brings to the process. It means that judgment gets applied once, written down clearly, and used the same way every time, instead of depending on one person’s memory of how it’s supposed to work.
What to ask before your team’s next agent conversation
Before your team’s next agent conversation, ask the question this piece opened with, but pointed at your own workflows: what would the agent actually be acting on? Papa Johns answered it by making the reconciliation logic visible before handing any of it to an agent. That’s the discipline worth applying to your close, your audit, or wherever your team is eyeing agents next.
Papa Johns, Alteryx, and Google Cloud are walked through exactly how they built that foundation, and what it took to trust an agent with the result, in this on-demand webinar . If your team is fielding the same agent questions, it’s worth your time.
On-Demand Webinar: How Papa Johns, Alteryx, and Google Solve 3rd-Party Reconciliation