There’s a version of this that plays out in a lot of finance orgs right now. A process has been slow and manual for years, everyone’s tired of it, and someone suggests the obvious fix: point AI at it. Skip the boring middle step and go straight to the thing that sounds like progress.
It’s an understandable move. It’s also the reason a lot of AI pilots in finance don’t go anywhere.
What has to happen first
AI needs a clear, consistent version of a process to work from. If that process is still scattered across spreadsheets without one clear way it’s supposed to run, there’s nothing stable for AI to learn from. It’ll still produce something. It just won’t be something you can trust, repeat, or explain to an auditor.
That first step doesn’t get much attention, but it’s the one that matters. It’s making the process visible enough that someone new could follow it, consistent enough that it runs the same way every time, and documented well enough to survive someone asking how you got the answer.
None of that is an AI project. It’s building the process into an actual workflow that is easily understood by everyone. Enough of that built out across a team, and that’s what people mean by the business logic layer. AI comes after, because AI needs something real to check its work against.
Three companies did this in very different corners of finance, and the pattern holds every time.
Sunbelt Rentals: governance before the AI conversation even started
The finance, accounting, and operations teams at Sunbelt Rentals had run the same manual, mostly-Excel processes for twenty years. When the company brought in a Director of Finance Technology and Transformation, she treated those twenty years as something worth building on, not starting over. “The goal is to leverage what people are already doing,” said LaShell Estes.
Estes started by getting finance, IT, and the business aligned, then trained people on the new tools until using them was second nature. Only once specific processes had been rebuilt as governed, repeatable workflows, with SOX controls built into how they ran, did the team turn toward AI, using it to read invoice PDFs and check whether use tax had already been paid. That last step worked because everything under it was already solid.
Crowe LLP: the same order, in a tax practice
The tax professionals at Crowe LLP were dealing with a more scattered version of the same problem. Different groups used different tools, some still lived entirely in Excel, and there was no consistent way to scale any of it. “There was no consistency or ability to scale,” said Hannah Brauer, Tax Transformation Manager at Crowe.
The fix wasn’t an AI rollout. It was the Alteryx Innovation Program, which trained more than 190 tax professionals and set companywide standards for how workflows got built, governed, and documented, so the same process ran the same way no matter who touched it.
That’s the business logic layer showing up again, just built for a tax practice instead of an equipment rental company. AI only entered the picture after that foundation existed. “We’re using Alteryx’s embedded AI features to rethink how workflows are orchestrated and delivered,” said Tracey Grant-Castleman, Tax Transformation Leader at Crowe, describing what became possible once the governance work was already done.
This is closer to normal than you’d think
None of these companies was unusually behind. PwC found that even among companies with more than $10 billion in revenue, 38% still manually collect and consolidate their forecasting data. Among companies between $1 billion and $10 billion, that climbs to 52%. Most of finance hasn’t finished the unglamorous part yet, which means most finance teams aren’t actually ready for the AI conversation they’re trying to have. That’s the current state of the industry, plain and simple.
Papa Johns: same order at a much bigger scale
Papa Johns ran into this exact sequencing question with something higher-stakes: reconciling delivery-partner payouts across more than 6,000 stores, where getting it wrong shows up directly in what franchisees get paid. The order was the same as Sunbelt’s and Crowe’s. Build the workflow, make it governed and repeatable, and only then hand part of it to an AI agent.
Papa Johns, Alteryx, and Google Cloud walked through exactly how they did that in this on-demand webinar.
If your team is eyeing AI for a process nobody’s actually made repeatable yet, it’s worth seeing what the order looks like before you skip the step that makes everything after it work.