A year ago, convincing a room full of finance and accounting leaders that AI mattered took work. Today, that conversation is over. Everyone's bought in, budgets are moving, and the pressure to deploy is real.

Which is exactly where things start to go sideways.

The finance teams struggling with AI moved too fast, skipped foundational work, and are now trying to reverse-engineer the basics while managing a live implementation. Those seeing real results did something less exciting first: they got the foundation right before they ever touched a tool.

The three places where that foundation tends to break down:

1. Teams trust the wrong data

The close runs on data, and finance leaders tend to overestimate how much of theirs is actually usable. Not all data is created equal. For example: 

• Bank feeds and credit card data are reliable. You might not love what those reports are telling you, but the numbers are externally verifiable.
• Data coming off platforms and sub-ledgers is where things get complicated. When data passes through multiple hands and systems before reaching a journal entry, an AI model will only process bad inputs faster and at a greater scale. It shows up most visibly in accruals, deferred revenue, and revenue recognition,  the areas where data touches the most people before reaching the books.

And it goes beyond accuracy. In finance, getting to the right number is table stakes. The harder requirement is being able to explain how you got there: whether controls were followed, whether the process is repeatable, and whether an auditor,  internal or external,  asking questions six months from now will get a satisfying answer. Bad data corrupts both the output and the audit trail before the process has even begun. This is where accounting-focused AI matters. A tool built for finance treats the control and the audit trail as part of the workflow, not something bolted on afterward. 

In other words, AI works best in accounting when the underlying data is trustworthy. That's the foundation on which the speed, automation, and controls are built. 

2. They automate before they reimagine the process

People, process, technology. It's a framework that gets invoked so often it starts to lose meaning,  until you watch an AI implementation fail because nobody thought seriously about the middle part.

The process problem in the close usually has two layers:

The standardization layer. A pattern that comes up repeatedly in multi-entity or multi-business-unit environments: the same accrual being run differently across a dozen or more entities, with different naming conventions, different owners, and different accounting logic. Nobody realizes how inconsistent it is until they try to automate it. At that point, standardization becomes an emergency, running in parallel with a live rollout rather than preceding it.

The redesign layer. Standardizing your existing process is necessary, but it only gets you so far. A process that works today was designed around the constraints of today's tools: ERP limitations, manual workarounds, and approval chains built before automation was an option. AI changes those constraints significantly.

The more useful question: what would your close look like if you designed it from scratch, knowing what AI can do? That's where the real gains come from, and it's why the scope no longer stops at the close. 

3. They chase the wrong use cases

Not every automation opportunity is worth pursuing. A use case that saves a team 30 minutes a month won’t justify the implementation effort, and it won't generate the internal momentum needed to sustain a broader AI program.

In practice, the highest ROI in the close tends to cluster around three areas.

• Journal entries, including accruals. One of the most time-consuming parts of the close and, for that reason, one of the highest-ROI places to start. More often than CFOs realize, a single person is quietly responsible for posting the majority of entries while the rest of the team carries a fraction of the load. Gaining visibility into that imbalance is valuable in its own right, and it points directly to where automation would have the most impact. Tools like FloQast can surface that concentration risk quickly, before it becomes a bigger problem.
• Reconciliations. Largely manual, largely repetitive, and a consistent source of close delays across organizations of every size 
• Variance analysis. Most teams can tell you what changed month over month. Very few can tell you why, because getting to the why requires time and headspace that a team mid-close simply doesn't have. AI can handle the mechanical work quickly, freeing people to focus on the narrative behind the numbers.

The starting point: get the right people in a room, map use cases against realistic effort and expected impact, and identify the two or three that will produce results fast enough to build confidence before taking on anything more complex.

What's waiting on the other side

When the data is clean, the process is sound, and the use cases are right, the close and reporting processes get faster. More importantly, the people running it get their time back and start spending it on the work that actually moves the business: FP&A, budgeting, forward-looking analysis, exception management, and strategic input into decisions. Finance's role becomes much more valuable when the team isn't buried in time-consuming, manual activities. 

For teams ready to start putting AI tools in place, accounting focused platforms like FloQast help operationalize and automate end to end workflows with AI once that foundation is solid.

This post was developed in partnership with Accordion. Accordion's team works directly with PE-backed CFOs to strengthen finance functions and implement data, technology, and AI across the investment lifecycle — which means they see these implementation failures up close. The patterns below reflect what both teams have observed working with finance leaders navigating AI adoption.