Your Data Doesn’t Need to Be Perfect. But It Does Need to Understand Your Business.

In nearly every conversation I have with finance leaders about AI, the same questions arise: Are we ready? What if our data isn’t perfect? Or worse, what if it isn’t even inside our financial system?

After almost a decade working in Finance AI and watching these concerns surface in real-world implementations, I've seen the same pattern play out. The data and systems you have today are almost certainly enough to get started, and the winning sequence is simpler: Start with what you have, learn what truly matters, and strengthen governance as you scale.

Where to Start: Follow the Friction

When evaluating where AI can create immediate, measurable impact, do not start with infrastructure. Start with friction.

Look for a high-value decision point where finance is accountable for outcomes but dependent on upstream signals. That intersection is often the strongest AI use case.

The CFO planning process is a clear example. It relies on upstream drivers that shape demand forecasts, pipeline expectations, and overall financial performance. Finance rarely owns those drivers, yet it is judged on their accuracy. In other words, Finance is accountable for outcomes, but dependent on inputs it does not control.

AI and machine learning are the unlock. They make it possible to convert opaque, time-intensive driver debates into transparent, measurable signals without waiting for structural perfection. So instead of pausing for a fully modernized data ecosystem, ask a simpler question: Can you extract a meaningful snapshot of historical data for that key driver from the past few years? In most organizations, the answer is yes.

With modern AI and machine learning capabilities, that is enough to begin. From a structured extract, you can build a finance-led demand forecast or pipeline projection, benchmark it against historical plans, identify structural bias or volatility, and quantify measurable improvements in accuracy. You move from opinion to evidence quickly, capturing early wins that guide the roadmap rather than waiting for one.

This process is not theoretical. One large organization I worked with needed to forecast consumption volumes across business units to support downstream pricing decisions. The data existed, but it lived across their data warehouse and a manually maintained model, creating bottlenecks and concentration risk. Their historical data pipeline was not yet fully automated, but they chose to move forward anyway.

Using historical volume extracts, weather patterns, population trends, and additional regional indicators, the team built a finance-led AI forecast within weeks. The result was a production-ready deployed model achieving 90%+ accuracy, over a 30% reduction in forecast error relative to the legacy process, and roughly a 90% reduction in cycle time.

The historical data pipeline was automated months later, with no material slowdown or rework during implementation.

The same principle extends beyond forecasting. A private equity-backed services organization managing hundreds of locations nationwide faced a different problem: How to drive consistent financial discipline across a portfolio that varied widely in size, cost structure, and maturity. Existing KPIs applied uniformly across the network were generating more noise than insight, making performance management reactive and inefficient.

Using AI-driven clustering, the team grouped locations into dynamic peer cohorts based on actual financial and operational behavior, not arbitrary segments. This enabled contextual benchmarking, early identification of overspend risk before it surfaced in month-end reporting, and location-level margin profiling that directly informed acquisition decisions. No net-new data infrastructure was required up front. The intelligence came from structuring and contextualizing data that already existed in the business.

The Readiness Question That Does Matter

The pushback is understandable. Many leaders believe early, bite-sized AI use cases are wasteful if a fully modern data ecosystem is not yet in place. The common refrain reinforces this caution: “Bad data in equals bad data out.”

But that framing oversimplifies the issue.

AI models do not operate on numbers alone. They operate on context. A machine learning–driven forecast must understand what the numbers represent within your business, not just the values in the cells.

That means recognizing which entity, product family, cost center, and reporting intersection the data represents, and the hierarchies those dimensions belong to and how they relate. It means understanding what “granularity” actually looks like in your organization: how you report, review, and make decisions. This includes your chart of accounts, consolidation structure, legal entities, reporting hierarchies, key planning intersections, and audit trail.

This is the financial intelligence layer; the structured representation of how your business operates financially and legally. And it is this layer, more than a perfectly engineered pipeline, that drives accuracy, efficiency, and insight in real-world deployments.

Refining and stewarding that layer belongs to the people who understand the downstream financial process and its nuances: Finance.

Price-adjusting demand into revenue, mapping deal volume into legal entities, weighing business units against enterprise objectives, these are finance-owned processes for a reason. The feedback loop between AI model output and financial decision-making, and back into the model, is where real success materializes.

Without the financial intelligence layer, and the right operators stewarding it, you are simply giving a model numbers without meaning. The output will reflect that.

What This Looks Like at Scale

The difference between AI that produces interesting experiments and AI that produces decision-grade results almost always comes down to governed context.

At scale, model accuracy is not just a function of the algorithm. It is a function of the features and predictors the model can access, the institutional knowledge extracted from your planners, and the organizational hierarchy that defines how performance is measured. As operators contribute patterns and feedback across the organization, the system compounds in accuracy beyond what any single forecast could achieve.

The same principle applies to insights and analytics.

The level at which you surface an insight matters enormously. Executives don’t need SKU-level variance in every review. Nor do they need the performance details of a single account executive when evaluating enterprise sales. They need insight at the level that maps to how they think about the business.

That level — the level at which executives report, plan, and allocate resources — is defined by the financial intelligence layer.

Without it, you either drown in detail or aggregate away the signal.

This layer becomes even more critical as finance moves toward agentic AI systems that don't just surface insights but take actions autonomously. The organizations building governed intelligence layers now are the ones that will be ready when that shift arrives.

Where Your Team’s Time Should Go

Your team's time is finite, and chasing perfect data at the source is not the highest-return use of it. The organizations moving fastest with AI in finance are concentrating effort in two areas: identifying data that is ready to support experimentation now, and governing data as it enters the financial intelligence layer rather than attempting to clean every upstream system first. The distinction matters. If forecasting occurs at the business unit and product family level, that is where accuracy and auditability must be strongest. Not every row of source data needs to be pristine, only the level at which decisions are actually made.

The organizations that understand this will not just improve processes. They will build a governed intelligence layer that understands their business. That layer, not the data itself, is what separates organizations that experiment with AI from those that operate with it.

Looking to learn more about advanced finance AI? Check out OneStream’s Finance AI Academy.