Every finance transformation presentation has the same slide: a tangled map of Excel files labelled "Current State: Manual and Error-Prone." The next slide promises AI-powered automation, real-time visibility, and the end of spreadsheet dependency.
Two years later, the team is still using Excel. They have added AI tools they do not fully trust on top of spreadsheets they do.
The Comfort of Control
Excel persists because it gives finance professionals something AI systems do not which is complete control and full transparency. Every formula is visible. Every calculation is traceable. Every error is fixable immediately. When a CFO asks why revenue is forecast at a particular number, you can walk them through exactly how you arrived there.
AI models cannot offer this. Even tools marketed as "explainable AI" produce feature importance scores and correlation charts, not the audit trail your CFO expects. This is a genuine tension between how finance builds credibility (show your working) and how AI operates (trust the pattern).
The answer is not forcing AI adoption everywhere. It is being deliberate about where AI genuinely adds value and where Excel remains the right tool. Use AI for pattern detection across large datasets, anomaly identification in high transaction volumes, and forecasting across many variables. Use Excel for scenario planning, executive-level transparency, and anything where the calculation logic needs to be clearly traceable.
The Integration Problem
What vendors do not always make clear is that their AI forecasting tool does not automatically connect to your ERP, your CRM, and your custom databases. Someone still needs to extract data, clean it, map it, and load it into the tool.
That someone uses Excel.
So instead of one manual process, you now have two: the legacy spreadsheet process and the data preparation process feeding your AI tool. The workload has not reduced, instead it has increased.
Before implementing any AI tool, map the full data flow from source systems to final output. If data connections cannot be automated, you are replacing one manual process with a more expensive one.
The Governance Problem
Excel sprawl happens when there is no governance. A model is built, modified by several people over time, and the original owner leaves. Nobody knows which version is correct or what the formulas actually represent.
Layering AI on top of ungoverned spreadsheets makes the problem worse, not better. If you cannot govern Excel effectively, you will not govern AI effectively either. Start with the fundamentals such as version control, documentation standards, clear model ownership, and review processes. These disciplines apply whether you are working with XLOOKUP or a neural network.
The Skills Gap
FP&A teams are typically strong at financial modelling, variance analysis, and business partnering. Most do not have a background in statistics, do not know what overfitting means, and could not explain a confusion matrix.
You cannot drop AI tools into that environment and expect adoption. Either invest in meaningful upskilling which requires months of development, not a two-day workshop, or hire people with data science skills who understand finance.
The model that tends to work best in practice is FP&A professionals who can partner effectively with data scientists. Your FP&A lead does not need to build the model. But they need enough understanding to ask the right questions: What data is the model training on? How is seasonality handled? What happens when market conditions shift significantly?
The Reality
Excel is not going away. It is too flexible, too familiar, and too embedded in how finance teams operate. The organisations making AI work in finance are not replacing Excel, they are being thoughtful about what stays in spreadsheets and what moves to more capable tools.
The right question is not "how do we eliminate Excel?" It is where does this specific process genuinely benefit from AI? Where does auditable calculation matter more than predictive accuracy? What is the true cost of ownership including integration, maintenance, and training? Wisdom is knowing which tool fits which task.