AI-driven forecasting was one of the first AI projects I ever tried—and it worked surprisingly well. We saved time, improved accuracy, and felt like we were stepping into the future of finance.

However, as I worked with more companies, I began to notice a pattern: it failed more often than it succeeded. Not because the tools were bad, but because most teams weren’t structurally ready. Forecasting turned out to be more art than science in many environments, and AI struggles with that.

So does that mean you shouldn’t bother? Absolutely not.

Even when AI doesn’t deliver an accurate forecast, it tells you something far more important: what’s broken in your current process.

Why AI Forecasting Fails (for Now)

Forecasting is Still an Art

Despite what software vendors promise, forecasting is often an art as much as it is a science, especially in the hands of a seasoned CFO.

Take cash flow forecasting. In many companies, key details live in your head:

• Verbal agreements with key clients
• Expected payment behaviors based on relationship history
• One-off events like supplier issues or last-minute tax payments

These things don’t live in ERP systems. And AI can’t use data it doesn’t have.

There Are No Real Patterns

AI thrives on historical data and repeatable patterns. But if your business is:

• Early-stage or still building a revenue base
• Pivoting its pricing, product, or customer segments
• Dealing with irregular sales cycles or high churn

… then patterns are either weak, inconsistent, or irrelevant.

The result? AI models that look confident but aren’t remotely accurate.

The Business Model is Still Evolving

If your company is shifting from services to SaaS, expanding into new markets, or testing new monetization models, your past data becomes less useful.

In those cases, human judgment matters more than machine learning. AI can’t forecast the future if the future doesn’t resemble the past.

When AI Forecasting Does Work

Despite the challenges, there are use cases where AI-driven forecasting shines.

1. Businesses with Clear Patterns

SaaS companies, subscription-based services, and businesses with predictable renewal cycles see the best results. Churn rates, expansion revenue, usage trends—these are AI’s sweet spots.

2. Well-Structured, Clean Data

Companies that treat their ERP or CRM as a source of truth—with clean, timely, and complete data—can feed AI tools the inputs they need to generate useful models.

3. Used for Direction, Not Precision

AI forecasting works best when you’re not expecting a crystal ball. It can help:

• Detect trends you might miss
• Create scenario ranges
• Draft first-pass projections for team discussion

When CFOs use AI as a second opinion, not the final say, the results are far more valuable.

So Why Try It Anyway?

Because even when AI forecasting doesn’t deliver accurate results—it delivers insight into your process.

Trying It Shows You What’s Broken

Run a forecasting model with AI and you’ll quickly discover:

• Data gaps in your CRM or accounting system
• Undefined business drivers (e.g., "What actually causes upsell?")
• Inconsistencies in how assumptions are applied

These aren’t AI problems. They’re forecasting maturity problems. And knowing where they are is the first step to fixing them.

Forecasting Standards Are About to Change

My prediction? Shorter, rolling forecasting cycles will become the norm, even at large enterprises.

Annual forecasts are already being replaced by:

• Monthly or bi-weekly rolling forecasts
• Driver-based models updated in near real time
• Scenario-based planning tools for board and investor reporting

Companies that can’t forecast quickly will fall behind. Trying AI now—even if it doesn’t work—prepares your team for the shift.

What to Do if You’re Not Ready Yet

If your company is too new or undergoing a major change, AI forecasting may not be the right choice yet.

But here’s what you should be doing:

• Keep your data clean. Even if you’re forecasting manually, structure your data like it will eventually be used by a machine.
• Document your assumptions. Move insights out of your head and into a system—so they’re accessible when AI is ready.
• Track manual overrides. If you regularly adjust forecasts by 10% "because you just know," start noting why. That’s how you’ll eventually train your models.

AI can’t forecast your business until you’ve built a forecast that your business can rely on.

AI-driven forecasting isn’t a silver bullet. For most finance teams, it’s a destination, not a starting point. But the journey is worth beginning now.

Because the real value of experimenting with AI forecasting isn’t always the forecast itself—it’s everything you learn about your data, your team, and your process along the way.