Beyond the Forecast: How to Lead Finance When the Future Can’t Be Predicted

Anyone who has ever prepared a budget knows the ritual. We look at last year, add a few percent, adjust the assumptions, and end up with a picture of the world twelve months from now. A picture that is calm, orderly and, unfortunately, more and more often an illusion.

Every such forecast rests on a few unspoken assumptions. First, that the relationships between revenue, costs, exchange rates, and demand will remain stable. Second, that a single number is enough to guide both strategic and operational decisions. Third, that the past is a reliable guide to what will happen tomorrow. Meanwhile, markets never stop moving.

Take the Polish economy. In the second quarter, GDP grew by 3.9%, driven by investment. At the same time, the manufacturing PMI remains below the 50-point threshold, and new orders have been falling for seventeen consecutive months. Within a single year, ten-year bond yields have swung between 4.88% and 5.92%.

These figures don’t add up to one coherent story. Which of them should go into the budget? And what do we do when, halfway through the year, it turns out we picked the wrong one?

This reality is increasingly described with the term BANI. It stands for four characteristics of the world in which companies operate today:

• Brittle: Systems look stable until they suddenly break. Leverage points are hidden, and shocks spread in an instant.
• Anxious: Contradictory signals lead to decision paralysis, and stakeholders’ worries are very real.
• Non-linear: Cause and effect no longer go hand in hand. A small change can trigger a huge effect, while a large one may produce almost none.
• Incomprehensible: Complexity exceeds our models, and historical data no longer predicts the future.

BANI differs from the earlier VUCA framework in one important respect. VUCA assumed that uncertainty could be managed with better data. BANI says something harder: more data won’t solve the problem, because the system itself has become more difficult to predict and control. And this is exactly the kind of world in which traditional forecasting starts to fail.

If the future can’t be predicted, what is left for finance? Quite a lot, as it turns out. But it requires a change in mindset.

Traditional planning is built on prediction. We forecast what will happen, build a single number, and defend it all year. We adjust only when reality clearly diverges from the plan. The goal is forecast accuracy.

Resilient planning is built on interpretation. We model what could happen, build multiple scenarios, and revisit the plan every quarter. We adjust each time new data arrives. The goal is rapid adaptation.

Same data, same people, different philosophy. You can predict inflation perfectly and get exchange rates completely wrong. You can estimate demand accurately and still be blindsided by the supply chain. That is why forecast accuracy is no longer the competitive advantage. Speed of interpretation is the ability to understand what is happening faster than others do.

Artificial intelligence seems like a natural ally in this shift. But there is a trap waiting here, and it is talked about far too rarely.

Harvard Business Review described an experiment that should interest every finance leader. It involved more than 300 managers and executives taking part in AI executive-education programs between June 2024 and March 2025.

The task was simple: based on recent share prices, predict Nvidia’s stock price one month out. Each participant first made their own prediction. The participants were then split into two groups. The first could revisit their answer with the help of ChatGPT, with no contact with others. The second had no access to AI, but could discuss the prediction with fellow participants.

The result? The group using AI became more confident and, at the same time, less accurate. The group that talked to each other was more cautious and more accurate. Better technology, more data, and more processing power produced worse forecasts, and people felt better about them.

Two mechanisms are behind this:

• Extrapolation bias: The model blindly extended Nvidia’s upward trend into the future.
• Authority bias: The confident tone of the AI’s answers overrode human judgment.

The conclusion is not “don’t use AI.” It is this: AI amplifies the way we think, including our mistakes. A model’s confidence is not the same as accuracy. Sometimes a model that claims to be 95% sure is more dangerous than one that honestly admits uncertainty.

So how do you use AI without falling into the confidence trap? Rather than simply accepting or rejecting a model’s output, it pays to interrogate it. These questions are not a sign of distrust. They are a way to separate insight from noise and to trust AI where it deserves it.

1. Does this forecast assume the past predicts the future? If so, what has changed since the period the data comes from?

2. Where does the model’s confidence come from? Is it based on the strength of the pattern, or simply on the volume of historical data?

3. What assumptions are baked in? Exchange rates, energy costs, demand elasticity: are they still valid?

4. Can you trace the logic? If the output can’t be explained, should it guide material decisions?

5. What if we are wrong about scenario probabilities? Does our strategy still hold up in the downside case?

In finance, the stakes are different than in most fields. A wrong number doesn’t stay inside the building. A bad AI output doesn’t just delay a decision; it costs credibility, audit standing, and sometimes far more.

No wonder that although as many as 89% of finance leaders are exploring AI, most struggle to identify safe, practical use cases. They want to reduce time spent on operational tasks, improve audit readiness, and free their teams for analytical work. What holds them back are concerns about data security, uncertainty about how AI fits into existing controls, and doubts about the accuracy of outputs without human oversight.

That is why the real benchmark for AI in finance is not speed. It is trust. AI that finance can trust has to combine three qualities at once:

• Human-First: A finance analyst can use it independently, with no IT ticket and no consultant. Finance defines the rules itself: variance thresholds, approval paths, exception handling. Without this, adoption fails.
• Genuinely Intelligent: It understands financial context: intercompany flows, consolidation rules, late adjustments, and unmatched transactions. It knows when it doesn’t know and flags uncertainty rather than returning a confident wrong answer. Without this, the tool becomes shelfware.
• Trustworthy: Every output can be traced back to source data, with no black box. Before it reaches board packs or filings, an output passes through human checkpoints. Without this, AI becomes a liability.

Underneath all three lies the foundation: AI-ready data that is connected, clean, consistent, and up to date. Even the most intelligent system will mislead you if it reasons on bad data.

This is exactly the approach behind Lineos, an AI companion that covers the entire finance cycle on one connected platform:

• Plan: Agentic scenario modeling with autonomous forecast optimization and collaborative decision-making.
• Reconcile: AI doesn’t just assist; it autonomously reconciles your balance sheet.
• Close: An intelligent consolidation companion with autonomous exception handling and validation.
• Report: Conversational analytics: a dialogue with your data, autonomous insight generation, and contextual commentary.
• Disclose: An intelligent compliance companion with intelligent narrative generation and regulatory updates.
• Tax: Autonomous optimization and proactive compliance monitoring.

Each module can work on its own, but they deliver the most value together, as a comprehensive view of the company’s entire finances.

We don’t know what the zloty exchange rate will be a year from now, when manufacturing will come out of its slump, or which shock will be the next to upend our assumptions. And we don’t need to know. The role of finance is shifting from an oracle expected to know the future to a guide that helps the company navigate uncertainty.

In that role, AI can be an enormous help, provided it supports human judgment rather than replacing it. The best decisions will not be made in the companies with the most accurate forecasts. They will be made in those that understand what is happening fastest, ask the right questions, and are able to change course early enough.