FX Forecasting: Why Accounting Teams Remain Reactive

22/09/2026

"How can we avoid FX losses during settlements?"

Don't ask me how to "accurately predict" tomorrow's exchange rate. No one can. Ask how to minimize damage when it fluctuates. That is the most practical question CFOs face when dealing with cash flow pressure.

The biggest mistake is relying on intuition and history

Many businesses still operate by the rule: buy when the USD rises, sell when it falls. It sounds logical, but financial markets are non-linear. You are comparing past data with a present that changes hourly.

What are the real consequences? A manufacturing company in Thailand I once advised lost about one-third of its quarterly profit simply by holding foreign currency too long based on manual forecasts. They assumed the rate would stabilize, but an unexpected macroeconomic news item reversed the trend. Manual processes are not fast enough to react.

The fix: Eliminate subjectivity. Use AI models to scan macroeconomic data, interest rates, and market sentiment in real time. AI doesn't tell fortunes; it calculates risk probabilities. If the probability of fluctuation exceeds a threshold, the system triggers an alert so you can hedge early.

Ignoring the opportunity cost of working capital

This is a more subtle error. You focus too much on the exchange rate and forget that foreign currency sitting idle in an account is dead capital. Forex management isn't just about avoiding losses; it's about optimizing overall financial health.

Example: A major export company in Vietnam held USD waiting for a better rate. Meanwhile, domestic borrowing costs were rising. They could have used that foreign currency as collateral or converted it more flexibly. Without scenario planning tools, you won't see this hidden damage. It's small each day, but compounded over a year, the number is enough to erode your entire gross margin.

We implemented a similar data analytics system with Masan for their supply chain, where optimizing cross-border cash flow is critical. The lesson learned: Tools must link exchange rate data with capital costs; they cannot be siloed.

Alert thresholds are set incorrectly

Many finance teams set alerts for 1% exchange rate fluctuations. It sounds sensitive, but in reality, this level of movement happens daily. The result? Alert fatigue. When a truly dangerous 3% fluctuation appears, they are used to the noise and ignore it.

Instead, use machine learning algorithms to determine "anomaly" based on relative fluctuations compared to recent history and the average volatility of that currency pair. Only alert when the market deviates from its normal trajectory. This reduces noise and increases signal reliability. I have seen a clear difference when applying this method for clients in the lubricant industry, where margins are thin and sensitive to every small fluctuation.

Not integrating with existing ERP systems

A classic mistake: Buying a standalone FX forecasting software that runs in parallel with SAP or Oracle. Data must be entered manually. Human error is inevitable. Decision-making is delayed by hours, which is enough to miss opportunities.

The solution isn't replacing the ERP, but connecting via direct API. The AI system must read foreign currency balances, payment due dates, and actual orders from the ERP to provide specific recommendations: "Buy 500k USD now to hedge risk for the November shipment." Without this integration, forecasts are just numbers on paper, useless for business decisions.

The trade-off between accuracy and reaction speed

Don't believe promises of "99% accuracy" in financial forecasting. That is deceptive marketing. In forex management, speed is more important than absolute accuracy. A forecast with 85% accuracy that updates every 5 minutes is more useful than a complex model that runs slowly every hour.

We must accept uncertainty. The goal is to optimize the timing of action based on the best available probability at that moment. AIVISION has worked with partners in the Philippines and Mexico, where markets are highly volatile and fast. The commonality is that they prioritize lightweight, cloud-based models with instant response over heavy models that are slightly more accurate but slow. This trade-off is intentional and needs to be clearly understood from the start.

What to do this week?

Don't rush to buy software. Do a quick test: Take the exchange rate data from the past 3 months and compare it with the actual times your company bought/sold foreign currency. Calculate the total loss if you had only bought/sold at the monthly average rate. This number is the "price of passivity." If that number is large enough to change your investment decision, start looking for an AI solution. If not, your problem may lie in approval processes, not the tool.

If you are weighing up a similar project, our team can help you scope it before you spend anything. See what we build or book a conversation.