Forecasting in the age of machine learning: are we there yet?

Forecasting has long been the heart of supply chain planning—and also its Achilles heel. For decades, businesses have relied on time-series methods, statistical models, and hard-earned human intuition to try and predict the unpredictable. But lately, the buzz around machine learning (ML) in forecasting has gone from quiet murmur to loud expectation.

So, is forecasting finally moving to machine learning? Or are we still admiring the idea from afar?

The Short Answer: Yes, But…

Machine learning is reshaping forecasting—just not always in the ways people expect. While the headlines promise AI-powered black boxes that can instantly divine the future, the reality is a little more grounded. Many companies in New Zealand and Australia are starting to explore ML-enhanced forecasting, but it's not about replacing traditional methods outright. It’s about augmenting them.

At Supply Chain Company, we’ve seen how ML can improve results when it's applied thoughtfully, especially for businesses dealing with high volatility, large product portfolios, or data-rich environments. It can uncover hidden patterns, detect anomalies, and adjust for non-linear relationships that classical models miss.

But we’ve also seen how “going ML” without understanding the underlying problem can lead to frustration. Not all datasets are ready for ML. Not all businesses need it. Sometimes, a well-tuned statistical forecast—guided by expert input—is still the better answer.

When Does ML Make Sense?

ML thrives in complexity. For instance:

• Retailers managing thousands of SKUs can use ML models to cluster products and detect seasonality or promotional effects more effectively.
• Food manufacturers can incorporate external signals like weather, social media trends, or commodity prices to improve demand signals.
• Distributors can better forecast slow-moving or “lumpy” items with classification-driven ML models.

One NZ-based food exporter we’ve worked with recently piloted ML-based forecasting to better capture demand shifts caused by changing customer behaviour in overseas markets. The results were promising—not revolutionary, but enough to demonstrate value.

That’s the key: incremental wins. In most real-world supply chains, ML doesn’t provide a silver bullet. It provides a smarter toolkit—better predictions in certain cases, and better explanations in others.

So What’s Holding It Back?

Adoption isn’t just about algorithms. It’s about culture, data, and trust.

Many planners are still wary of ML because it can feel like a black box. “Why did it predict that?” is a fair question—and too often unanswered. The rise of interpretable ML models (like XGBoost with SHAP explanations) is helping, but we’re not quite at the point where everyone is comfortable handing the wheel to the machine.

There’s also the question of data. Clean, rich historical data is critical—and let’s face it, many companies are still wrestling with fragmented systems and inconsistent master data. Garbage in, garbage out still applies, perhaps more than ever.

Lastly, integration matters. Forecasting doesn’t live in a vacuum. If ML-based forecasts aren’t feeding smoothly into supply, finance, and S&OP processes, then their value is limited—even if they’re more accurate.

What’s Next?

We’re at a fascinating crossroads. The technology is here. The results are emerging. The barriers are increasingly cultural and operational.

In New Zealand, we’re seeing forward-looking businesses—particularly in FMCG, food export, and retail—begin to pilot ML tools alongside traditional systems. The goal isn’t to throw out the old models. It’s to combine them with machine learning in practical, explainable, and measurable ways.

We believe the future of forecasting isn’t about replacing humans with machines. It’s about empowering planners with better insights, automating the repetitive, and allowing more time for strategic decision-making.

Final Thought

So yes, forecasting is moving to machine learning—but not with a bang. More like a steady, thoughtful march. For companies that embrace this shift wisely, the rewards will go beyond accuracy—they’ll gain adaptability, speed, and a stronger pulse on what’s coming next.

The question isn’t whether ML will play a role in your forecasting future. It’s: Are you ready to lead that future—or follow it?

Deep Dive?

Interested in a deeper dive into the transformative potential of machine learning across the entire supply chain? The recent Harvard Business Review article “How Machine Learning Will Transform Supply Chain Management?” (March–April 2024) provides a comprehensive overview. It highlights how ML is enabling more agile, resilient, and efficient supply chains by directly integrating data-driven decision-making into inventory, logistics, and risk management processes—complementing the incremental, practical advances described in this article.

Siamak Moradi and Eddy de Haas are Senior Supply Chain Consultants at Supply Chain Company. They can be contacted at siamak.moradi@supplychain.company and eddy.dehaas@supplychain.company.