Maintaining Competitiveness and Spotting New Opportunities

Dr Peter Catt, Chair and Director of Virtual Blue, presented a practical framework for how organisations can respond to uncertainty through applied AI and predictive modelling. Rather than focusing on hype or emerging technologies still in development, his session centred on tools that businesses can use now to reduce risk and improve planning.

He opened with a straightforward point: unpredictability is not new, but most organisations are still using planning methods that assume stability. “The issue isn’t turbulence itself,” he said. “The issue is how we respond to it.” He challenged the audience to examine how many of their forecasting tools rely on static spreadsheets or fixed assumptions, which often fall short in dynamic environments.

Drawing from real New Zealand case studies, Catt showed how AI-powered forecasting has helped companies improve stock alignment, free up working capital, and respond to supply chain pressures. These examples were not positioned as breakthroughs, but as evidence that practical use of data can lead to better day-to-day decisions.

He also clarified that AI is a broad field, and generative AI is only one part of it. “There’s a lot of AI out there that’s very robust—it doesn’t hallucinate,” he noted, encouraging the audience to look beyond chatbots and toward tools that support explainability and planning.

Catt emphasised that building predictive systems does not require a complete digital overhaul. With the right data and leadership backing, meaningful improvements can be made in short timeframes. “Planning always requires prediction,” he said. “You’re anticipating a future state, then you plan, ideally act.”

His talk offered a measured, evidence-based approach to using AI for decision support, with a focus on integration rather than reinvention.

Takeaways

Prediction requires more than instinct

Businesses traditionally rely on static tools like spreadsheets, which are no match for the sheer volume and complexity of modern variables—such as freight movements, exchange rates, and social sentiment. Dr Catt urged firms to shift from “yesterday’s logic” to AI-enabled forecasting that can handle nonlinear, multi-variable relationships and present probabilities rather than fixed figures.

PAC learning provides confidence

One of the most compelling ideas introduced was “Probably Approximately Correct” learning. “It does sound like a joke,” Dr Catt said, “but it’s actually a serious framework. You might be 90% confident that your prediction is within 5% of the actual outcome.” This gives leaders a statistically valid basis to act, even in uncertain scenarios.

AI is more than GenAI

Catt warned not to conflate generative AI with all of AI. “There’s a lot of AI out there that’s very robust—it doesn’t hallucinate.” Symbolic AI and machine learning have long played critical roles in decision-support systems, especially when explainability and compliance are key. He forecast that “artificial general intelligence” is likely to emerge not from GenAI but from “neuro-symbolic AI”—a blend of neural networks and rules-based logic.

Causal AI and ensembles are the future

Beyond correlation, businesses should explore causal AI models—built on the work of Judea Pearl—that discern whether variables are truly influencing outcomes. Combining multiple models (ensembles) is also critical: “Two weak models averaged together can outperform one strong model,” he said.

Statistics still matter

Despite the buzz around AI, Catt reminded the audience that “statistics are your friend.” Many breakthroughs in machine learning echo findings from decades-old statistical models. Marrying both disciplines can help tease out seasonality, trends, and hidden market signals.

Actionable deployment is key

Catt emphasised embedding predictions directly into workflows, CRM, ERP, customer churn alerts, so that insight becomes action, not just insight. “If you can’t do anything about it, it’s totally academic,” he noted. “But if it plugs into your existing systems, it becomes transformative.”

Your data is strategic capital

His closing call to action: treat your data like a strategic asset. Merge internal data with external indicators to build richer models. And with the right structure and sponsorship, a breakthrough predictive system can be operational “in as little as six weeks.”