Therese Walsh: AI is a boardroom issue, not just a technology problem

What evidence do you need to see before you believe an AI investment is genuinely delivering a return?

We are increasingly moving beyond broad organisational uplift in capability and reviewing use cases to looking for evidence at the boardroom of a genuine return to the business. So there is a call from boardrooms to executive teams to prove an uplift in customer experience, quicker more efficient processes, and economic value delivered. For example Air NZ is now developing AI tools to take dozens of pieces of information and recommend optimal decisions in severe weather impacts to allow normal operations to restart more quickly. No single human or team could optimise across the same number of variables.

From a banking perspective, external fraud and financial crime are on the increase and AI is highly effective at identifying patterns, anomalies and emerging threats in large data sets so it is, and will increasingly be, a formidable tool to use against an increasingly difficult societal issue.

Are boards demanding enough accountability from CEOs for the productivity gains they expect from AI?

Probably not. While the right questions around productivity gains and economic value are starting to be asked more consistently, we are only at the beginning. So key focus areas have tended to be more about adoption, exemplars and systems. However, there are two key issues broader than short term productivity gains that demand attention. Firstly, AI is creating real and present security risks on a significant scale and these are currently dominating boardroom conversations. Although cyber risk has been around for a while, we have managed this through controlling as best we can the high risks as they are assessed. Emerging AI models turn this on its head where combinations of low risks can expose a company to significant impacts. How to govern for this is being grappled with, and while models are starting to emerge this has not been solved for yet.

Secondly, boards require their CEOs to articulate the operating model needed to take advantage of AI within the organisation. Enterprise wide transformation is often required to take advantage of the opportunities AI presents. Simply inserting AI into current business structures and processes is unlikely to maximise its benefits.

If AI delivers a major productivity dividend, who should capture it — shareholders, customers or employees?

Ultimately all of the above. Aligning the commercial drivers for shareholders, customers and employees will always be the most sustainable and productive way forward. However, it is early days and things are volatile so value could skew in different directions while it is building.

Does AI ultimately mean fewer people doing more work, or more people doing higher-value work?

Both. AI already is and has the potential to further eliminate administration tasks and more quickly analyse data. It will shift job opportunities. It should allow people to spend more time in roles exercising judgment, solving problems, and serving customers.

However, it isn’t a linear transition, so a board’s responsibility is to manage the transition thoughtfully and transparently. The key question for a board is whether we’re deliberately redesigning work, roles and capability rather than simply layering AI on to today’s organisation.

What worries you more: that New Zealand businesses will adopt AI too slowly, or that they will adopt it without changing how they operate?

Both are risks, but adopting AI without changing the operating model will likely be the more expensive mistake.

New Zealand businesses cannot afford to be slow adopters. Things are moving too fast for that. However, superficial adoption may be even more dangerous, because it creates the illusion of progress while competitors fundamentally reinvent how work is done. Boards need to grapple with this point and understand where the executive teams are driving the future operating model.

New Zealand has had a productivity problem for decades. Can AI genuinely change that equation?

AI adoption does not equal productivity gains. AI adoption alone won’t change our productivity problem, using it alongside enterprise transformation, however, creates a real opportunity. We have a small population, so AI allows us to essentially grow the capability we can’t solve through immigration. It allows us to do more with the population base we have.

What is the biggest AI risk that boards are currently underestimating?

The fastest-growing threat that we are all grappling with is cyber security. Anthropic’s Mythos work is quite sobering. AI changes the economics of cyber attacks. For me it is not about boards underestimating the risk, it is about how to mitigate the risks that is the hardest part. The basic issue is: Can our cyber defence keep pace with the offensive capability of AI? I don’t believe there is a company in New Zealand who can answer that question confidently.

However, perhaps the most underestimated risk is not moving fast enough on enterprise transformation. It is a bit like buying a tractor to plough the fields faster, and then we harness the cattle to the tractor so they can both pull it only to slow things down.

Are New Zealand companies becoming too dependent on a handful of overseas technology companies for critical AI infrastructure, models and data?

Global company concentration risk is an issue that deserves board attention.We already manage dependency on critical suppliers in many parts of our businesses. AI should be no different. We are unlikely to build many foundation models at global scale in New Zealand, so partnership is inevitable.

While partnership is important, dependency without options is something quite different. So the strategic question for boards to consider is what capability, data, IP, and optionality we need to retain ourselves.

Five years from now, what will we look back on and realise we completely misunderstood about AI?

It is hard to know but, if I had to pick something, it’s that there is a big focus on what AI can do not how to change organisations to use it. But I think we will all look back and think we should have moved faster in both public and private sectors to use AI as a way to get ahead.

Looking across aviation and banking, two highly data-intensive industries, where do you see AI having the greatest potential to change the economics of the business?

Both industries have enormous amounts of data, complex operations and thousands of decisions being made every day. The largest prize is probably not one spectacular thing but thousands of decisions being made better and more easily.

 At the moment there are many areas across aviation and banking exploring or using AI to improve processes especially in customer self service and knowledge building with customer facing teams, through to engineering modernisation, and weather disrupt decision making.

What can a bank or airline safely delegate to AI — and what should always remain a human decision?

Overall AI powered decision-making at pre-approved levels is already starting to be implemented in many organisations. But for me — and I would guess for many organisations — material decisions or ones which require a lot of judgment are not being delegated.

For aviation, any decisions affecting safety or airworthiness would have to meet a high bar as efficiency will never outrank safety.

In banking, while AI may increasingly make recommendations, customers must have confidence that decisions made are fair, explainable and subject to appropriate oversight. So human accountability will always be retained.

Would you be comfortable with AI making a decision that materially affects a customer’s financial position or travel experience without human intervention?

Potentially, but it depends on the consequence and the ability to remedy it. Automation is not new and so AI doesn’t change that, but customers need to know that there are safeguards and an effective path to human intervention. Whatever technology or way of working any organisations uses, efficiency cannot come at the expense of fairness or trust.

What have you learned from seeing AI adoption from two different boardrooms — ASB and Air New Zealand — that you would apply to other companies?

There is no one formula, and experimentation is important to get things going.

But the key thing I have seen at both Air New Zealand and ASB is that AI should not be treated as a technology programme, but a business transformation programme, and thinking that way at a governance level is key to unlocking the benefits of AI. Ultimately the question for boards is no longer whether AI will reshape their industry but whether the organisation is prepared to reshape itself.

And finally security, security, and security. There is a whole tsunami of unknown security risk coming at us from AI models externally and we need to be as ready as we can be.

After 15 years in governance, for me this is one of the most significant strategic shifts I have encountered. What makes it different is the implications are well beyond any single company, industry or country.

It is challenging us in governance to rethink how to create and sustain value.