AI investment needs a more informed boardroom

Artificial intelligence has moved onto the board agenda faster than many boards have developed the skills to oversee it. The issue for boards is no longer simply whether their organisations should invest in AI, but whether they have sufficient literacy to distinguish a compelling demonstration from a sound investment.

By Dhesen Ramsamy, group chief technology and data officer at Old Mutual

In South Africa, the timing is particularly relevant, as King V applies to financial years beginning on or after 1 January 2026, against a backdrop of rapidly rising investment in and adoption of AI.

Look beyond the demonstration

One of the most important leadership disciplines is distinguishing an impressive AI demonstration from a production-ready capability.

AI can now dramatically accelerate software development. A team can build and test a feature much faster than before. However, moving from a working demonstration to secure, resilient software that serves customers at a regulated financial institution remains a demanding process. The same principle applies to operational automation. Saving three minutes at one point in a process creates little value if the next stage cannot absorb the additional volume.

Leaders therefore need to understand both the economics and the technology. What does each interaction cost? Are we matching the task’s complexity to the model in use? What infrastructure is required? What additional security, assurance and monitoring are needed? What is the opportunity cost of a pilot that remains a pilot for 18 months? The objective is not to deploy the most AI, but to identify where AI delivers the greatest lift in customer outcomes and business performance.

That distinction is particularly important in financial services, where the upside is substantial. Better use of previously underused data can improve underwriting and potentially extend financial products to customers poorly served by traditional risk measures. Natural-language interfaces could lower barriers for people who struggle to navigate conventional banking applications. AI can also help customers enter conversations with advisers better informed and more confident.

But inclusion cannot come at the expense of trust. Poorly governed data can just as easily perpetuate historic biases in credit or underwriting, deepening the digital divide. Trust is the new currency, and leadership must protect it.

Governance now requires literacy

Good governance does not mean every director needs to become an AI engineer, but that they need sufficient literacy to interrogate what is put in front of them. The difference with AI is the speed, reach and proximity of the technology to customers and employees. That becomes even more important as we move towards agentic AI.

An AI system that generates an answer creates one category of risk. An agent that retrieves information from several systems, applies rules, decides and triggers an action creates another. If its data is inconsistent or poorly governed, it does not merely produce an inaccurate report. It can take the wrong action, potentially affecting thousands of customers.

Boards therefore need visibility across the entire value chain, from data provenance and model performance to customer outcomes. The starting point should be the business case, not the technology itself. That means focusing on the problem being solved, the evidence supporting the expected return, whether the underlying data is fit for purpose, and the cost of the technology as it moves from pilot to scale.

Those questions matter because the gap between AI adoption and AI value remains significant. PwC’s 2026 research in Africa found that 82% of organisations are running AI pilots, yet relatively few have scaled AI across the enterprise. Therefore, the organisations that succeed will not necessarily be those deploying the most AI, but those that can identify where it delivers measurable value and govern it effectively.

AI is moving from tools that retrieve, analyse, and generate information towards systems capable of taking actions across multiple systems. This shift changes both the opportunity and the level of oversight required.

Three things’ leaders can do now

First, build AI literacy into board and executive development. Directors do not need to know how to build models, but they should understand data provenance, model limitations, AI economics, privacy, security, and the difference between generative and agentic systems.

Second, demand evidence, not demonstrations. Every significant AI investment should link the use case to a measurable business or customer outcome. Ask what changes from start to finish, what it costs at scale, and how success will be measured.

Third, govern what AI produces as rigorously as what it consumes. As systems become more autonomous, organisations must consider not only the information fed into AI but also the recommendations, derived data, decisions, and actions it produces.

AI literacy is therefore not another specialist skill to delegate to the technology function. It is becoming part of the basic toolkit of modern leadership. The institutions that understand that distinction will not necessarily move fastest; they will be better placed to know when to move fast, when to proceed carefully, and when to say no.

In financial services, that judgement may prove considerably more valuable than the technology itself.

The post AI investment needs a more informed boardroom appeared first on IT-Online.