Advisers must experiment with AI or risk being left behind
More than two thirds of the work being conducted by financial services professionals today will be automated within the next five years. This according to Khadeeja Bassier, Chief Operating Officer at Ninety One, who made the bold prediction during a presentation to the 2025 PSG Financial Services Annual Conference titled ‘The next frontier in asset management.’
The opportunity in AI-backed automation
The consensus emerging from the presentation to the two-day-long conference was that the AI-backed automation of repeatable processes was an opportunity rather than a threat. Bassier unpacked the present day arrival and uptake of AI in the context of the Guttenberg Press, introduced in the mid-1400s.
“The people of that time had no idea of the transformational impact that the printing press would have on their society, and I would argue that we are sleepwalking into a similar inflection point today,” she said. Why? Because everybody has a smartphone, and access to the internet, and the potential to use ChatGPT or Claude or Gemini.
A memorable analogy was used to describe the widespread adoption of large language model (LLM) AIs: “It is like having an arrogant intern at your fingertips.” The presenter argued that the compute power in these generative AIs will forever change the cost of distributing knowledge, and help advisers to inject more value through the advising process. She warned, however, that early adopters of AI would have to pay attention to how the rules governing society were being redrawn to accommodate technology disruption.
The advent of real-time price discovery
Those with two or three decades’ experience in financial markets will not be unfamiliar with rapid change. In the 1990s, it could take time for bond and stock price information to disseminate, and in some cases, there were notable delays in price discovery for trades conducted in over-the-counter markets. “Today, we have price latency of milliseconds, a construct that fundamentally changed how we traded,” Bassier said. As one example, the shift to near-instantaneous pricing data enabled the rise of algorithmic and high-frequency trading.
The disruption caused by the fast-tracked adoption of AI across financial services shares similarities with the shift to real-time pricing, but a close comparison reveals important differences. First, AI is emerging at a time when society generates and stores an unprecedented volume of data, creating both opportunities and challenges for decision-making. Second, the sheer pace of technological adoption and change in the current era is unlike anything witnessed during the transition to real-time market pricing.
Navigating a deluge of unstructured data
The first point speaks to the exponential increase in data creation and storage, enabled by a range of pre-AI technologies. As Bassier illustrated with her personal example, the shift from a handful of holiday photos in past decades to thousands of digital images today reflects the broader explosion of data across all aspects of life and business. This data deluge is often unstructured, creating vast opportunities for organisations that can apply AI tools to extract meaningful insights. “If you can use compute to unlock meaningful insights from data, you have a game changer; generative AI lands at a time where we are a data rich society,” she explained.
The second point centres on speed. As Bassier noted, “the pace of change is frankly staggering.” She referenced a popular meme about the overwhelming experience of trying to keep up, joking that the past three years have aged her by 15 years. This sentiment captures the intensity of AI-driven change: tools like ChatGPT, Claude, and Gemini have been embraced by millions of users within months of launch, reshaping workflows and expectations in record time. You adopt, or risk getting left behind.
Further unpacking the ‘pace of change’ angle, the presenter noted that the cost of deploying LLMs at scale has plummeted. In November 2021, it cost around USD60 for a million LLM tokens; today, those tokens would cost just USD1.50. Bassier illustrated this with a comparison of the potential costs for producing a 2000-word market wrap: while a research assistant plus resources might cost the firm a couple of thousand rand, the compute cost for the ChatGPT LLM to generate the same output would be around 10 South African cents. A senior person would have to check and verify either piece.
An otter using Wi-Fi on a plane
Generative AI is making storytelling in real time possible, and its capabilities are improving at breakneck speed. To illustrate, the presenter shared the output from an image generate in April 2022 versus April 2025, given the simple instruction: draw an otter on a plane using Wi-Fi.
In its earlier attempts, the AI made noticeable errors such as leaving out bits of the plane or drawing the otter inaccurately. Today, the image is spot on. A photo-realistic otter is seated, working on a laptop, while fellow travellers look on in amazement. Change Wi-Fi to Bluetooth, and the AI seamlessly replaces the laptop with a smartphone and air pods.
Today’s financial services professionals can use their smartphones to access cloud-based LLMs that process hundreds of billions of parameters to make predictions, yet there is no instruction manual to teach them how to ‘drive’ them. “What AI does is shift the paradigm, [requiring us to] learn by experimentation,” Bassier said. Users have to figure out how to prompt each model for optimum outputs, all while keeping in mind that the model will respond differently each time it is prompted. Again, this reality is best presented as an example.
The case study shared during the presentation involved 800 knowledge workers who were divided into two groups. Group A did not use AI, and Group B did. These groups were then given two problems: the first involved a creative product innovation assignment, the second involved solving a business problem. Group B outperformed by 40% on the creative product innovation, but did 23% worse on business optimisation. Why? According to the researcher behind the study, Group B relied on the machine too much. The solution is to view AI as a cybernetic teammate, and be vigilant of the AI hallucinations and other issues that might afflict it.
You will not be replaced…
Bassier identified three barriers to AI adoption: intellectual arrogance, fear of displacement, and fear of the unknown. “You will not be replaced by AI, but you will be replaced by the human using AI,” she warned. Ninety One has embraced experimentation, concluding a contract with OpenAI for Enterprise and exploring tools like Microsoft Copilot to support internal processes. “This is Einstein in your basement; the only limit is your imagination,” she added.
AI’s capacity to simplify complexity has transformed how research and client solutions are developed. Bassier closed with a personal story of her three-year-old son choosing ChatGPT over Google or books to answer a difficult question. When asked why he chose Chat GPT, he responded: “Because ChatGPT knows how to explain it, so that a three year old will understand.” Her closing advice to attendees was simple: “Spend 30 minutes a day just bouncing ideas off your preferred LLM as you would a colleague; you will be surprised on the upside.”
Writer’s thoughts:
The rapid rise of generative AI feels both exciting and unsettling for financial advisers. Are you experimenting with the technology, or still sitting on the sidelines? And what risks do you see in using AI as your assistant? Please comment below, interact with us on X at @fanews_online or email us your thoughts editor@fanews.co.za.
Comments