Exploring potential of large language models for investment firms

Expectations of rapid return on investment generated by new innovations in artificial intelligence can lead to frequent changes of direction, which wealth managers must resist at all costs.

More than two years after the beginning of the explosion of generative artificial intelligence (AI), the enthusiasm has not waned. Large language model (LLM) providers compete every day with announcements and new releases that turn heads and promise a bright tomorrow.

This frenzy is hard for users to follow. Fear of focusing on the froth when a groundswell can sweep you away has never been stronger. But frequency of change actually invites us to develop stability and continuity at the heart of business approaches, to avoid pitfalls that come with these rapidly changing trends.

The observations made in the first months of AI can still be used to navigate this journey. At the fundamental level, these are: the map, to chart your course with a solid decision-making framework to deploy generative AI use cases; the compass to assess your course and manage risks; and planning, to ensure you don't forget the fundamentals, data, team training, plan B.

These instruments are key to supporting businesses, especially asset and wealth managers, in use and deployment of generative AI in a challenging and fast-moving environment.  They will help them to stay on course in sometimes violent winds, and their usefulness remains even as new challenges are added to the course.  While risk of misinterpretation, regulatory uncertainty or the energy consumption of LLM’s was quickly identified, other challenges have become more important.

The geopolitical situation is not immutable and the weighting of certain parameters in the AI deployment equation varies. For European companies, including the interpretation of their dependence on non-European models, solutions and infrastructures in their reading grid is becoming an issue that has never been more topical. The time when some people theorised about companies without people working in factories seems to be over.

Autonomous systems

The stellar deployment of generative AI also causes perverse results, which can happen quickly. An increasing proportion of the content published today is generated by AIs, this content is then scanned and used by AIs who in turn create derivative content, new content creation and ideas.  These risk drowning out genuinely original concepts, leading to an impoverished marketplace of ideas. Agentic AI, which operates autonomous systems and makes decisions independent of human interventions, will only exacerbate the issue.

Finally, the combination of these two problems also shows the extent to which models configured deliberately to produce certain outcomes will be the source of massive disinformation. Building on a biased body of information produces and regenerates into further biased results that can in turn feed AI content generation (deep fakes, erroneous information) which will multiply the harmful consequences of the models deployed.

The final challenge is that the expectations of rapid return on investment generated by new AI innovations and their media echo chamber can lead to sudden and frequent changes of direction. However, if technology evolves very quickly and long-term promises are not fulfilled, the scope for cost-saving is diminished. For many organisations their deployment of AI is not yet mature and until they have implemented a coherent long-term strategy, their expectations for the ROI potential of the technology will be disappointed by reality.

Steering the ship

It is clear that the use of artificial intelligence within many asset and wealth managers does not yet generate massive savings.  For many organisations, their deployment of AI is not yet fully mature and they have not committed to invest regularly. Until they have implemented a coherent long-term AI strategy, their expectations for the ROI potential of the technology will be disappointing, and tangible results will not be achieved.

Scaling AI will not come for free and it will not happen tomorrow.  This is not a reason to abandon ship, On the contrary; clear goals, consistency and perseverance will pay off. The Apollo programme started in 1961 for a first moon landing in 1969, Amazon made its first profit in 2001, seven years after its creation, and surpassed Walmart's sales in Q4 of 2024.

Companies must think and behave like explorers, developing their experience and use or create tools to help them navigate along the way. Those that succeed will have been able to adapt to these new parameters as they go along, while keeping to a clear course, seizing the opportunity to realise the potential of a truly transformative technology for the benefit of all.

Edouard Legrand, chief data officer, BNP Paribas Asset Management