Implementing artificial intelligence in the asset servicing industry

Arnaud Misset – Chief Digital Officer at CACEIS looks at the use of AI in the asset servicing industry covering AI hype and fears to operational efficiency and staff considerations. This exciting new technology might revolutionise the industry, or it could just help increase efficiency – this article seeks to explain the practical realities of leveraging this extraordinary technology.

AI, and especially its close cousin - machine learning, has long been implemented in specific areas of the finance world. It has helped the industry gain operational efficiencies, provide simple chatbots, and sort through vast collections of data to identify trends, report on performance and uncover fraud. However, today, AI is once again right at the centre of the hype machine, with Generative AI and GPT as the new tech buzzwords. Generative artificial intelligence consists of algorithms that can be used to create new content, and a Generative pre-trained transformer (GPT) is a type of model that specialises in generating text. Hype usually leads to over-promising and under-delivering so it is essential to look in depth at the technology’s real-world capabilities and identify practical use-cases. Although the possibilities of generative AI seem endless, our industry, with its reliance on procedural accuracy, speed of operation and huge transaction volumes may well be more suited to traditional or discriminative AI than to generative AI. The key component of both traditional and generative AI is without doubt the datasets it has access to. The now classic “garbage in – garbage out” concept remains entirely valid in the age of AI, where stand-alone off-the-shelf algorithms require quality data in a quantity that allows for accurate training and assessment. Good data governance, boring as that may sound, therefore remains at the heart of the entire headline-grabbing AI revolution.

All companies in our industry will have access to the same standard ‘blank-slate’ AI algorithms promoted by technology companies. This means that the differentiating factor for companies looking to leverage AI tech is less in the algorithm acquired and more related to available datasets, training methods and mastery of prompts or queries. However, there is a huge range of available AI algos, so the selection process is very complex especially as each one needs to be rigorously tested for specific tasks using a large set of quality training data. This could be very time consuming and may lead to decision inertia if the goal is to find one model that is able to perform all required tasks instead of multiple models for multiple tasks. Internal decision-tree procedures may be key to resolving this by routing tasks to the right AI model in a company’s algorithm arsenal. Acquiring off-the-shelf tech is one possibility but there may be significant benefits to developing AI tech in-house, in terms of customising it to specific tasks, maintaining data security, and retaining independence, especially during these unstable times when AI start-ups are popping up everywhere. Time-to-implementation is also an essential consideration which may be very high on many companies’ priority list.

The industry talks a lot about AI in terms of its positive impact on efficiency – essentially performing tasks more rapidly and accurately with fewer resources – with a subtle subtext that hints at fewer ‘Human’ resources. CACEIS’ however is certain that AI will not be a replacement for our expert staff as they represent the foundation of sound AI implementations - customising the models, providing training data, and assessing the accuracy and validity of the output. The experts define the datasets upon which AI’s accuracy is based – and we must remember that AI is never 100% accurate, often suffering from ‘hallucinations’ when training data or poor prompting queries leads to confusion but an answer is nevertheless required. So experts are more or less safe but does this mean that junior staff can be laid off en masse? To do so would be very short-sighted, and short-termism causes problems down the road. Who were the experts before they gained their expertise? Junior staff. With no pool of junior employees climbing the ranks and gaining expertise, who will replace the expert as they leave or retire? Cutting staffing costs to find yourself in a position of having to rely on the expertise of an unchecked and potentially inaccurate AI system is not an advisable business strategy – plus, we mustn’t forget that new technology brings a need for expertise in different disciplines such as data science, machine learning, and querying.

To truly leverage the power of AI at a cross-company level, staff must be well educated on the benefits and possibilities it offers – and essentially, have an opportunity to experiment and perform trials using sample datasets to discover areas of added value as well as limitations. Staff themselves are best placed to know how tasks are performed and how they could be performed and verified more efficiently. CACEIS’ innovation division operates under a ‘Build-Operate-Transfer’ model, in which a solution is co-developed with a specific team, it is then implemented for the team, and once operating correctly, responsibility for the solution is transferred over. So instead of replacing the human element of our company, AI will augment it, staff will refine outputs, and the positive feedback loop will see the accuracy, speed, efficiency and scale of our services continue to improve.