Our report Demystifying Artificial Intelligence in Risk and Compliance examined how Artificial Intelligence (AI) is being used in these two key areas, illustrated with a number of case studies. This Spotlight focuses on broader AI applications within the finance industry, and considers the most successful uses of AI tools in this sector, as well as possible future applications. A subsequent report due later in 2018 will examine the technical aspects of AI in more detail.
The way that most people think of AI1 is wrong. For many it is a field of unimaginable complexity, in which scientists create artificial machines to solve problems beyond humanity’s capabilities. In reality, however, AI is neither an ‘overseer’ nor a method for solving complex multi-factor problems. Rather, it is a ‘servant’: a cog in an often much larger analytical machine. And while AI is certainly immensely powerful, the idea of overarching genius-level systems is a pipe dream, albeit one that many technologists and companies are still chasing.
As Chartis defines it, AI has many strengths, but it also has many weaknesses. It does not adapt easily to drastic changes in datasets and, in an environment of continuous data inflow, it must be constantly ‘maintained’ – monitored and updated – to stay efficient. And with every improvement that adjusts an AI tool to its changing environment, the algorithms involved also become more complex. In fact, introducing new, different data may actually lead to a drop in the quality of an algorithm.
Away from the hype, in this report we focus purely on where AI can offer genuinely profitable and effective solutions in finance, delivering the technology behind automation in a variety of applications. As AI is highly efficient in environments with large data sets and a narrow focus, it has had a significant impact on the development of data transformation applications in financial services. While the
most successful financial use cases for AI have so far been in the retail banking space, compelling applications have also become established in the areas of so-called ‘alternative’ data (or ‘alt-data’) and data quality.
- Improving data quality using Machine Learning (ML) tools is by far the most common use of AI in financial services.
- Alongside AI applications designed to improve data quality and operational efficiency, technologies such as voice recognition and so-called ‘chatbots’ are being deployed in a range of successful customer-facing applications in retail and private banking.
- The area of alternative data is one of the fastest growing – and lucrative – for AI (and especially ML).
Elements of success
For FIs to successfully implement AI tools, we believe they must consider five crucial elements when developing an AI process from end to end:
- Data quality. The better the quality of the data being fed into an AI process2, the easier it is to tag the data – a vital step for effective analytics. Poor analytics working with good-quality data will always give a better result than good analytics working with poor-quality data.
- Transparency. As more importance is attached to AI, the whole process must stay transparent. Firms are not willing to risk their money and reputation on systems they don’t understand.
- Clear responsibility. If a serious error does occur, FIs must be able to trace the chain of events and link processes to people, ensuring they resolve issues rapidly.
- Adequate hardware and software. The success of AI relies heavily on the hardware and software it operates with. Parallel processing (using Graphical Processing Units [GPUs], for example) and specific data management systems (such as NoSQL or graph databases) are increasingly being applied to AI use cases.
- AI specialists. As with any technology, AI should be constructed, operated and developed by relevant experts (such as data scientists). These individuals must combine domain knowledge of the area they operate in with technical knowledge of how AI works: often a rare combination of skills.
Many FIs will find that, rather than being best deployed as an overarching decision-making machine, AI will instead be used to add value across many small, predictive processes. The paradox of AI is that, while it relies on vast quantities of data to operate well, it is best applied to small use cases, or the ‘cogs’ in bigger processes (see Figure 1).
While AI will continue to expand the assortment of possible ‘cogs’ that FIs can use, it will not necessarily outperform the cogs they already have in their decision-making processes. As an example, it may take years to prove how effective a credit-scoring mechanism is in the real world (i.e., when debtors start to default), so it could be hard to tell if an ML-driven replacement was actually more effective.
AI often works better as a sorting mechanism that feeds into rules-based or user-based decision- making, rather than as a decision-maker itself. So perhaps we should think of it less in terms of free- standing artificial intelligence and more as a tool providing assisted intelligence.
Longer term, the use of AI will be focused on data analytics, data cleansing, and data transformation and modification. The digital conversion and loading of documents into legal or compliance software and workflows is one area where significant headway is being made. Finally, and notably, because the most effective use of AI is in transforming data from an unstructured to a structured form, there are plenty of development opportunities still to be explored.
1Note that when we talk about ‘AI’ in this report, we mean AI in an industrial/business context, not the AI of science fiction (robots, intelligent machines and so on).
2By ‘AI process’ we mean the process of constructing and maintaining an AI tool.