By Dr Bernd Schopp is Chief Commercial Officer at Squirro

Artificial Intelligence (AI) is no longer the new kid on the block regarding the hottest enterprise technologies. It’s been around for many years and has been deployed for numerous use cases in many different sectors, including Financial Services (FS).

AI can be transformative for banks. It helps locate and manage data, providing actionable insights that allow banks to manage and mitigate risks more effectively, better understand customer requirements, improve the sales pipeline and much more.

But AI is evolving. While it remains effective, certain limitations have emerged. Especially when it comes to Machine Learning (ML). These limitations can be addressed via Composite AI, a new approach that fuses different AI disciplines and techniques to generate better results and deeper impact for users. In a world with more data than ever, Composite AI is becoming increasingly essential to success in banking.

The rise of data and actionable insights

The enterprise world is overflowing with data. It comes in so many forms and types – much of it unstructured – that working with this data, and extracting tangible and actionable insight from it, is a daunting prospect. Around 80% of enterprise data is unstructured, meaning vast insights go untapped.

Insight is everything in FS. Without insight into the potential risks and threats facing a bank, insight into service levels or customer requirements, any bank will struggle to maintain its position. Decision-making relies on a deep understanding of a topic, so banks must do everything they can to find and use such insight.

AI techniques have been a key technology in helping banks locate and manage unstructured data, such as email correspondence, call transcripts, earnings announcements, news reports, and premium data sources. But with ever-growing volumes and types of data, combined with an almost insatiable requirement for insight from that data, is the current state of AI sufficient?

The limitations of AI

Typically, banks use AI to gather information from various sources. They utilise ML and Natural Language Processing (NLP) techniques to find the signals and then present those findings to the user. And ideally these tools are integrated into existing systems and workflows. Most sales teams work with Salesforce, for example.

ML in particular, has its limitations because it is purely statistical. It may cover 90% of the information, but sometimes the right answer to a user’s query is found within the 10% that is not covered. Furthermore, when ML provides you with a recommendation, it does not explain why – this context can be essential. Finally, it cannot model explicit domain knowledge into the search bar or insight engine. To solve this, modern AI combines statistical AI with symbolic AI in a new way.

For an asset manager approaching sales, many questions need answering: what products should they recommend to clients; where is that client in buying cycle; do they have a strong understanding of how products relate and compare with existing products; what can they recommend?

ML creates and identifies signals but cannot take into account the representation of knowledge and reasoning behind it. The full picture – expanded queries and expanded results – gives a greater understanding. This requires a different approach.

What is Composite AI?

In simple terms, a Composite AI platform generates insights from any content and data by fusing different AI technologies such as machine learning and semantic AI with graph technologies. Gartner first coined it in 2020, stating that Composite AI refers to the ‘combination of different AI techniques to achieve the best result.’

Typically, it adds to techniques such as ML and NLP with knowledge graphs to deliver faster and more accurate results, and a reduced time to value. Adding knowledge graphs means a user can explicitly model the domain knowledge in that field, significantly advancing what is achievable without Composite AI.

Reaping the benefits of Composite AI

Knowledge graphs are much more symbolic, explicitly modelling domain knowledge and, when combined with the statistical approach of ML, create a compelling proposition. People are more realistic now about what ML can (and cannot) achieve – with Composite AI, there are no such barriers.

For example, we work with European Central Bank on monitoring risks. We can model their supervisory knowledge explicitly, then we can expand the queries and subsequently expand the results, giving them much greater awareness of the potential risks to the organisation.

A further benefit to Composite AI is that you can use pre-built industry-standard knowledge graphs. Users can then adapt those and develop your models for a specific industry. They need improving, but they are already applicable to most companies.

ML remains very powerful as a technology. But it needs a lot of time and effort to make it accurate to train the models, and it must be integrated with existing tools and systems. Adding knowledge graphs to ML – the Composite AI approach – is a much more effective way of extracting actionable insight from data.

Composite AI expands the quality and scope of AI applications and, as a result, is more accurate, faster and delivers better results to the user.