Data-driven technologies underpinned by rapidly evolving AI, are set to be placed at the heart of firms’ operating models. It emphasises the need for Financial Services Executives to have a clear understanding of AI, and how they can be embedded in their company processes. As an executive or board member responsible for AI investment decisions, you should be knowledgeable about AI tools and technologies. It’s important to know how to leverage them for maximum efficiency and productivity.
In our recent webinar for Board and Senior Executives: ‘AI: Why an executive understanding is so important,’ world-leading NLP scientist, Dr Lexi Birch and Aveni COO, Jamie Hunter, covered the practical applications of AI in financial services, its challenges, data ethics, human impact and what to expect at each stage of the AI maturity curve.
Watch the full webinar replay here.
Here are the 5 key takeaways from the session:
1. Defining AI: it’s the mechanising of human thought allowing computer systems to perform tasks that usually require human intelligence such as speech recognition.
The question AI seeks to provide an answer to is, ‘how can we mechanise human thoughts?’ Artificial intelligence has undergone a lot of stages since the 1950s, from the development of information theory and early computational devices that were the foundation of artificial intelligence in the last century, to its current stage of immense advancement. Currently, there is a paradigm shift towards neural networks and deep learning which are algorithms that train the model to process data the same way the human brain would. These transformational technologies are modelled on the human brain and nervous system.
A standard way for a machine to ‘learn’ is to adopt supervised learning which means training the machine using labelled data. A simple example would be providing a neural network with the words, ‘I have cancer.’ This signal is propagated throughout the network and the model is able to predict if having cancer means the person is vulnerable. If the model gets it wrong, then the weight gets updated by a human and the model improves.
Instead of relying on large amounts of labelled data which is resource intensive, with unsupervised learning, the network can learn from unlabeled data. This has been key to propelling us to a new level of performance and capabilities. It also means that practical applications of AI in industry are now more realistic and don’t require the previously large amounts of human and technical resources to deploy.
From Aveni’s point of view, we depend on speech recognition systems to unlock audio data. We’re able to train useful models with much less data. We don’t need thousands of instances of customers admitting they are vulnerable for the model to learn from it. Instead, we can cluster sentences that have similar properties and with fewer instances of labelled data, create a class label also known as classifier from that.
This has had a massive impact on us as a RegTech company whose focus includes the identification of vulnerability, agent performance and coaching, expression of dissatisfaction, and the automation of quality assurance process. We’re able to use very few examples of these issues to build a model and provide accurate solutions.
2. Natural Language Processing (NLP) lets us unlock the value from unstructured data and this is key to the future of financial services.
Financial institutions have got to grips with leveraging structured data such as customer data and credit data. However, unstructured data such as audio from customer interactions, videos, emails, PDF, etc remain both under utilised and under analysed.
Customer communication is key to detecting when there are problems, for example, with offerings and processes within the organisation. Unstructured data from interactions provides insight into customer preferences, and change in customer needs. This insight allows firms to stay ahead of future trends and competition. It equips them to deliver good outcomes to customers. As a result, firms can focus efforts on specific areas, ensuring continuous improvement of the customer experience.
NLP also allows firms to better comply with regulatory requirements. One current high profile example is Consumer Duty. As regulation becomes more data-driven, this technology lets you identify risks and get a true understanding of your customers and their needs. The insight can help you provide a good customer outcome and comply with Duty requirements.
RegTech companies like Aveni, utilises NLP to help businesses optimise low level functions enabling efficiency savings. It can also allow firms to better understand trends and drive insights.
3. Inadequate/inaccurate training data, bias and inadequate resources are some of the challenges in deploying NLP solutions
Though there are a lot of opportunities, there are also some challenges in deploying AI. AI projects can typically be long, complex and costly. Their return on investment needs to be very clearly specified. Key performance indicators must be well defined and compelling enough to warrant these investments. A focus on strategy will limit the possibilities of wasting time and money on pet projects.
Another challenge is inadequate/inaccurate training data. Although models can be trained with less data, data is still a key factor in all successful AI projects. In some cases, companies struggle to provide the quantity and quality of data needed, especially for conversational data.
Data used to train models can also result in bias. A model is biassed when it provides accurate results for one set of data and inaccurate results for other sets of data. This often happens when the training data isn’t of good quality. For example, transcriptions of women’s voices are less accurate when studies have shown women’s voices are generally clearer. This indicates the NLP models have been trained more on men’s voices than women’s.
Companies may lack the resources to properly implement AI. There are cases where firms lack the technical talent and tools to implement and scale these smart systems to provide proper return on investment. This emphasises the importance of low-resource NLP.
4. Adopting human+ AI can help to mitigate risk and make AI systems more beneficial
To deliver safe and beneficial AI systems, we should be putting humans at the centre of the process. Firms should leverage artificial intelligence where appropriate, but under human supervision. The most effective AI systems should try to harness the best aspects of both human and artificial intelligence. We call this human+.
Artificial intelligence is not typically suited for situations where things can change rapidly or problems can not be clearly defined. Leveraging both human and AI is a way of overcoming this challenge. For example, creating models that clearly explain to humans why they are making the predictions that they are making. It should also give some indication of how much certainty they have in these predictions.
A key feature of human+ is the idea of human-in-the loop. In this concept, while the models are being used, humans tune them by giving feedback on the predictions. This leads to models that are really robust and reflect what humans think the right answers are for the particular cases.
5. It has become vital for financial services Executives to embrace a culture of practical implementation of technology
As an executive, it’s important to understand the practical aspects of implementing AI solutions, as well as having realistic KPIs and expectations. This can help you drive more value by positioning your business to experiment and leverage the benefits of this technology. Technology is able to drive a culture of innovation, remove blockers, and empower team members. It also encourages collaboration between the company and technology providers. With tightening regulations and growing adoption of technologies like this in the financial services industry, an executive understanding of AI is key. This is the time to explore and participate in AI investments.
Watch the full webinar replay here.
