Blog

While the idea of modern-day Artificial intelligence (AI) has been around for years, rapid innovation in the last couple of years has led to immense growth and potential for AI. Today, chatbots and robots take the lead in AI headlines, highlighting their power to lower costs and the debate on whether they threaten jobs, or help create them. However, the real power of artificial intelligence lies in deeper data science capabilities backed by real business understanding and problem solving.

Financial institutions can collect billions of data points from their technology infrastructure every day, with most of it going untouched. This is where AI can make a significant impact. Data is invaluable, but only if it can be assigned meaning. By employing deep data analytic tools and cognitive machine learning to structured and unstructured data, businesses can have more insight than ever before and turn their data into meaningful assets to apply to problem solving, consumer insights and more.

Arguably, the financial services industry has been employing data science techniques for decades in risk and pricing calculations like Monte Carlo simulations, Black-Scholes models, credit risk scoring, and more. Today’s technology landscape and recent progress in technology and data science innovation allows for financial institutions to leverage data science techniques to delve in to deeper use cases. Deep data science is already at work for the largest technology companies, like Facebook, Google, Amazon, and more. Google has a smaller IT department than HSBC, for example but has 100+ teams that are focused on AI. Financial services is in a position of playing catchup to remain current with the latest innovations, as the space is moving quickly, and waiting too long could have sizeable disadvantages, especially as these advanced technologies are becoming more personalized and more consumerized.

Challenges to AI Adoption in Financial Services

The financial services industry has three unique challenges ahead to apply the innovations being made by the tech giants in this space to its domain-specific problems, including these key three challenges.

1. Security

Banks have greater security, architecture and regulatory oversight which naturally slows things down. They cannot afford to fail and so by that nature will be slower and more conservative.

2. Unmined Legacy Data

Banks have a huge legacy of data, data sources and applications due to their heritage and M&A activity. Combined with often short-sighted focus on individual projects and requirements, and this makes for an incredibly complex data landscape to harness.

3. Legacy Regulation

Financial services companies have heavy regulatory burdens that tech companies currently don’t have impeding them. Financial companies typically have to be able to tell the regulators exactly why they have done something. Many of the latest data science/machine learning techniques such as Deep Learning are inherently ‘black box,’ and so it’s very difficult to provide that full audit trail of why an action has been taken.

However, these challenges are not unsurmountable for the financial services industry to delve into deeper data science and machine learning techniques to apply to business problems and create more cohesive solution sets.

How the Buy-Side and Sell-Side Are Using Data Science and AI

The buy-side is currently leading the efforts within the financial sector. For example, hedge funds are trading on real-time sentiment analysis of social media and news sources. With such large historic data sets of trading available, there is in turn a huge amount of data to be analyzed for patterns and insight, which can assist with predictive pricing and defining trading strategies. Pricing and risk calculations can be and are being made deeper and more complicated now that the compute power and data is readily available. Taking open-source a step further, Quantopian is even now crowd sourcing algorithmic trading strategies.

The banks are also starting to utilize data science techniques for banking, with some key use cases being the following:

• • Personalized tracking and alerting on spending and payments

Machine learning can be applied to recognize scheduled payments and incomes, and comparing them to monthly spending patterns so that banks can alert their clients if they predict they might overspend or that they could be investing more. Taking this a step deeper, banks could compare clients to each other/to client segments to provide feedback on their spending behaviors. For example, if a client spends more on electricity than their peers, but less money on groceries. This provides value add to customers and offers up-sell & partnership opportunities.

• • Matching customer behavior (and that of their customer segment) to real time activity Analyzing news, market movements, etc. can be used to predict and suggest new investments. E.g. A client invests in Apple shares, and many people in their customer segment also invest in technology stocks. If/when a comparable stock beats its earnings forecast and/or the banks research department has upgraded their outlook the bank could proactively suggest the client invest here. This creates up-sell opportunity, but also allows salespeople to work more intelligently and efficiently while providing a more tailored and personal offering to clients.

• • Real time/near time analysis of production systems and infrastructure
    By running analysis in real time, issues can start to be predicted or detected at a very early stage so that they can be proactively addressed. For example, a combination of many minor events could work as a predictor of a larger event coming in the near future. Providing network analytics can also start to show hidden interdependencies and links between applications/servers that weren’t previously obvious.

• • Improved fraud/AML/FinCrime/KYC search and detection
    The application of network analytics can better understand links between people/companies and events. This can go even deeper by adding non-traditional information, such as psychological profiles. Even further, adding unstructured data to this – for example, if a trade looks suspicious statistically, but by adding in communications like chats or emails related to the trade there is a much larger picture to analyze. Voice can be analyzed not just for the conversation being spoken, but also to identify stress levels to add even more insight.

Getting Started

Applying data science to financial services means utilizing the vast amounts of both unstructured and structured data created every day to gain insight, and thus give the data value. Data science employs techniques such as machine learning, predictive analysis, statistics, and others with the goal of using the data to address a business question or problem. In getting started with AI, CTOs should formulate a plan looking at which use cases are safest to start with, how to figure out which algorithms work best for which use cases, and how to source the talent or tools needed for said use case. For the first project, firms should start small, combining tech talent with domain knowledge rather than just relying on developers or engineers to run with the use case. The best results will come from a smaller start that builds up from there, as the firm gains credibility and evolves the technology and builds the infrastructure over time to support more complex use cases. Initially, firms must define what the problem is to be solved, and then define that use case from there. There is also an initial decision of whether to build in-house, or to utilize an external vendor. If choosing a vendor, firms should do a deep assessment of vendors. Firms looking to implement deeper AI techniques should bear the following in mind as they begin their projects:

• An acceptability of failure, to the point of encouraging it, is needed within the firm as data science and machine learning, are by their nature experimental and not all proofs of concept will be successful.
• Start with the ‘low hanging fruit’ to build knowledge and experience while still providing quick value.
• Look at what data they have and how accessible it is to the data scientists involved. Flexible and scalable architectures will be needed to rapidly iterate and test hypotheses while keeping costs low. However, they should not allow a focus on data cleanliness and a central ‘data lake’ to slow the adoption and use of data science.
• Firms need to keep a business focus on their data science activities and make sure that activities are driven by the business, with measurable success criteria at frequent checkpoints. Partnering with experts can be a great way for firms to rapidly get up to speed and to avoid common pitfalls, and at the same time can use this to upskill and build internal capability.

Financial services firms that look towards innovation to deepen their AI capabilities and use cases via deep data science and machine learning applications have significant challenges to overcome, however, there are also significant positive gains they could reap from these innovations. These deep learning and data science applications and analytics allow institutions to provide more tailored, personalized service for clients, which is especially important for difficult or complex clients. Cost reduction and increased efficiency is also a major benefit, stemming from more targeted client acquisition and more efficient use of applications/infrastructure/operations and IT time and resource. Additionally, many of these use cases for these technologies create up-sell opportunities for institutions, and the data analytics will tell a lot about the client, meaning the better a firm knows their client and their specific, unique needs, the easier it will be to sell to that client and not waste time on products that they do not need. Financial services firms can build on the groundwork laid by the Googles and Facebooks of the world to bring unique benefits to their businesses and clients powered by AI and deep learning.

Come and find out more about Synechron at the AI Expo Conference & Exhibition North America in Santa Clara on the 29-30 November. Synechron will be exhibiting and a senior representative, Peter Memon, will be speaking within the AI in the Enterprise track on Day 1 (29 Nov) about how AI powers Business Intelligence.

(c) Synechron