At a high level, Artificial Intelligence (AI) is a branch of computer science that makes machines imitate intelligent human behavior, simulating (and often exceeding) human performance. AI has finally emerged as the future, after unfulfilled hype that goes back to the 1950s, due to developments such as the availability of an immense amount of data, the open-sourcing of ML algorithm development, and advances in high-density parallel processing infrastructure. In fact, IBM now believes the technology solutions market for AI amounts to a staggering $2 trillion over the next decade. Data is the new oil, and 90% of data in the world right now has been created in the last 2 years alone. The power of data has actually lagged the technical capability to monetize it efficiently and effectively, in a world where the use of data is moving from a competitive advantage to a requirement to compete. As such, enterprises are now shifting development focus from software engineering to data engineering, and a wave of AI M&A has hardly begun as companies in every sector will ultimately need an AI solution.
AI-based cognitive systems offer many defining elements, one of which is machine learning (ML), with deep learning as a subset of ML. That said, there is not yet clear consensus on the definitions of terms under the AI umbrella, with many terms now used as buzzwords, and it is hard for many to decipher what is true AI. For example, machine learning is a type of AI that uses algorithms that are self-teaching as the data changes, building a pattern recognition model with subsequent sets of data. ML gives computers the ability to learn without being programmed, iteratively adapting the algorithm for each new incremental piece of data added. For example, Netflix has continually adapted its machine learning capabilities to personalize recommendations to its subscribers, which has dramatically increased its market capitalization by reducing user churn. The largest pain point for machine learning is having large data sets of training data, which often requires manual tagging until the algorithm begins to learn. While startups are typically equipped to innovate at a faster pace than established large players, it is the large companies that have the advantage of existing large labeled data sets (i.e. Google images), including Facebook, Google, Amazon, and Twitter. Deep learning, on the other hand, is a cutting-edge subset of machine learning and involves using a model of human neural networks to make predictions about new data sets. The most impressive breakthroughs going forward will likely be in unsupervised ML or reinforcement learning, which can self-learn without the manual training or labeled data.
While many fear that AI will mean the inevitable end of jobs, AI will more so augment human workers in many ways rather than replace them. When ATM machines were introduced, for example, there were similar fears on job losses. In the end, bank branches became more efficient and profitable, allowing banks to invest in building more branches and hiring more. Consumers are starting to get acclimated to newer AI technologies in small doses as well, including the likes of Google Home and Amazon Alexa, although much of it is not as visible.
AI in Financial Services
The market for AI in financial services is expected to grow from $1.3 billion in 2017 to $7.4 billion in 2022, at a CAGR of 40.4%, according to Research and Markets. The reason for this is that the financial services industry is facing both numerous opportunities to innovate and challenges ahead that will determine the future landscape of the industry post-disruption. Financial services companies that wish to adapt will also need to consider storing information on a blockchain-based distributed ledger, protecting client information with the latest cybersecurity innovations, automating process work, and providing a full set of solutions via mobile banking. As such, the financial services sector has adopted AI at a more rapid pace than most sectors and the most optimistic forecasts on future adoption outside of the technology sector itself, per the McKinsey chart below.
Investment into AI outside of the US, particularly in Asia, is moving at a faster pace with strong government support in many instances. For example, the Monetary Authority of Singapore recently announced a $20 million (USD) investment to deploy AI technologies in financial institutions across the country, subsidizing up to 50% of projects, albeit not the largest government initiative. Interestingly, the market for AI in financial services is growing particularly faster in Asia, with many of China’s financial service companies making huge investments in fin tech, AI, and blockchain, with China exceeding the US in fin tech investment recently. With the Chinese government recently outlining a road map to prioritize the development of AI with a detailed plan to “lead the world” in AI by 2030, it surprisingly already has published more academic papers on AI than any other country to date. AI is the new space race.
Anthony Jenkins, former CEO of Barclays, recently predicted that bank branches could become as commonplace as a Blockbuster store if they do not keep up with technology. Challenges persist, however, in implementing AI at financial institutions, including legacy systems that do not communicate, privacy concerns, data silos, a lack of trained staff, the laborious effort of training supervised models, lack of cultural alignment, and potential bias of machine learning. For an AI implementation to succeed, it requires large sets of proprietary data or a virtuous cycle of data creation, talent, and patents. Banks also need to decide whether to use major cloud vendor, internal build, open source tech, or proprietary tech as part of their AI road-map.
Key Use Cases of AI in Financial Services
1.) Asset Management and Robo-Advisors
Robo-advisors provide automated, algorithm-generated financial planning services and portfolio optimization. While those with a certain level of wealth do not mind the fees of having a trusted wealth manager, the rest of the market is vulnerable to AI disruption as robo-advisors can provide the same index fund weighting and portfolio diversification without the overhead or human error potential. There are some limits for the indefinite future as many serious investors will be reluctant to effectively bet against themselves, even though research shows that evidence-based algorithms predict the future more accurately than human forecasters. The market size of robo-advisors could eventually approach the AUM of the entire asset management industry in time. Similarly, AI-enabled personal finance intelligence applications are helping consumers manage their finances, analyze spending, automate tax form filing, and make financial recommendations with a business model not predicated to generating fees from investments.
2.) Data as a Service
AI can glean insights that can increase productivity, maximize human talent, increase revenue, and reduce costs. This technology will be able to be used at precipitously lower marginal costs over time, leveling the playing field so that even small businesses can use it. One could see an “AWS of AI”, essentially AI as a Service, which will be more of a plug and play for those lacking data science teams. Open source algorithms also lower the cost of building AI applications, including Tensor Flow from Google, H2O, DSSTNE from Amazon, and various Python ML libraries, coupled with broad platforms such as IBM Watson, Amazon Machine Learning, Google Cloud Platform, and Microsoft Cognitive Services.
3.) Workflow Automation and Chatbots
One of the key tangible use cases of AI to date has been in replacing some of the workflow of analysts, back office, and even research, particularly as banks globally seek to cut costs due to regulatory burdens and disruptive fin tech companies compressing their own margins. Natural language processing has enabled machine-based interactive customer service, which is a huge opportunity to engage with consumers in a more intelligent way, increasing revenue and customer satisfaction while reducing customer service employee costs. Bank of America unveiled their intelligent virtual assistant named Erica at Money 20/20 last year, which uses cognitive messaging to provide financial guidance to over 45 million customers. Finn.ai offers a white-labeled chat bot that integrates with integrated with existing messaging platforms. Further, JPMorgan has developed a platform called Contract Intelligence (“COiN”) that can review contracts such as credit agreement in seconds, as opposed to the 30 FTE hours of a single agreement typically takes. Banks can also leverage customer data to personalize experience and target them better, by understanding their habits, financial needs, and stage of life to recommend products appropriately There are also use cases in monitoring regulatory compliance to ensure adherence to thousands of regulations in real-time, further automating workflow.
4.) Fraud Detection
Fraud detection technology is certainly not new. PayPal was developed out of advanced fraud protocols by coupling a tech platform with human judgement, which was later adapted by Peter Thiel and other PayPal alumni in founding Palantir. What is new is that machine learning can mimic the associative memory of the human brain to identify likely fraud with an infinitely larger data set. Machine learning will continue to reduce false positives that annoy consumers, and the accuracy in successfully identifying bad actors will increase as the models self-learn with further transaction data, a virtuous cycle. Deep learning will further reduce fraud as it can take in thousands of variables versus a few dozen. Fraud detection machine learning models are fed a high volume of historical transaction data with numerous variables, self-improve its algorithms as anomalies or fraud are flagged, and compare flagged items with account history to assess the likelihood of fraud, looping in human assistance below a certain confidence threshold. There is also potential to combine fraud detection AI with the blockchain to provide a bulletproof unchangeable transaction ledger.
5.) Hedge Funds and Private Equity
Quantitative hedge funds have been among the most advanced in developing algorithms historically. Machine learning will allow include data such as financial statements, natural language processing to monitor news, and external sources such as every web page on the internet, social media discussions, job boards, transactions, and customized feeds. Machine learning seeks to identify investment trends and stock price changes at a faster pace, in some cases updating itself to adapt to changing conditions, reducing the need for analysts, with some new AI-focused funds charging half the fees and without much of a team. In private equity, tools such as CB Insights are gathering wide sets of data on the universe of private companies with predictive machine learning to surface the ones that are most likely to be a fit for the fund, assessing the prospects for high risk-adjusted returns. For example, Deep Knowledge Ventures, a Hong Kong-based VC firm, has developed an AI system that makes investment decisions based on available data and the software itself has a vote on the investment committee.