Fintech Empowerment: Data Science, AI, and Machine Learning
This article looks at data science and AI/ML. These are two of the foundational technologies that empower fintech. They’re foundational because they’re universally applicable and therefore ever-growing. The authors look at data’s role in all fintech transactions. They describe AI and ML as enablers and amplifiers. The financial institutions that adopt emerging fintech technologies (like AI and ML) have a competitive advantage, though there are adoption challenges for even the most adventurous companies. The authors take a detailed look at fintech and marketing and how big data, analytics, marketing, and financial services can be leveraged.
As the world becomes increasingly connected through the Internet of Everything, the emergence of fintech is beginning to disrupt the financial world with transformative changes. The unique consumer behavior and banking habits of millennials, coupled with their pro-technology attitude, facilitate the rapidly advancing disruptive revolution of fintech. Banks and financial institutions alike have been adopting innovative technology to cater to the ubiquitous use of mobile devices. As many fintech companies accelerate their presence via online operations — not restricted or constrained by time and place — they can provide their customers with more convenient financial services experiences at much lower costs. According to FinTech Global, the “global fintech sector raised [US] $41.7 [billion] in the first half of 2018, surpassing 2017’s record total,” and “global fintech investments increased steadily between 2014 and 2017 from $19.9 [billion] to $39.4 [billion] at a [compound annual growth rate] of 18.5%.” To stay competitive, traditional banks are now facing the challenge of offering innovative fintech products and services to their clients. If the world’s banks cannot find a way to compete with fintech startups, which capitalize on new technologies, they stand to lose $1 trillion in profit.
In this article, we discuss the impact of data science, artificial intelligence (AI), and machine learning (ML) on the revolution and evolution of fintech. We explore the competitive advantages for financial institutions that embrace the latest fintech technologies, challenges they face in adopting these technologies, and future opportunities in fintech with AI and ML.
The Emergence of Fintech
Fintech, a concatenation of finance and technology, refers to technologies used and applied in the financial services sector. Fintech applications cover a wide range of areas addressing the needs of consumers, investors, and regulators. Companies utilizing fintech consist of both startups and established financial and technology conglomerates with the commonality of seeking to replace or enhance the appeal and power of their financial products or services.
Increasingly, fintech has begun to disrupt traditional financial services such as mobile payment, money transfers, loans, fundraising, and asset management. Even currencies are not spared from fintech’s onslaught. The use of digital currencies (any currency in a digital rather than physical form), such as Bitcoin, has, in some cases, become an alternative to traditional cash or check payment. One popular form of digital currency is cryptocurrency. Cryptocurrencies are considered reliable because they are based on cryptography.
As the value of a Bitcoin rocketed above $19,000 in 2017, the cryptocurrency caught the attention of investors. Built on advanced technologies such as blockchain and distributed ledger, Bitcoin improves transaction speed and reduces service fees. Embedded in a peer-to-peer network, every account in the system has a complete record of history of all transactions, so double postings can be effectively prevented. After the confirmation of each transaction, blockchain nodes spread the transaction throughout the network, and the new record becomes part of the blockchain. Digital currencies, including cryptocurrencies such as Bitcoin, Litecoin, and Ethereum, could potentially transform the financial industry.
Empowering Fintech with Data Science: Big Data and Analytics
Fintech is driving financial services innovation, with big data and analytics as two key enablers. Big data has been a ubiquitous term in business and IT news for more than a decade. One of the fundamental issues facing fintech today is how to manage big data and, more important, how to capitalize on data science. To see how fintech affects the business process of marketing, see sidebar, “The Evolution of Marketing.”
The Evolution of Marketing
Approximately 70% of big data usage is for marketing applications. To see how fintech intertwines with big data, analytics, marketing, and financial services, let’s consider an example.
Suppose that Nancy carries a credit card issued by ABC Credit Corporation (a fictitious financial company). ABC would like Nancy to use the card more often because it generates revenue for ABC in several ways. ABC receives a flat fee from the vendor for the use of the card for a purchase and additional interest if this purchase becomes part of an unpaid balance. The financial institution also has the option to impose an annual fee for the credit card.
Traditional marketing methods begin with mass mailings, which have a fixed cost for each person solicited by the marketing effort. The amount spent on the marketing effort is wasted unless the recipient increases the use of the credit card; hence, it can be expensive, as many people will have no interest in increasing their use of the card and will simply ignore the marketing mails.
With fintech, ABC can reach out to potential customers using electronic mail by sending an email to every card holder at a significantly lower cost than postal mail. Unfortunately, most computer users will never see the marketing offer among the tidal wave of daily spam. According to a 2014 Cyberoam report, an average of 54 billion spam messages were sent out every day. Spam messages accounted for 39.2% of email traffic worldwide in 2017, down from 59.8% in 2016, but still a significant percentage.
ABC finds the email campaign response rate disappointing. The Direct Marketing Association indicates that the average response rate for an email campaign in 2017 was only 0.12%. As the price of memory storage has decreased such that more complete customer information can be captured, ABC is now able to store more data and learn more about its customers, which allows ABC to switch from mass marketing appeals to directed campaigns, sending marketing emails only to those card holders most likely to take advantage of the offers or promotions. The combination of fintech, big data, and analytics allows companies to use mathematical models to predict the likelihood of a purchase. By targeting only those customers predicted to have a higher probability of using the card, overhead costs are reduced while profits increase. The overuse of these targeted advertising appeals, however, has quickly hardened the population of computer-literate users. For example, ads on Web pages, even those directed toward a particular consumer, have been countered by ad blockers.
Business analysts using big data and data science stepped up the game. The next approach has been to target smaller and very similar groups of people. Prediction models now gather an increasing number of client characteristics, processed by ever-more sophisticated modeling techniques and faster computers. Personalized marketing means that Nancy receives advertising designed specifically around her purchasing interests. Marketing algorithms not only work to predict her next purchase but also to recommend that next purchase. Suppose Nancy uses her card to purchase a pair of running shoes. Within seconds of receiving this information, a marketing algorithm predicts (or recommends) that her next purchase will be a pedometer, as she is classified as a middle-income, gadget-oriented consumer who is athletic or fitness-conscious. The algorithm, using data from past purchases captured by fintech, shows that many customers who purchase a pair of running shoes eventually purchase a pedometer. Moments later, thanks to AI and ML, Nancy receives a message that good pedometers are available at Joe’s Sports Store just around the corner from her current location. She is given information about Joe’s inventory, which happens to include other items also recommended to her. She is given pictures of Joe’s along with a video advertisement. The fintech application then offers her 20% off the purchase of the pedometer today if she uses her ABC card.
Think for a moment about the complexity of this situation. ABC receives millions of credit card transactions a second from stores located worldwide offering every type of product or service. Nancy’s transaction is instantaneously associated with her personal demographics and purchasing history. Meanwhile, ABC’s data science group has utilized AI and ML to mine text reviews, photos, and videos found on social media to identify a cluster of people with buying interests similar to Nancy’s. All the information is drawn together to provide an estimate of the probability Nancy might next buy a pedometer.
Moving forward, ABC is further exploring the use of data science, AI, and ML to enhance its market share and innovate new financial services and products for its customers.
At the heart of big data are significant changes in the cost and capacity of storing information. According to Computerworld, in 1967, the cost of a megabyte of hard drive storage was about $1 million. By 2017, “that same megabyte of capacity on a hard disk drive costs about two cents.”
A large proportion of new data generated every day is directly or indirectly related to fintech and can be utilized in fintech. Data centers and cloud services developed by Amazon, Google, IBM, Microsoft, and Oracle have facilitated the storage of fintech data. Such cloud computing development has hugely lowered operating costs and enables startup fintech companies to focus their efforts and financial resources on specific applications via the Internet. To be useful, data must be analyzed to extract meaningful information that can be applied to transform business solutions. This transformation is facilitated by increasingly powerful graphical processing units and advanced microprocessors enabling the explosive growth of big data analytics and data science.
The Four Vs of Big Data and Fintech
With the exponential growth of big data in volume, velocity, variety, and veracity (the four Vs), it becomes imperative for the financial industry to efficiently recognize interesting patterns from financial data, enabling business executives to make discoveries from, support decisions with, or provide explanations about patterns, trends, clusters, gaps, and anomalies. Let’s look more closely at each of the four Vs:
Volume. The financial industry sits on a huge amount of data that continues to grow every second. Financial institutions have their data generated from various activities ranging from financial transactions to online consumer reviews. Data volume has been growing from terabytes to petabytes and even zettabytes. With the growth in data availability, a sophisticated team of data scientists and analysts in fintech can do more in-depth and insightful analyses with the data.
Velocity. Having more data does not necessarily lead to faster and more efficient decision making unless that data is also processed and analyzed. A key element of successful fintech applications is the capability to capture and process significant volumes of information on a millisecond basis. For example, institutions using fintech for credit score management face the challenge of producing credit ratings for their clients on a real-time basis. In general, stream processing involves a computing platform enabling analysis of high-velocity data so that timely responses are possible. Data integrity can be improved by continuously analyzing and transforming information in memory before it is stored in the database. Fraud identification requires real-time actions and immediate responses, making developments in fintech an ideal aid. The quicker a fraud is detected, the greater the chance that losses can be controlled.
Variety. Powerful insights come from efficiently processing data in both traditional and nontraditional formats. While most data-driven initiatives focus primarily on structured data stored in databases, unstructured data gathered from social networks, weblogs, and/or other sources contains enormous value for financial institutions. For example, textual data collected from social media can be used for sentiment analysis to gauge customers’ attitude toward a financial company’s products and analyze its product offerings. Fintech needs to be able to combine various data types and sources to yield valuable intelligence from big data and analytics for real-time and effective decision making.
Veracity. Understanding the importance of data veracity is the first step in differentiating the signal from the noise when it comes to big data. Along with opportunities, the huge amount of data poses many challenges for data credibility and quality. For example, using input with bad quality and low accuracy is likely to result in unreliable predictions in financial models. IBM estimates that poor data quality cost companies $3.1 trillion in 2016. Ensuring data reliability and quality, and transforming big data into useful information, are the prerequisites for turning mega data into value-adding “smart” data in fintech.
Descriptive, Predictive, and Prescriptive Analytics in Fintech
Obtaining a competitive advantage is what businesses hope to achieve with big data. As discussed above, to be useful, data must be transformed into practical and useful information via business analytics. Analytics strives to provide insight into the business relevancy of data. Analytics is not useful in and of itself unless the analysis yields actionable insight based on the appropriate performance/value measurement, which can be accomplished via a combination of descriptive, predictive, and prescriptive analytics:
Descriptive analytics in a fintech context provides a summary view of an institution’s financial health based on past financial data. It presents historical data in an easily digestible format for the benefit of business and financial executives. Primary analytical methods include data aggregation and data mining. Information visualization is especially important, particularly as the dimension and size of data sets increase. Visualization of financial data (e.g., sales expenses, profit margins, and net income over time) is an example of descriptive analytics. The ease of comprehension and accurate representation of data analytics results enable their efficient and effective utilization for timely decision making.
Predictive analytics focuses on predicting the future with available data. Many financial institutions have widely used time series, regression, and advanced ML methods to predict outcomes of interest. Deep learning, an advanced form of ML, is particularly important for prediction with large-scale or streaming data, which usually includes a very large sample size and/or a very high dimensionality (i.e., a large number of variables or features). Making financial forecasts based on past trends is an example of predictive analytics in the fintech context.
Prescriptive analytics goes beyond predictive analytics and evaluates new ways to conduct business. Optimization, simulation, and decision-modeling approaches are used to assess possible business scenarios and to find ways to achieve more desirable outcomes. Prescriptive analytics is one of the newest and least fully explored areas in fintech. Although business analytics has been vastly developed and many tools and platforms for analysis are readily available and accessible for free (e.g., R and Python), companies are confronted with the challenge of translating analysis into actionable insights that can be applied to improve competitiveness and drive business value creation. The use of AI and ML to automate financial decision making is an emerging area in prescriptive analytics within fintech.
Challenges and Opportunities of Data Science in Fintech
Big data and analytics present tremendous opportunities for fintech to innovate existing products and services. However, many challenges come along with the opportunities.
Data quality is paramount in analytics. “Garbage in, garbage out” is to be avoided at all costs. Data capture and cleansing are essential and are the most time-consuming parts of a fintech analytics project, especially when big data is involved. Raw financial data can be extremely messy, noisy, and dirty, as it may be obtained from different sources, some of which are ill-structured and untrusted. A well-designed system architecture can enhance the accuracy, reliability, and efficiency of the data-capturing and cleansing process. As the fintech industry moves to real-time analytical processing, data quality becomes an ever-more challenging issue. In fintech applications, big data platforms such as Hadoop can be used as a data hub that integrates data from various sources and provides a single data environment for descriptive, predictive, and prescriptive analytics.
Textual and Sentiment Analytics in Fintech
Given that a large percentage of business-relevant data is stored in a textual format, textual and sentiment analytics are essential for fintech data analysis. A nontrivial exploration of textual data (e.g., Twitter feeds, online customer reviews, and news) requires an appropriate way of representing texts as numbers, which are mandatory when using most advanced analytics methods. Text mining applies a set of algorithms to quantify unstructured text as structured data and then uses quantitative methods to analyze the structured data.
“Bag of words” is a popular approach to quantify textual data by treating a document simply as a collection of words. A general text quantification used in the financial industry, it involves the preprocessing of textual data, such as tokenization, part-of-speech parsing, stemming, and stop-words filtering, followed by representing the text by term count, term-by-document matrix, and other techniques. However, as bag of words ignores the meaning, syntax, and context of the text, the results from this approach could be less satisfactory than that of the natural language processing approach. Advanced text analytics methods (e.g., word cloud visualization, text classification, and sentiment analysis) can be conducted to gain insight into the contextual polarity of emotional reaction to certain financial services, products, and/or processes.
Information Visualization in Fintech
Fintech requires creative approaches to the visualization of financial information for descriptive, predictive, and prescriptive analytics. Visualization can expand human working memory and amplify human cognitive capability. Visual communication has been regarded as “a must-have skill for all managers.” Information visualization is still a challenge in fintech, and more research is needed to ensure effective representation of information for fast and accurate interpretation — especially in the case of multidimensional data and data of multiple types and structures.
Understanding Investor Risk Tolerance and Mitigating Risk Exposure
Fintech, far from being a silver bullet, neither resolves all the woes of the financial industry nor is it a panacea for financial investors. Fintech companies still need to understand and consider investor risk tolerance and educate investors on risk exposure. Ignorance can be very costly, making education and training necessary for financial institutions and investors. The evolution of fintech products needs to be closely monitored and controlled, as fintech companies may be compelled, given the fierce competition in the fintech market, to accelerate and roll out the development of products without thorough conceptualization and testing.
Fintech Governance and Regulation
Governance and regulations typically advance more slowly than the pace of technology advancement. Fintech is no different, meaning that governance and regulatory bodies need to catch up with the rapid developments in fintech. Some organizations that have been enhancing or further refining fintech regulations include:
The US Financial Crimes Enforcement Network (FinCEN) — part of the USA Patriot Act that addresses issues such as money laundering.
The US Federal Reserve’s Comprehensive Capital Analysis and Review (CCAR) — a framework for preventing financial institutions from being undercapitalized.
The US Financial Industry Regulatory Authority (FINRA) — provides a set of rules regarding the transparency and validity of transactions.
The US Office of the Comptroller of Currency (OCC) — regulates fair access to banking and enforces the US Bank Security Act.
Given that cybersecurity and data privacy have been headline news lately (e.g., the Marriott data breach that involved about 500 million guests), government entities also need to expeditiously put in place financial regulations to reduce the occurrences of cybercrime. In addition, regtech, which is emerging as an important solution using innovative information technology to manage the complex array of financial compliance demands, should also be a focus of attention.
Fintech Business Models
It is vital that fintech startups, entrepreneurs, and companies find a successful and profitable fintech business model. Simply adding fintech to an existing business model is not going to work. At best, business as usual can only provide suboptimal outcomes. The business models and processes need to be reengineered to capitalize on the innovations enabled by fintech.
Even for startup funding, fintech entrepreneurs can utilize the innovations enabled by advanced technologies. For example, one popular fintech funding model is crowdfunding. It provides entrepreneurs with a platform to readily raise money at a low cost and from all over the world. Kickstarter, founded in 2009, is a popular crowdfunding platform that focuses on helping people achieve their fundraising goals; the platform has helped fund numerous business ideas such as films, music, comics, video games, and food-related projects. Indiegogo is another crowdfunding platform aiming to help people solicit funds for idea development, charities, or startup businesses. Via these platforms, fintech startups and entrepreneurs can capitalize on using websites for their fintech projects, which they can then publicize through social media such as Facebook, Twitter, and similar platforms. Campaigns that successfully reach their funding targets are assessed fees of a small percentage of the funds raised. For unsuccessful projects, the sites allow campaigners to refund contributors, at no charge, any money raised.
Continuous innovations and adaptations are necessary to survive and excel in fintech, and fintech business models and processes need to evolve as technology advances. Fundrise is a good example. Fundrise is an online investment platform that helps democratize real estate investing. With a minimum investment of $500, Fundrise makes investing in commercial real estate accessible to the masses. An eREIT (electronic real estate investment trust), a Fundrise invention, pools the funds of many individual investors and allows investors to purchase a diversified mix of commercial properties without high capital commitment. By the end of 2017, Fundrise had originated $343.8 million in both equity and debt investment across more than $1.9 billion in real estate property.
The gulf between those who have ready access to computers/mobile devices and the Internet and those who do not poses a challenge to fintech adoption. Another challenge is cultural differences, which should be carefully considered when pushing fintech services from one country or region to another. The generation gap is another stumbling block, with the two main factors being trust in and familiarity with technology. Generations Y and Z readily adopt fintech, while some members of Generation X may face challenges. Moreover, a significant group of pre-Generation Xers may not be fintech adopters. For this group, which includes many wealthy individuals, human touch, rather than fintech automation, may still be necessary.
The Future of Fintech: AI and ML
The use of AI and ML in fintech has mushroomed. For example, robo-advisors are a class of financial advisory services that provide, with minimal human intervention, investors with online automated asset management. Based on a client’s risk preference and desired return rate, robo-advisors employ methods such as modern portfolio theory and investment analysis to allocate the client’s money among a variety of financial products (e.g., stocks, bonds, futures, commodities). Online delivery and the ability to eliminate human interaction through the application of AI offer a unique customer experience that keeps cost at a low level.
AI and ML are expected to be the next drivers of fintech. Machine learning aims to give computer systems the capability to find patterns in and derive insights from data. ML techniques, such as support vector machine, ensemble, and deep learning (recently becoming popular with the advancement of computation hardware), can be used for classifying financial data and performing predictive analytics.
For instance, these techniques can be easily adapted to a financial setting to predict whether a financial customer is likely to cancel financial services (aka “customer churn analysis”) based on the customer’s demographic information and past use of services. Such analysis can help financial institutions optimize their resources to prevent customer churn. Another example is fraud detection. With historical data and ML, an analytical solution can be embedded in the operational process and automatically isolate or minimize financial fraud. Using historical transaction data, the algorithm, together with the training process, is expected to automatically differentiate fraudulent activities from legitimate transactions. AI and ML can help fintech companies detect suspicious incidents instantaneously and expedite the time to respond. Risk monitoring, based on real-time big data processing and ML, can capture fraud signals more effectively and efficiently than traditional rules-based systems or manual processes. Unsupervised ML algorithms (in which data is unlabeled), examples of which are association rules and clustering, have also been widely used in applications such as market basket analysis as well as in customer and product segmentation.
Getting more from pattern recognition in fintech will increasingly depend on the use of ML techniques. To utilize these techniques effectively, organizations need to ensure that the data is not only of good quality but is also cleansed and categorized appropriately for data analysis. As with big data and analytics, the processes to examine data and feed it into ML algorithms will be key. At the same time, caution is needed as there is still a lot of hype about AI and ML. Predictions and prescriptions must be validated, and any biases imposed by AI and ML must be offset.
Although AI and ML provide tremendous opportunities to further advance fintech, they are evolving rapidly, and most of the recommendations provided by AI and ML algorithms are not easily understandable or explainable. Research efforts are ongoing to make these “black boxes” transparent or at least translucent.
Combined with data science, AI, and ML, fintech creates a competitive advantage for financial institutions. The revolutions and evolutions in fintech are accelerating and will continue to drive innovations and create new financial services and products. Financial institutions need to embrace fintech and evolve with it — or be eliminated from the marketplace, passing into history.