How Ayasdi’s machine learning is giving banks an analytical advantage

As the famous adage goes, “the future is already here — it’s just not very evenly distributed.” This is especially true for banking and the use of machine learning. Banks are increasingly using machine learning to power part of their operations, but the adoption of these new technologies is not uniform. The road to full implementation is fraught with technological and organizational challenges. Machine learning is a collection of techniques for understanding data, including methods for visualization, prediction, classification and other tasks relevant to data analysis. Banks are implementing machine learning in three main stages. The first stage is automating manual processes that are currently handled by an army of quants, such as risk modeling, stress testing or Anti-Money Laundering monitoring. Top data scientists are employed by financial institutions and working with programming tools like SPCC and R, they filter and analyze huge data sets in order to perform analytical tasks. “No human being can wrap his head around that amount of data,” said Daniel Druker, CMO of Ayasdi, a machine learning company that partners with financial institutions, like Citi and Credit Suisse. Instead, using machine learning algorithms, a computer can surface insights and recommendations from those data sets, while the quants examine and take actions based on those learnings. “It is not a person guessing what part of the data set is relevant. With machine learning, 5 thousand man hours go down to an afternoon,” Druker explained. For AML, human beings currently monitor and check suspicious transactions, which is very time consuming. With machine learning, however, a computer can use historical data to filter out the easy cases, leaving the hard cases for humans to analyze. According to McKinsey’s 2015 Global Banking Report, banks that have replaced older statistical-modeling approaches to credit risk with machine learning techniques have experienced up to 20 percent increases in cash collections from outstanding loans. Out of over 20 banks that work with Ayasdi, Druker said, 100% are either already operating in this stage or actively exploring implementing such technology. The second stage of implementing machine learning is blurring the line between humans and machines in areas such as customer support. With Natural Language Processing, a computer can easily answer the majority of customer questions, leaving humans to deal only with more complex customer service problems. Druker estimates about 50% of banks are exploring or have already implemented such applications. The highest level of machine learning application is the fully automating business processes. Take a life insurer, for example. When a customer applies for a policy, he might be asked to fill out a 40-page long form and get a physical examination. That information is then sent back to the company for approval. The entire process can take over a month to complete. With a fully automated service, a computer can use personally-identifying information about a prospect to search everything it knows about him, including credit scores, purchase history, social media profiles. It can then instantaneously approve an application. Good analysis of customer data, using advanced analytics techniques, can help banks further personalize their offerings to clients, increasing customer loyalty and boosting revenues. According to McKinsey, some European banks using these techniques report 10 percent increases in sales of new products, 20 percent savings in capital expenditures, and 20 percent declines in customer churn. “Banks know this is coming, but these are still board room discussions,” Druker said. IP Soft, a company that automates IT and business processes for enterprises across a wide range of industries, recently published a white paper on a fully AI-enabled bank. Not unlike Druker's explanation, this chart from the report offer a nice summary. In a tough banking environment, banks are looking to machine learning to reduce costs and increase retention. Smaller banks with less resources are understandably having harder time implementing machine learning solutions. According to Capgemini’s report, “Big Data Alchemy: How can Banks Maximize the Value of their Customer Data?”, silos across business lines or data distributed across different systems such as CRM, portfolio management and loan servicing, hamper proper integration. “Banks lack a seamless 360-degree view of the customer," claimed the report. "Further, many banks have inflexible legacy systems that impede data integration and prevent them from generating a single view of the customer." A common pushback from banks is regulatory concern. Machine learning applications for banking are uncharted territory and neither banks nor regulators fully know yet how to handle them yet. There is a lot of work and discussions currently underway. In addition, these tools threaten the jobs of people who traditionally handled analytic tasks for banks. “Nobody want to fire all the quants,”Ayasdi's Druker said. Forward-looking banks see that with the integration of machine learning, their top talent will be freed up to spend time on more value-added tasks for the organization. CB Insights has identified 41 companies providing machine learning solutions in the financial industry. Together with the explosion of general applications of  AI, deals and investments in AI companies reached record levels in 2016. Since the beginning of 2016, over 15 fintech AI companies have closed investment rounds. Among the main areas of interest are analytics, assistance bots, market research and credit scoring.