Free Novel Read

Bank 4.0 Page 22


  In deep learning networks, we’ve created artificial neurons called perceptrons. These artificial neurons are the brain child of Frank Rosenblatt of the Cornell Aeronautical Laboratory, who designed these way back in 1957. Initially designed for image recognition, the first perceptrons were hard-wired logic circuits, and not the software-based code they are today.

  In true computing terms the perceptron works with inputs applied to an algorithm as a binary classifier as it learns. For example, if the algorithm is being used to learn to distinguish between a cat and a dog, the algorithm applies varies vectors (inputs) against a bias to create a linear decision boundary. Simply stated, the algorithm filters inputs to produce a one or a zero output, but over time the algorithm can adapt its bias (shifting the linear boundary) as it learns to produce more and more accurate results. The capability to correctly identity the difference between an image of a cat and a dog gets better and more precise over time.

  Figure 3: Deep learning neural networks use artificial neurons called perceptrons (Image credit: Christoph Berger).

  Figure 4: Perceptron updating its linear boundary as more vectors are added (Image credit Wikipedia).

  Historically, if we talked about how humans differentiate from technology, it was always about our ability as humans to recognise patterns, think creatively, understand abstract concepts, etc. By teaching machines to learn, it’s clear that our ability to recognize patterns or to reason on things is no longer the clear advantage it once was.

  There are a variety of techniques used in deep learning, such as single versus multi-layer perceptrons, backpropagation networks, alternate step-functions and linear vectoring, but you don’t need to be an AI expert to understand that AI is already starting to impact broad swathes of society.

  We are already losing out to machines

  In the European Union, United Kingdom, United States, United Arab Emirates, Singapore, Hong Kong, Australia and many other countries today, if you enter a port with a biometric capable passport, you’ll have the option to go through an e-gate or similar. It might be a fairly obvious statement, but the reason for this is simple—computers today do a much better job at recognizing a face or verifying your identity than a human customs officer ever could. Research shows that facial recognition software is 15–20 times more accurate at identifying a customer than a typical face-to-face interaction9. A fact that would indicate face-to-face bank account opening is no longer safe, incidentally. Statistically based on software comparators, it is probably the single riskiest thing a bank could do in this day and age.

  China has developed a national identity database that can identify any one of the 1.4 billion Chinese citizens via software in two-to-three seconds10. While many in the West might scream about the civil rights issues associated with such policies, the fact is we’ve been seeing this sort of tech fictionalised in movies and so forth for decades now. Most modern policing organisations already have this capability, and the technology is maturing for a very simple reason. Governments can trust this tech to work better than human eyeballs.

  How many of us would want our borders compromised by inferior technology today? Wouldn’t we all want the best chance of catching a criminal or identity thief? In these scenarios, it’s pretty straightforward to prove that algorithms, biometrics and identity databases can consistently outperform human workers.

  In airports the applications are straightforward. Airlines like JetBlue and Finnair are trialing facial recognition systems to bypass checking your boarding pass at the gate. Before long you may be able to enter the airport, board your flight and pass through customs at the other end just by using your face. The golden age of travel may return simply thanks to biometric tech powered by an algorithm.

  So what are we to make of the insistence by those banks and regulators that to open a bank account you must have a human physically present themselves in a branch? In the light of broader trends in identity verification, a requirement for a human bank officer to facilitate account opening is an anachronism. Very soon, based on both cost and performance, humans won’t be competitive when it comes to the front line on the basis of identity verification alone. If your business is built on in-branch customer acquisition, you will find that AI capabilities in general are a big threat to your primary acquisition approach.

  Some of the broad areas where artificial neural networks are already outperforming humans include:

  •Image and pattern recognition

  •Board and video games

  •Voice generation and recognition

  •Art and style imitation

  •Prediction

  •Website design/modification

  Between 2009 and 2016, machine intelligent HFT algorithms accounted for 49–73 percent of all US equity trading volume, and 38 percent in the EU in 2016. On 6 May 2010, the Dow Jones plunged to its largest intraday points loss, only to recover that loss within minutes. After a five-month investigation, the US Securities and Exchange Commission (SEC) and the Commodities Future Trading Commission (CFTC) issued a joint report that concluded that HFT had contributed significantly to the volatility of the so-called “flash” crash. A large futures exchange, CME Group, said in its own investigation that HFT algorithms probably stabilised the market and reduced the impact of the crash.

  For an industry that has developed trading into a fine art over the last 100 years, HFT algorithms represent a significant departure from the trading rooms of Goldman Sachs, UBS and Credit Suisse. The algorithms themselves have departed significantly from typical human behaviour. Very different behaviour and decision-making has been observed when analysing HFT trading patterns. What has led to this shift?

  Perhaps it is the fact that HFT has neither the biases that human traders might have (for instance, staying in an asset class position longer than advised because the individual trader or asset manager likes the stock or the industry) nor the same ethical basis for making a decision. While some might argue that Wall Street isn’t exactly a bastion of ethics, the fact is, an HFT algorithm simply doesn’t have an ethical angle for decision-making (unless those skills have been programmed in). Those deep-learning algorithms have created different linear boundaries from humans doing the same job.

  While HFT has been pioneered by the big trading companies, and has certainly helped them, what impact are algorithms having on investment portfolios and wealth management?

  Robo-advisors, robo-everything

  As with the other trends we’ve seen in the Bank 4.0 world, the first movers in the robo-advisor space were the FinTech startups. Betterment launched in 2010, and for CEO Jon Stein, “one of the most satisfying results of the work we started seven years ago is seeing the entire industry change”. That change is a tacit acceptance that human advice is a marginal value add, and when it comes to portfolio performance over the medium-term, robo-advisors may offer an opportunity for rebalancing and optimization consistent with your return expectations, that humans won’t efficiently be able to match.

  I’ve met Jon numerous times, interviewed him on my radio show, and I like the fact that he’s largely a quiet achiever. You don’t hear a lot from Jon in the media for months at a time, and he lets the results of Betterment speak for themselves. I am also a big fan of Betterment’s startup story, as the first robo-advisory firm, because it demonstrates his tenacity.

  Figure 5: The Robo-Advisor revolution (Credit: Barrons).

  Jon Stein and his roommate, Sean Owen (a Google software engineer), started building Betterment’s platform back in 2007. Stein taught himself to code in order to build the early prototypes behind the platform. However, starting a business in a highly regulated industry that would require licensing and other compliance-related competencies required more than technical competency. Before starting Betterment, Stein had attended weekly poker games with Eli Broverman (circa 2003–4). While Eli had come out of those poker games better off than himself (according to Stein), it was a relationship that allowed Stein to tap Broverman, a
securities attorney, for help with the startup in the early stages. In 2007, while Stein was still studying at Colombia University’s Business School, he and Eli met up for lunch at a Dominican restaurant on the Upper West Side and sketched out a plan to move forward with the ugly regulatory stuff that would otherwise have held Betterment back.

  By 2008, the small team, including Jon’s girlfriend (his wife today) doing graphic design, were working on funding and the launch platform. The licensing and business formation followed in 2009, and then in 2010 they launched at TechCrunch, with Chris Sacca (of Shark Tank fame), leveling some pretty tough criticism their way: “I worry that it’s too simple. People don’t always trust it. People expect a little bewilderment that gives it credibility. This starts to feel a little like a toy.”

  Today that toy manages more than $10 billion in AuM (assets under management), and Betterment’s growth is estimated at around 106 percent annually, although it appears to be slowing as they get larger (it was around 300 percent just three years ago). Stein says he is aiming for $1 trillion in AuM, so they have big aspirations and more growth to go. To reach that goal, however, there will need to be a substantial shift in behaviours around investing.

  Today, we’re at a pivot point for personal investing. In the past the assumption was you’d need both advice and financial literacy in order to be able to successfully invest as an individual. That’s a problem today as the data shows that financial literacy amongst Millennials is actually significantly worse than that of their forebears11. A survey back in 2015 conducted by Bank of America U.S. Trust found that just 47 percent of multimillionaires aged 18–34 use a financial advisor12. For those Millennials that aren’t multimillionaires, the statistics are even worse.

  Assuming that Millennials will be both literate enough to invest and seek out human advisors in the future is a big assumption. The emergence of automated investing tools like Stash, Digit and Acorns, and the development of robo-advisory tools seems more likely to fill this gap in skills and behaviour.

  When doing research for this chapter, I tried to find portfolio return comparisons between human advisors and robo-funds. From a moderate-risk portfolio perspective, when I used to work with private bankers and wealth advisors back in the day, we’d look for 10–12 percent annual returns as a safe assumption on a longer-term investment horizon. Typically, this would be a mix of equity and income producing bonds.

  Robo advisors today are performing right in that range of expected returns. Barron’s conducted a survey of robo-advisors over the 2016 calendar year and found annual returns for the better-performing robos were in the 11–12 percent range13. That’s also consistent with the average annualised return of the S&P 500 Index, which was 11.69 percent from 1973–201614. BI Intelligence forecasts that robo-advisors will manage around $1 trillion of AuM by 2020, and around $4.6 trillion by 202215. As a trend, by 2030 we would expect robo to dominate the mass market investment industry.

  On a portfolio performance basis then, the difference between a robo-advisor and a human-based asset management firm are negligible. Certainly, if you are willing to take greater risks with your portfolio, or you are investing larger amounts in more diversified pools or structured products, then you may find that a human team can perform with higher results. However, the firms and advisors that produce those results typically require a minimum investment that is out of reach of 99 percent of the population. Thus, it seems entirely reasonable that robo advice will come to be seen as one of the greatest tools for large-scale affluent wealth management since the creation of “premier” banking. Accessible, automated portfolio management without the friction.

  Seeing this trend emerge, ICBC in China has made a big bet on AI and robo advice. Their robo-advisor tool doesn’t require a traditional risk profile questionnaire to get started. It learns from your investment style along the way and teaches you how to invest in your optimal range for your level or risk tolerance and your return expectations.

  Figure 6: ICBC’s Rong-e line of products, including AI投 or robo-investment.

  ICBC’s AI投, or AI invest, represents what is certain to become a baseline capability for wealth management capabilities moving forward. It also will fundamentally change the legal and compliance requirements we have around “risk” for basic investment or wealth management. For the last three years, every time I’ve done an annual “risk” review with my bank in Hong Kong, HSBC, they’ve used a telephone to record our conversation so there is an audible record of me accepting the risk conditions. Each time I go through my annual review (which generally takes about an hour), at least 75 percent of the time is spent on compliance-related activities. They do all of this for regulatory and legal reasons.

  With AI managing risk tolerance and optimising your portfolio for your required investment horizon and return expectations, the whole regulatory process required by the FSA, SEC, etc involving signing a piece of paper or a telephone system legally recording my formal response to a risk tolerance questionnaire will be quickly undermined. Human advisors will just look slow, clunky and bound by friction. Robo advising will quickly become the benchmark on experience, and then on asset management performance. Regulators will be forced to adapt too.

  For those of you still sceptical about robo advice generally, it would be helpful to step back and see where AI-based advice fits from a technology perspective, rather than simply trying to articulate it as humans versus machines.

  A bank account that is smarter than your bank

  If you can imagine technology like Siri, Google Home or Alexa maturing in the next five years and being able to order you not just socks16, but a pizza, and book Uber rides, flights, restaurant reservations and doctor’s appointments. Once commerce is integrated into our tech so seamlessly, the next obvious area to tackle is day-to-day money interactions and financial advice.

  If you think this sounds like a science fiction story, you are in for a big shock. Remember that back during the dot-com boom (or bubble), the majority of non-tech press was extremely skeptical about the effect e-commerce would have on retail businesses. Today online shopping dominates choice, with many categories of retail showing 50 percent or more of sales are either influenced by or initiated via the web or mobile17. For Christmas 2017, it was projected that almost 40 percent of all sales were done online18, and Amazon owned the largest percentage of that. That shift in consumer behaviour has been devastating for retailers, with 7,000 stores closing in the US alone in 2017 (which is a 300 percent increase from 2016). In the UK it is expected that more than 5,000 stores will close in 2017, but that is actually down on previous years.

  In markets like China, mobile commerce now dominates day-to-day retail activity for a wide range of segments, and today 75 percent of all e-commerce is mobile-led in China19. In parallel to a growing middle class, this mobile bias is creating slower growth in retail stores than we would have expected, given China’s economic growth. The big growth is certainly centered around online portals more than physical retail, and the erosion in physical retail is plainly apparent20.

  In the near future you’ll be making these everyday purchases increasingly via voice on a smart assistant built into your home and smartphone. Voice assistants are already being used to make purchases by 40 percent of Millennials, with that number expected to exceed 50 percent by 202021.

  So why is this trend towards mobile and voice commerce so important for banks to take note? If you live in a developed economy or an urban centre like Tokyo, New York or London, there’s a fairly good chance that if you were ordering a take-out dinner for delivery, that you’d be using an app. If I asked you to check your balance, chances are you’d likely use the same approach. Today more than 50 percent of customers in most developed economies use their mobile for checking their balance versus any other bank channel. Twenty years ago it was dominated by ATM or phone banking. In 10 years it will be dominated by voice-based or agency-based commerce engines22.

  Consumer: “Alexa, what
’s my account balance?”

  Consumer: “Siri, has my salary hit my account yet?”

  Consumer: “Google, how long will it take me to get to the office if I leave in two hours?”

  Figure 7: How people use smart speakers on a daily basis (Source: NPR and Edison Research).

  Now, you would be wrong to dismiss this as simply another channel in the bank arsenal, because this is the start of actually redefining your day-to-day relationship with technology, not just your bank account. Voice has the potential to become the underpinning of day-to-day advice for you and your money, but increasingly it will be just the way you access a range of basic technology capabilities. ComScore says 50 percent of search will be voice-based by 2020, and commerce is obviously going the same way. But search leads to conversational commerce, which is more than just asking a question—it’s a dialogue.

  Increasingly we’ll be asking our bank, via Google, Siri or Alexa, whether we can afford to go out for dinner; or, at my current rate of savings, when I can afford a deposit on a house or to buy that replacement vehicle I’ve been looking at; or what I need to do to pay down my credit card debt faster (if you still use plastic). Ask and ye shall receive. Voice will combine natural language, search and AI to provide answers to these questions much faster than through a branch or web channel. Primarily because voice will emphasize the utility of the bank to solve these problems, not directing you to a product to download via a channel.