by Brett King
The growth in capabilities behind smart assistants like Alexa is frankly unbelievable.
Figure 8: The growth of Amazon Alexa skills (Credit: Voicebot.ai).
At current growth rates, Amazon Alexa will have approximately three million skills by September 2018, and 10 million by the end of 2018. That growth is obviously unsustainable, but it illustrates the massive potential of the technology in terms of capabilities and it does closely mirror the growth in apps on app stores over the last decade (albeit faster than mobile app store growth currently).
The capabilities go far beyond skills, it also speaks to the capabilities of machines to understand us when we talk to them, or to have a conversation that is human equivalent.
Figure 9: Google’s voice recognition accuracy compared to humans (Credit: Google).
This all adds up to one undeniable trend. The capabilities of conversational commerce on smart assistants is growing at such a rate that its impact on the way we use computing technology is greater today than the internet’s potential for impact back in the year 2000. The frictionless, conversational nature of this technology will absolutely force service providers to adapt to a world where their services will have to be delivered via a voice-based technology.
Figure 10: Various categories of personal AI impact emerging over the next 5–10 years.
The seamless nature of voice will force us to create compelling, frictionless experiences where advice and utility meld together. The movements toward “open banking” will give Google, Apple and Amazon amazing abilities to incorporate this data into voice assistants. You won’t even need a bank app—these will likely become native services within the next 10 years.
Alexa: “You can’t afford to buy a new car right now, but if you sign up for Uber today, Uber will cover half of the next two years’ lease payments. You just need to agree to drive for Uber at least four hours a week. Is this something that interests you?”
Siri: “You are paying too much using a credit card today, I have other options for financing that could save you $230 per month that I could automatically link to your Apple Pay wallet. Would you like to see them?”
Voice will lead to customers learning to trust their AI assistant to recommend a day-to-day financial solution to them, rather than them going to look for it themselves. It will be like we trust Wikipedia or Google search today. We might still get an “offer” via voice, but given context and behaviour, voice interfaces might lead us in a new direction in terms of thinking about our finances that don’t fit with our current banking relationship or the products the bank currently offers. For example, if your bank doesn’t offer voice-based credit lines, then an offer for a new credit card might simply lose out to a bank who can do that via voice, on the basis of context alone.
Siri: “That’s been taken care of. We’ve extended you a $730 line of credit to pay off your son’s school fees for the quarter. The line of credit will be paid off monthly from your account, unless you tell me to pay it out. I can also suggest when you have enough funds to make an additional payment. Would you like me to do that?”
Frankly, if a bank doesn’t start thinking about the digital bank account, accessible through voice and mobile, as their primary channel for day-to-day access and advice to their customers, then they will be caught off guard in the same way banks were when both internet and mobile apps first appeared. This time, however, the risks are much greater, because the shift from product to experiences will dramatically erode the ability to simply retrofit voice onto the existing channel middleware or bank core systems architecture.
What will you need to make voice and conversational AI work? Data. To start with…
The larger problem for banks is that Alibaba/Taobao, Tencent, Apple, Amazon, Baidu, Google, and other platforms incorporating payment gateways, will often know more about their customers than the banks do. If a Beijing car dealer uses a bank debit card for a business trip to Shanghai, the bank knows what airline he or she flew, as well as the hotel and restaurants patronised. If he uses a mobile super-wallet like Alipay or Tencent WeChat, the bank knows nothing about that trip and the bank is data poor.
“If the customer ‘interface’ is happening elsewhere, the bank has zero visibility over transactions,” said James Lloyd, Asia-Pacific FinTech Leader at EY. “That’s not a good situation to find yourself in.”
—Wall Street Journal, “The Cashless Society Has Arrived—Only It’s in China”, January 2018
Voice as the customer interface will result in increasing pools of financial-related behavioural, merchant and location data that sit outside the bank ecosystem within voice or aggregated technology platforms (mobile, augmented reality glasses, etc). For banks to be able to respond to your needs, they’ll need the data that captures real-time behaviour—but Alexa, Google and Siri may not share what led up to an API request for a credit facility, they may just share the request.
Today we have three overarching pieces to the voice stack. We have the core VoiceOS and services layer, which is what handles natural language processing, search, weather, time and basic enquiries, along with installed skill activation. We also have the skills or apps that sit on top of the Alexa platform. Lastly, we have APIs that give access to smart sensors, home automation and other extensions of the platform.
Figure 11: How Amazon skills and services sit on the Alexa voice architecture (Credit: Amazon).
So, first and foremost, banks are going to have to get comfortable with working in the cloud. They can have a private cloud connected to voice services like Alexa and Siri, but they’ll get much faster capabilities on Amazon’s own architecture, which is built for purpose. The reality is that Amazon’s cloud is, in almost all instances, going to be faster and more secure than a bank’s internal, on premise architecture23.
Secondly, bank’s need a data pool that can be queried across the voice layer. For this data pool they will need to have cross-silo data integration, what we used to call a 360-degree view of the customer. But this is more about anticipating natural language queries and customer behaviour where a voice event might be triggered.
Thirdly, banks will need broad data-based and technology-based partnerships that lead to better integration of their financial services capability into real-world, real-time scenarios where they can add value easily.
Finally, banks will need voice-based and behavioural-based design teams that are intimately familiar with how people use tech like voice day-to-day and where technology fits into their life. This is a completely new skill set for banks. This is not mystery shopping one of your investment products or trying to come up with demographic-based or psychographic-based credit card offers. This is behavioural gamification, economics and psychology as a design competency. In the voice world you are an experiential solutions provider. You are not pushing an offer for an existing bank product down a new channel—if you are, you will fail!
The only way voice works for banks as a business tool is if they accept that Alexa is an extension of their voice to the customer—but it only works in a conversational manner. Pitch me a product that I don’t immediately need, and you will lose access to the channel, because I’ll block you faster than a bad Tinder date. The key skill will be anticipating the customer’s needs and responding in a frictionless manner, whether via voice, mobile, in an augmented reality head-up-display (smart glasses circa 2022–25) or similar.
Where automation will strike first
In our bank we have people doing work like robots. Tomorrow we will have robots behaving like people. It doesn’t matter if we as a bank will participate in these changes or not, it is going to happen… The sad truth for the banking industry is, we won’t need as many people as today.
—John Cryan, CEO of Deutsche Bank, September 2017
Consumer trends are clearly driving adoption of technologies like voice-based smart assistants, but from an overall perspective we can see that there will be multiple market forces pressuring FIs to adopt artificial intell
igence.
Organisational Area AI Competency/Classification Adoption Drivers
Regulatory Compliance Machine/Deep Learning Regulatory, Cost
Technology Improvement Various Supply-side Pressure/Savings
Infrastructure Advancements Cognitive, Machine Learning Competitive (FinTech), Economics
Marketing/Sentiment/Brand NLP, Machine Learning Competitive, Responsiveness
Onboarding/Acquisition NLP, Machine Learning Economics
Trading Signals Machine Learning Economics
AML/KYC/Fraud Protection Machine Learning Regulatory, Economics
Credit Scoring/Risk Assessment Machine/Deep Learning Competitive, Behavioural
Pricing/Underwriting Machine Learning Economics, Profitability
Portfolio Management Machine Learning Performance, Productivity, Consistency
Optimizing Back-office Cognitive, Deep Learning Economics, Demand-side
Procurement Machine Learning, Cognitive Productivity, Economics
Algorithmic Trading Machine Learning Competitive
Data Analysis/Personalisation Data Modelling, Deep Learning Competitive, Supply-side
Table 1: AI competencies and drivers in banking.
Whether on the supply-side, demand-side, competitive, legal or economic, there will be consistent pressure over the next decade to invest in Artificial Intelligence for profitability and best-practice operations. Broadly speaking, the top four benefits driving AI adoption will be:
1.Identifying new business opportunities
2.Automating repetitive tasks
3.Improving workforce productivity, and
4.Competing with peers
The impact will be broad in scope, but is centered initially around IT, finance/accounting, customer experience/engagement and fulfillment.
Figure 12: Where AI will have an impact on competitiveness in financial services by 2020 (Source: consultancy.uk).
As regulatory and consumer-facing technology pressures have come to bear on financial services over the last 20 years, we’ve seen a very purposeful shift to technology as a core competency. Artificial Intelligence accelerates this trend of reliance on technology for profitability over the corporate levers of asset management, marginal interest rates and so forth. As we’ll see in the next chapter, as we move away from universal banking models we’ll discover that banks that had operational advantages based on quasi-governmental or regulatory protections are heavily exposed in an arena where delivery of banking services is technology-dominated.
As these technologies come into play, the requirement to reduce fixed costs and improve agility in service delivery will be acute. Essentially, we’ll see banks increasingly having to compete with the likes of Ant Financial and neo-challenger banks whose economics are vastly different because they’ve already automated out humans for acquisition and delivery. By 2025, stock markets will be consistently asking the question of banks that have retained branch networks (or insurers/wealth managers with frontline representatives), if the economics of that real-estate can be justified when other digital competitors are scaling faster, have better cross-sell and up-sell ratios, and have higher margins due to their lower fixed costs. Defending existing distribution systems will become more difficult as AI impacts the way we think about the core operations of a bank or institution.
The reason AI will hit acquisition and customer relationships harder than back-office solutions in the near-term is not just about channel mechanics like voice. It comes down to the fact that technology in the onboarding and relationship arena actually creates a magnitude of other benefits rather than just a full-time equivalent (FTE) replacement.
We can expect an automated onboarding process, for example, to pay for itself in year one of deployment, where FTEs has been released. In year two, it means we’re already getting economic benefits from the investment. However, a robo-process for onboarding a customer can work 24/7/36524 without holidays or weekends, where time taken to open an account is typically less than one third of that of a human-involved process; it has much greater volume tolerance and scalability, and frankly will be less error prone. There’s immediate FTE benefits to be sure, but the improvements in client servicing and risk are hard to argue against.
The only remaining argument becomes: what if our customers like talking to a human?
Redefining the role of humans in banking
For decades as IVR systems were deployed you’d hear people say, “I just want to talk to a human!”. When we outsource call centre operations to offshore centres like India, you’d hear anecdotal criticism of the fact that the customer service representative answering didn’t have a local accent or local knowledge. Thus, for many years having a real, local human answer your phone when you called was considered a competitive differentiator. The principle here was both service- and advice-related—you could expect better service than a clunky IVR that you had to navigate through, and because they would have local knowledge that would lead to better advice.
In fact, broadly speaking, information asymmetry has been the foundation of financial advisory, insurance sales and most front-line customer service roles in financial services. In contract theory and economics, information asymmetry deals with the study of decisions in transactions where one party has more or better information than the other. Thus, when you wanted a mortgage to buy a property, you wanted to invest in the market, or you wanted advice on even the best credit card for you to apply for, there was a human in a bank that knew more about those products or services than you could know. Having said that, in pure informational terms, this asymmetry was often heavily biased towards products offered by the bank.
In the last 30–40 years the advice we received from a bank branch wasn’t, in fact, advice on how to buy a home or how to invest your money, as much as it was which product the bank could offer you to buy a home, or which investment product or asset class you should put your money into. If you wanted true advice independent of bank products, you would have to go to a broker, but even then brokers received commission on introductions, so their advice wasn’t unbiased. You could engage the services of a money coach or similar and then you’d get unbiased advice on money management, but that came with a direct cost. The bigger problem for humans in the financial services advisory space, however, is that the information asymmetry that justified their existence is now coming to an end due to the advent of AI.
Figure 13: Lidar point-cloud mapping in an autonomous vehicle setting (Image credit: Lidar News).
As discussed earlier, autonomous vehicles are increasingly being integrated onto our roads. Today, humans remain competitive with autonomous vehicles due to our ability to analyze and make decisions regarding driving conditions, obstacles and road markings, but our advantages are quickly being eroded.
More data, faster processing and cognition times, means better advice for the end consumer.
It won’t be long before the improvements in both sensors and cars “brains”, or processing, means that they consistently outperform humans when it comes to driving. At this point, we’ll see reduction in road deaths, decreases in insurance premiums for autonomous driving vehicles and even changes in road usage biased toward non-human drivers. It won’t be long before autonomous vehicles can process more information, more quickly to make a decision, than a human—classic information asymmetry.
Getting back to financial services though, the same will apply. As algorithms, AI customer interfaces like smart assistants gather tons of data—they will soon have much more data than an analyst or customer service representative could ever hope to absorb. With machine learning techniques and increasing error correction capability, these algorithms and AIs will quickly be able to improve the application of this data in giving customers advice in real-time that fits with their life. Whether portfolio management or investment advice, day-to-day money coaching, or help navigating credit options, algorithms will simply have information asymmetry over the human agents. They’ll have
better, more perfect data that can be applied in real-time compared to a human service professional. At this point, the roll of humans at the front-end “advice” portion of financial services will be facing the same long-term threats as truck drivers will face from self-driving, autonomous vehicles. Not to mention that the advice an AI gives us will be much more consistent from customer to customer, and won’t rely on the individual knowledge (and lack of bias) of the financial advisor.
Designing these systems, these machine-based interactions, understanding customer behaviour and creating new experiences based on emerging technology are going to be critical creative skills for the financial institutions of the future. These will remain human areas of differentiation over the next couple of decades at least. AI today and in the short term remain a collection of capabilities that humans used to do—driving a car, assessing risk, reviewing an identity document against a human, reading an email and executing a trade, etc. The leap from watching behaviour or observing an issue and redesigning a system or product or solving a design/process error is going to be beyond machines for a few years yet. AI will certainly impact elements of design also, but the overall interface between customers and the bank will be dominated by human creativity, and will be a massive area of change as banks transition from products to experiences. Your hiring practices shift focus out of the back office and related processes, and into design for the front office.
If you believe that, just because you are in a position of leadership in a bank, this doesn’t concern you, the challenges will still be based on how well you work with AI.
How to lead when your employees are Algos