20190604 Artificial Intelligence In Banking SA

EU MonitorGlobal financial marketsArtificial intelligence in bankingJune 4, 2019A lever for profitability with limited implementation to dateAuthorOrçun Kaya 49 69 910-31732orcun.kaya@db.comArtificial intelligence (AI) is a significant step forward in the digitalisation andtransformation of modern businesses. In short, it refers to computers’ capabilityto acquire and apply knowledge without programmers’ intervention.EditorJan SchildbachInvestors are lining up to be part of the imminent change. AI attracted USD 24bn in investments globally in 2018, a twelvefold increase since 2013. US startups received the most attention, followed by Chinese, which already outpacedEuropean AI start-ups.Deutsche Bank AGDeutsche Bank ResearchFrankfurt am MainGermanyE-mail: marketing.dbr@db.comFax: 49 69 910-31877www.dbresearch.comDB Research ManagementStefan SchneiderWithin Europe, Germany, France and the UK are the frontrunners inexperimentation and in the implementation of AI. In light of intensified globalcompetition, the European Commission proposed a EUR 9 bn budget to fundAI-related projects between 2021 and 2027.Similar to earlier examples of information technology (IT) implementation infinancial services, AI promises great efficiency gains and potential revenueincreases. To date though, AI implementation in banking has been modest. AI isbeing tested for real-time identification and prevention of fraud in online bankingas well as in know-your-customer (KYC) processes. Robo-advisors are alsoevolving over time to become true AI solutions. Looking forward, regulatorymeasures around data privacy and concerns regarding cybersecurity mightcreate obstacles to AI use in banking. In addition, the highly regulated nature ofbanking may cancel out some efficiency gains of AI.AI’s potential contribution to bank profitability should not be underestimated.Empirically, AI has a significant positive impact on European banks’ return onassets (ROA). By increasing labour productivity, AI technologies couldstructurally reduce costs in the banking sector. Rapid implementation of AItechnologies is, therefore, central to fighting persistently weak profitability and toremaining competitive.

Artificial intelligence in bankingIntroductionHuge progress in computer hardware, software and internet technologies haveirreversibly changed our societies. It is now difficult to imagine an economicagent without computers, internet or mobile devices. The pace at which IT isevolving offers great opportunities to expand the client base, introduce newproducts or improve existing ones and to increase efficiency in a relatively shortperiod of time. On the other hand, if companies miss out on the current IT wave,they might be overtaken by events soon.Among the various IT breakthroughs of recent years, the advancement in AI isparticularly remarkable. In short, AI refers to computers having cognitive skillssimilar to humans, which could result in immense efficiency gains for firms andtheir clients alike. The financial sector has been one of the early experimenterswith AI technologies, not least due to its likely contribution to strongerprofitability. It is therefore essential to take a closer look at the potential role ofAI in banks’ digital transformation.Artificial intelligence: A giant step beyond standard IT applicationsMachine learning1There are various approaches to programmingcomputers so that they mimic human decisionmaking. Decision trees, ranking or prioritisingare among the more established solutions. Arelatively new approach is machine learning(ML). ML is a subset of AI and refers tocomputer programs that recognise patternsand make predictions based on them. Typicalexamples are internet platforms thatrecommend particular products or news storiesto users who might like them based onprevious preferences. By continuouslyanalysing new data and scenarios, ML toolsmake adjustments to decision-makingprocesses without being specificallyprogrammed to do so. They are therefore ableto learn from data. Subcategories of ML aredeep learning, as well as supervised,unsupervised and reinforcement ML.ML tools process vast quantities of datathrough neural networks. In short, these areprocesses which classify data on successivelayers. In doing so, they rely on theprobabilities of possible outcomes. They makedecisions based on the most likely outcome,even though it might turn out not be the perfectchoice in the end. However, neural networksinvolve a feedback loop. Depending on theaccuracy of the outcome from previous trials,they update their approach to perform betterthe next time.Source: Deutsche Bank ResearchTo date, IT solutions in the business world have by and large focused onautomating repetitive tasks that would otherwise require human involvement.The boundaries for these IT applications have been set by their developers, andby design these solutions have been limited in their capabilities. They havelargely been static and unable to comprehend or act on their own. Withtechnology evolving rapidly, however, this is increasingly changing.Artificial intelligence refers to the ability of computer programs to acquire andapply knowledge without human intervention and involvement. By observing theworld around them and analysing information autonomously, AI systems drawconclusions and take appropriate actions. They learn from their previousjudgements and, depending on the level of accuracy, improve their performanceover time.AI as a term was first coined at the Dartmouth Conference in 19561 and is notnew per se. In recent years, though, some breakthroughs in IT have allowedtremendous momentum in AI’s capabilities:i) The expansion of internet usage has led to huge amounts of digitalinformation being generated and stored. In about 10 years, the amount ofdata generated worldwide grew some 17 times. Forecasts point to anotherfivefold increase between now and 2025. This large volume of information,once cleaned and structured (i.e. big data), is at the core of data-drivendecision-making.ii) There has been a colossal increase in the processing power of computers. Astandard measure of that, the number of transistors, has increased 10 mtimes since the 1970s. The speed of central processing units, anotherelement contributing to processing power, rose by a factor of 6,750 over thesame period.2 This enables algorithms to process information at much fasterrates and contributes to the accuracy of their decision-making.iii) Other developments – such as the reduction in data storage costs,advancements in data mining processes or an increasing number of ITexperts – have further fuelled the feasibility and capability of AI. While the122 June 4, 2019See McCarthy, Minsky, Rochester, Shannon (1955).Moore’s Law states that the processor speed of computers, or the number of transistors on anaffordable central processing unit, doubles approximately every two years.EU Monitor

Artificial intelligence in bankinghard drive cost per gigabyte has come down from around USD 5,000 in 1990to some USD 0.025 today, the number of IT specialists grew by 50% in theeuro area between 2007 and 2017, for example.Big data as input, data identification methods such as machine learning and thegreater affordability of these tools have been the driving factors behind AI’srecent rapid success in understanding languages, recognising objects andsounds, and observing and solving problems autonomously.Artificial intelligence investments on the riseStrong growth in VC investments in AIstart-ups globally2USD bn2520151050131415161718**estimatedSources: OECD, Deutsche Bank ResearchPatents in AI: China outpacing the EU3% of total global patents in AI*4030201001015US1015EU-281015JP1015CN*AI-related patents are patents in cognition, meaning andunderstanding, large-capacity and high-speed data storage,high-speed computing, large-capacity information analysis.Thanks to its rapid evolvement in recent years, AI is being experimented withand implemented in several areas. Due to measurement issues, however,quantifying its deployment is hardly a straightforward task. Indeed, firms mightdeploy AI to increase efficiency in their processes, which is not directlyobservable for analysis. Moreover, it is sometimes difficult to differentiatebetween more standard IT solutions and sole AI applications. To partlyovercome these drawbacks, information on venture capital (VC) investments inAI start-up firms may be useful.3 In 2018, AI start-ups received a staggeringUSD 24 bn globally, up from less than USD 2 bn in 2013. Growth in VCinvestments over the past two to three years has been particularly strong. AIfirms have also increasingly become acquisition targets. Over the last 20 years,a total of 434 companies in the AI sector have been acquired, 220 of them since2016 alone.4Of the total VC volume in 2018, almost USD 15 bn went to AI start-ups in theUS, and another USD 6.5 bn went to Chinese firms. In 2017 and 2018, thenumber of VC deals flattened out. Yet the average volume of VC investmentssurged, an indication of VC flowing into more mature AI firms whose capitalneeds are larger than those of typical seed stage start-ups. In China, forexample, SenseTime Group, a computer vision and deep learning technologydeveloper, raised USD 1.6 bn in VC funding in 2018. With the new capital, thevalue of the company rose to over USD 6 bn, making it the world’s mostvaluable AI unicorn. In the US meanwhile, it is primarily large tech firms whichinvest in AI start-ups.To VC investors, AI appears to be a truly transformative technology withsignificant potential, like the internet and mobile revolutions in past decades.How do AI start-ups use the funds they receive? First observations indicate thatthey hire new AI talent (which proves to be costly and difficult to find) andexpand their services. Investors might therefore need to wait a while before theysee meaningful returns on their investments.Artificial intelligence and intellectual property rightsSources: OECD, Deutsche Bank ResearchA technological field usually is more useful and has a greater value for theeconomy in the years to come if there is a substantial increase in the number ofpatents filed in this particular field.5 There were some 20,000 patent applicationsin AI-related technologies in 2016, double the figure of 2010. Around 50% ofthat was accounted for by AI patents in computer vision. This technology ismostly used in self-driving cars and shows how intense competition in this areacurrently is. Of the total AI patent applications in 2015 (the latest available data3453 June 4, 2019The OECD identifies AI start-ups as firms whose business model focuses on i) “artificialintelligence”, “machine learning” and “machine intelligence”; ii) “neural networks”, “deep learning”,and “reinforcement learning”; and iii) “computer vision”, “predictive analytics”, “natural languageprocessing”, “autonomous vehicles”, “intelligent systems” and “virtual assistant”.See WIPO (2019).See Inaba and Squicciarini (2017) for a detailed explanation.EU Monitor