www.ifc.org/thoughtleadership NOTE 85 • JUN 2020 Artificial Intelligence Innovation in Financial Services By Margarete Biallas and Felicity O’Neill Artificial intelligence technologies are permeating financial services sectors around the world. The application of these technologies in emerging markets allows financial service providers to further automate their business processes and to leverage new and big data sources to overcome obstacles— including the high cost of serving rural and low-income customers and establishing customer identity and creditworthiness—that prevent the delivery of financial services to many consumers. Realizing financial inclusion benefits through the adoption of artificial intelligence relies on its responsible adoption by firms, on competitive market settings, and on continued investment in the necessary infrastructure. Artificial intelligence (AI) was established as a discipline providers (FSPs) to begin integrating AI technologies into some 70 years ago, but its applications have accelerated in their service offerings. A recent survey of 151 firms, which recent years, supported by an evolution in machine learning was conducted jointly by the World Economic Forum and and improvements in computing power, data storage, and the Cambridge Centre for Alternative Finance and included communications networks. This note defines AI as the both financial technology (Fintech) firms and incumbent science and engineering of making machines intelligent, banks, suggests that this is indeed happening, with 85 especially intelligent computer programs (see EM Compass percent of respondents saying they are “currently using Note 69).1 AI can therefore be characterized as a series of some form of AI.”3 systems, methods, and technologies that display intelligent In emerging markets in particular, the need for AI stems behavior by analyzing their environments and taking from the fact that individuals and businesses are often actions—with some degree of autonomy—toward achieving underserved because they lack the traditional identification, prespecified outcomes. 2 collateral, or credit history—or all three—needed to access Reductions in the cost of Internet connectivity, increased financial services. AI can address this problem by providing mobile device penetration, and increased computing analytically sound alternatives to determining the identity power over the past decade have helped digital consumers and creditworthiness of individuals and businesses, based and businesses generate a wealth of new and real-time on alternative data collected from mobile phones, satellites, data through mobile phones and other digital devices. and other sources. Concurrent advances in data storage, computing power, An additional obstacle to emerging market customers energy reliability, and analytic techniques have made it accessing financial services is cost—the cost to reach and cost-effective for businesses to analyze this wealth of real- serve these customers is often too high relative to the size of time and alternative data. As a result, it is now increasingly their financial transactions and the revenue they represent. commercially viable for emerging market financial service About the Authors Margarete Biallas, Digital Finance Practice Lead and Senior Operations Officer, Financial Institutions Group, IFC. mbiallas@ifc.org. Felicity O’Neill, Global Partnerships Associate, Partnerships and Multilateral Engagement, IFC. foneill1@ifc.org. 1 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. AI can help address this problem, too, by automating various canvasses challenges to the responsible and sustainable processes—customer service and customer engagement use of AI by emerging market FSPs. It also outlines what are a few obvious ones—to reduce costs. In this way, AI actions investors and development finance institutions like can enable higher volumes of low-value transactions and, IFC can take to ensure that AI is deployed to maximize by doing so, begin to turn these formerly underserved financial inclusion. individuals into potentially profitable customers and include them in the contestable market for FSPs. AI applications: Analyzing New and Complex Data Sets Thus, to the extent that the use of AI by emerging market FSPs results in the extension of services to previously The first broad application of AI by emerging market FSPs underserved individuals or underfunded businesses, these is to analyze alternative data points and real-time behavior technologies have the potential to enhance financial to more effectively: (1) improve credit decisions; (2) identify inclusion. Yet the pace and extent of adoption, and hence threats to financial institutions and help meet compliance the degree to which inclusion benefits are realized, relies obligations; and (3) address financing gaps faced by on efforts by government, businesses, and investors to businesses in emerging markets. generate institutional and market settings that facilitate Improving credit decisions. Lenders and credit ratings the responsible and sustainable integration of AI into agencies routinely analyze data to establish the financial services. This includes FSPs generating trust by creditworthiness of potential borrowers. Traditional data lending responsibly, addressing algorithmic bias and error, used to generate credit scores include formal identification, managing cyber risk, and striving for informed consent bank transactions, credit history, income statements, in the use of consumer data. These settings also rely on and asset value. In emerging markets, underbanked supervisors enhancing their capacity to regulate algorithms, individuals—and particularly women—do not always have and authorities continuing to foster a competitive access to the traditional forms of collateral or identification environment for financial services. that creditors need to extend financial services. By using This Note explores the early applications of AI in the alternative data sources—public data, satellite images, financial services sector in emerging markets, and company registries, and social media data such as SMS and BOX 1 FarmDrive Kenya-based FarmDrive is an agricultural data FarmDrive’s algorithm is currently in its second stage. analytics company delivering financial services to During the first phase (the pilot), which ran between unbanked and underserved smallholder farmers, December 2015 and December 2016, the company while helping financial institutions cost-effectively collected environmental data (weather and climate increase their agricultural loan portfolios. Using patterns and soil data), economic data (income and simple mobile phone technology, alternative credit market data), and social data such as social network scoring, and machine learning, FarmDrive closes the information including apps usage and individual data, data gap that keeps smallholder farmers from the from participating farmers. The aggregated data is fed financial services that would allow them to grow into FarmDrive’s algorithm, which generates credit their agribusinesses and increase their incomes. scores that can be used by financial institutions. FarmDrive collects a farmer’s data using questions In its next phase of development, FarmDrive will seek and answers via text messaging. The questions to expand the environmental arm of the algorithm are designed to identify the farmer’s location, by incorporating more alternative datasets, including crops cultivated, farm size, assets such as tractors, satellite imagery and remote sensing data. In and farming activities. This data is combined addition, FarmDrive plans to use these environmental with existing agricultural data to develop a credit datasets in combination with crop cycle data to predict profile. FarmDrive also uses testing to determine seasonal yields and influence agricultural insurance the likelihood that a farmer will repay a loan. The products. When smallholder farmers have access to aggregate profile is then shared with lending credit, they can sustainably contribute to economic institutions for credit assessment and funding. development while improving their livelihoods. 2 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. portfolios. For example, African FinTech MyBucks BOX 2 Branch provides microloans and insurance directly to customers in 12 countries, including Zambia, Malawi, and Uganda, Branch is a mobile app digital lender operating in by applying its AI technology Jessie to scrape data from a Kenya, Nigeria, Tanzania, Mexico, and India. Since potential borrower’s phone to generate a lending profile.5 its establishment in 2015, Branch has provided The use of predictive scoring reduced the default rate on more than 15 million loans to over three million MyBucks’ loan portfolio in South Africa by 18 percent customers, disbursing a total of $350 million. over the 2017–18 financial year.6 However, it is too early Branch applies machine learning to create to tell whether AI will facilitate a net improvement in an algorithmic approach to assess the non-performing loan (NPL) rates, with responsible lending creditworthiness of potential borrowers based practices by FSPs still critical to market sustainability on thousands of data points on the individual and regardless of whether AI is used to assess creditworthiness. the accumulated experience across borrowers. A potential borrower downloads the Branch app, Identifying threats. Financial institutions are vulnerable verifies his or her identity, and provides consent to a wide range of risks, including cyber fraud, money for Branch to access the customer’s smartphone laundering, and the financing of terrorism. In order to data. Branch applies its algorithms to data like combat these threats, financial institutions undertake text messages, call logs, contacts, and GPS, know-your customer (KYC) and anti-money laundering and combined with a borrower’s loan repayment countering financing of terrorism (AML/CTF) compliance history, to make a lending decision. The system activities, among others, to verify the identity of their creates personalized loan options in a matter customers, to understand the purpose and intended nature of of seconds, allowing Branch to approve a loan transactions between individuals and businesses, to conduct within minutes. Loan durations range from a few ongoing due diligence to ensure that transactions match weeks to more than a year, with a typical loan of customer profiles, and to meet regulatory requirements. around $50. Underwriting loans of this size would Detection of fraud and anomalies is among the most not be viable at scale using traditional credit commonly cited reasons for adoption of AI by financial assessment methods. service providers.7 And risk management is currently one of the most common uses of AI technology in financial services sectors. According to IFC research, more than 250 regulatory technology companies (RegTechs) provide their messenger services interaction data—AI can help lenders services worldwide. A strong focus of these technologies is and credit-rating institutions assess a consumer’s behavior on suspicious transaction monitoring, where AI is used to and verify their ability to repay a loan. For example, Box 1 identify anomalies in user behavior.8 In emerging markets, illustrates how FarmDrive aggregates alternative datasets to KYC compliance is difficult because many individuals build credit scores for smallholder farmers in Africa. lack primary identification documents, registries are often Start-ups like Kenya’s mSurvey, a mobile survey platform patchy, and there is a lack of confidence in some sources of that drives decision-making for businesses across Africa government data needed for verification. Yet it is critical and the Caribbean, are using mobile phone applications that emerging market FSP’s meet KYC requirements, to generate the requisite data needed to feed scorecards because they underpin correspondent banking relationships and build real-time profiles of local consumers.4 As that allow individuals and businesses to send and receive the results of consumer scorecards are fed back into a payments across borders. This matters for financial inclusion machine-learning system, the algorithms improve, refining because remittance flows between markets are now the which data points are the most predictive in assessing largest source of foreign exchange earnings in low- and creditworthiness. Once the AI system is in place, predictive middle-income countries, excluding China.9 scorecards are inexpensive to generate for consumers who AI-enabled compliance technology can reduce the cost have access to mobile phones, which means the addressable for FSPs to meet KYC requirements and decrease false market for financial services is greater and the cost and positives generated in banks’ monitoring efforts by sifting speed of underwriting loans is lower, enabling FSPs to through millions of transactions quickly to spot signs extend services to underserved consumers (Box 2). of crime, establish links, detect anomalies, and cross- Early evidence suggests that predictive scorecards may check against external databases to establish identity also help reduce rates of non-performance across loan using a diverse range of parameters. McKinsey estimates 3 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. that AI-algorithms can help reduce the number of false credit risk, and help predict fraud and detect supply chain reports by 20 to 30 percent, though they also observe that threats in real time and cost-effectively. many financial institutions have been slow to adopt these For example, Tradeteq is a platform that provides investors techniques because the algorithmic “black box” is often too and SCF originators with the technology to negotiate, analyze, difficult to validate for the purpose of meeting supervisory and manage trade finance investments, using alternative data requirements.10 In addition, to date, the cost of the enabling to provide credit analysis and facilitating originators to pool software is beyond what many emerging market FSPs can assets, with the objective of reducing the structural costs that afford. However, research conducted by IFC indicates that drive the trade finance gap.13 Although costly, the accessibility for emerging market FSPs, the cost can be reduced through of services like Tradeteq for FSPs have been improved through shared services arrangements (see IFC EM Compass Note software pricing models based on optional use-of-service, 59, “How a Know-Your-Customer Utility Could Increase rather than upfront capital expenditure models. At the same Access to Financial Services in Emerging Markets”). time, AI solutions in trade finance are limited by the extent The volume of digital financial transactions—remittances, to which SMEs along the supply chain have digitalized their savings deposits, and online purchases—is growing at operations.14 Nevertheless, continued innovation to reduce double-digit rates annually. The growth in the value and the structural costs that sustain financing gaps in emerging volume of these transactions exposes financial services markets, such as trade finance, is a nascent benefit of AI in the firms to fraud and cyber-attacks, with downside risk financial services sector. to firms’ reputations. A 2017 CGAP survey of digital financial services companies in Kenya, Tanzania, Zambia, AI Applications: Automating Business Models to Uganda, and Ghana found that unplanned system outages Differentiate Services and Capture Market Share due to events like cyberattacks decrease customer trust.11 The second broad application is the use of AI by emerging Like KYC compliance, leveraging the predictive and market FSPs to automate business models and processes learning capabilities of AI through security software to to lower the cost of transacting with a wider range of identify and manage cyber threats will help FSPs maintain consumers. This includes lower-income consumers and confidence in the security and integrity of transactions for businesses who are benefitting from access to financial customers and correspondent banks. However, software- products that are tailored to their specific needs through as-a-solution packages to monitor and address cyber and the use of AI. fraud risk are currently prohibitively expensive for many emerging market FSPs, preventing the potential benefits Increasing access through process automation. AI software from being fully realized. helps automate aspects of digital financial services such as customer engagement and customer service, reducing Addressing financing gaps: the case of supply chain finance. the cost to FSPs of extending tailored support to a wider Globalization has increased the scope and complexity of range of consumers. Juniper Research estimates that banks supply chains. FSPs take on credit risk for supply chain globally will save $7.3 billion in operating costs by 2023 transactions by intermediating the financial instruments through the use of chatbot applications.15 An example is such as loans and cash management that enable trade Bank BCP in Peru, which has partnered with IBM Watson between buyers and sellers. The Asian Development to develop a personalized chatbot, Arturito, that facilitates Bank estimates that there is a global trade finance gap of customers in converting currencies, meeting credit card $1.5 trillion, which is driven by the relatively high cost repayments, and accessing 24-hour customer support via of assessing firm creditworthiness and meeting KYC and Facebook. Similarly, Brazil’s Banco Bradesco has worked AML/CTF requirements, particularly for emerging market with IBM Watson to develop a chatbot that answers small and medium enterprises.12 The application of AI by 283,000 questions a month in relation to 62 products, with originators of supply chain finance (SCF) has the potential 95 percent accuracy.16 to help bridge this trade finance gap. This tailoring and automation has financial inclusion Originators of supply chain finance now have access to a potential if it facilitates the extension of financial services greater wealth of data about the behavior and financial to individuals and businesses that might have been deterred health of supply chain participants. Machine learning from accessing financial products due to an inability to algorithms can be applied to these alternative data- transact in their own language or to physically access a points—records of production, sales, making payments branch or banking agent. For example, IFC client MTN on time, performance, shipments, cancelled orders, and in Cote d’Ivoire is working with tech company Juntos to chargebacks—to create tailored financing solutions, assess 4 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. incorporate AI-support into its digital wallet MoMo, so that customers can better understand their financial products and obligations. To date, 95 percent of MTN’s digital dialogue conversations have been successfully automated. This use of chatbots and language processing to help address trust and financial literacy barriers for consumers in accessing financial services remains an underexplored application of AI in emerging markets. Personalized Banking. To date, the high cost of developing FIGURE 1 Example of a Chatbot in Peru personalized relationships with clients has restricted “Chatea con Arturito” or “Chatting with Little Arthur” is an automatic response channel working as a chatbot that is offered by Banco de Crédito “relationship banking” by financial institutions to large (BCP), the largest bank in Peru. Its conversations can be followed at companies and high-net-worth individuals. FSPs are https://www.facebook.com/pg/ArturitoBCP/posts/. increasingly looking to differentiate their services to Source: Banco de Crédito (BCP), Peru. attract greater market share by using AI and big data (sets of structured and unstructured data) to automate inclusion barriers. For example, weather risk transfer an assessment of consumer behavior to provide simple contracts are financial tools that protect farmers from savings and investment advice, often for free. Such “robo- climate risk by triggering a payout for predefined weather advice” has financial inclusion potential if it can automate events.18 WorldCover is using AI to assess satellite, weather various processes and by doing so lower the costs of serving station, and agronomic data to determine the risk of customers with low-balance accounts. weather “events,” and is working on smart contracts that To date, the deployment of robo-advisors in emerging leverage AI and blockchain to trigger automatic payouts.19 markets is largely limited to Brazil, China, and India, The automatic disbursement of payouts via nonbank where there are significant savings pools. India-based payments providers like M-Pesa will allow farmers without ArthaYantra, for example, aims to circumvent the culture bank accounts to access insurance cover. Identifying and of accepting financial advice from family and friends as addressing barriers to the scalability of insurance solutions well as the commission-based model of existing financial such as weather risk transfer contracts will be critical service brokers, both of which result in suboptimal savings to meaningfully addressing the insurance protection outcomes. Instead, the company’s AI robo-assistant gap in emerging markets, which currently accounts for Arthos analyzes customer data to recommend mutual $160 billion, or 96 percent, of the total global insurance funds matched to each consumer’s risk profile and track protection gap. 20 financial decisions to generate monthly rebalancing options.17 Careful analysis of early attempts to automate Managing the Risks that AI Poses wealth management advice will help determine if robo- Integrating AI into financial services presents sector-specific advice provides better savings and investment outcomes privacy and algorithmic bias challenges. The International for consumers, on average, than human advisors. Unlike Committee on Credit Reporting (ICCR) has identified a chatbots, which are interactive systems conducting a number of risks associated with credit scoring models, conversation via text or audio designed to simulate how a including: data inaccuracies; the use of data without human would behave, Robo advisors are highly specialized informed consumer consent; the potential for bias and bots mostly employed as automated financial advisor and discrimination in the design and decisions of algorithms; investment platforms. The system uses a software algorithm and heightened exposure to cyber risks. 21 These risks are to build and manage portfolios.  enhanced in AI models where data is fed back into systems to refine decision making. More Complex AI Applications Are Under Development Additionally, early adopters of AI in financial services may be able to leverage their head start to generate ever These early examples have illustrated how FSPs are larger data sets on which algorithms can be further trained integrating narrow AI—such as machine learning and refined. An early mover may get so far ahead, and algorithms—into their services to reduce business costs be able to tailor finely priced offerings so much better and overcome operational hurdles in order to serve more than competitors, that it captures an outsized market- customers. Still under development are more complex AI share, resulting in a winner-takes-all scenario. This applications with greater potential to address financial 5 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. would reduce competition for services, with the risk that successfully managing a system using AI. However, there consumers lose choice and price competition in the longer is a question about where those jobs are created, with some term. An alternative scenario is that AI adoption creates EM countries potentially losing out on human capital and new business models that enhance cost-competitiveness knowledge transfers if the jobs are created in company among technological suppliers. Avoiding a winner-take-all headquarters rather than EM subsidiaries. There are also scenario, through efforts by government and regulators some jobs required for financial service delivery in lower- to monitor anti-competitive effects, will be important to income contexts, such as banking agents, that are still largely maintaining consumer benefits of AI in financial services. outside the digital realm and are therefore much further from displacement via technological advancement. As FSPs adopt AI, they need to attract staff with the right skills to understand how AI technologies, like Facilitating Responsible and Sustainable credit-scoring algorithms, work, so that lending is issued deployment of AI responsibly. Otherwise, there is a risk that AI innovations do more harm than good by increasing indebtedness IFC’s digital financial services and fintech practice has for vulnerable consumers and eroding consumer trust invested in and provided advice to over 150 financial in the industry, which in aggregate may increase services providers since 2007. Through its investment systemic risk. Adopting responsible lending and risk and advisory services, IFC has considerable experience management practices like the ICCR will be important in in assessing how new technologies, including AI, can be avoiding overindebtedness for EM consumers. deployed in the financial services sector to help achieve the World Bank Group’s twin goals of ending extreme poverty These risks raised through AI adoption require FSPs to and boosting shared prosperity. For example, IFC client carefully assess and actively govern their operations in Yoma Bank in Myanmar has developed a scoring algorithm terms of data ownership, privacy, security, and biases. This to provide loans to suppliers and distributors, leveraging task will require coordination between FSPs and others— their payment and order data to build a loan book that international organizations, governments, and industry—to funds micro, small, and medium enterprises (MSMEs). develop robust privacy, data management, cyber security, Yoma Bank’s nonperforming loan ratio is well below one and supervisory regulations/processes to facilitate AI percent. adoption across the sector. As early as 2015, through a partnership between IFC, In contexts where the digitalization of financial services—a Ant Financial, and Goldman Sachs, IFC provided $245 prerequisite for AI adoption—still lags, additional efforts million in financing to Ant to launch a data-driven lending are needed by governments and investors to develop the product for women-owned small businesses in China. prerequisite settings. For example, CGAP has identified Although MSMEs account for 90 percent of all Chinese interconnected and open digital platforms, shared market firms and 60 percent of employment, only 30 percent of infrastructure and data, and support for public goods like foundational IDs as structural requirements for digital financial services innovation.22 In addition, government and private sector investors must continue to invest in telecommunications and energy infrastructure to improve the enabling environment for the digital economy. Without this enabling support, there is a risk that digital financial services, whether using AI or not, will continue to be commercially and practically infeasible, leading to a deepening of the digital divide. As with any process automation, the integration of AI into financial services is likely to displace jobs in EM countries. For example, natural language processing could replace outsourced customer care services, which is an industry that employs thousands of workers in countries like Vietnam, South Africa, and Morocco. Alternatively, some jobs will be created in technology companies and large financial institutions to meet the aforementioned governance, FIGURE 2 Example of an AI-Supported Digital regulatory, and maintenance obligations associated with Dialogue Source: Juntos. 6 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. formal banking system loans are disbursed to them. With hurdles, with the result that it is now commercially feasible 560 million people in China connected to the Internet and to extend financial services to more people. However, the small firms increasingly operating online, Ant Financial (a early use of AI by FSPs is still narrow in scope, with many subsidiary of Alibaba Group) saw an opportunity to apply unexplored opportunities to use the technology to enhance machine learning that leverages online transaction data to development impact, such as improving consumers’ assess the creditworthiness of loan applicants, even those financial literacy. In addition, many of the lowest-income without collateral.23 While collateral provides comfort to consumers will still remain out of reach of FSPs where there lenders, relying on it for lending decisions excludes millions is low smart-device penetration and unreliable Internet of small businesses with high potential. Instead, Ant connectivity and energy supply. Financial was able to apply AI to big data to make lending Investors and development finance institutions like IFC assessments based on actual payment history, enhancing can mitigate risks associated with the deployment of AI its competitiveness by bringing high-performing small by adhering to the Guidelines for Responsible Investing businesses into its customer base at a more rapid pace and in DFS, 27 an industry standard developed under the at lower cost, which would be hard for traditional banks to leadership of IFC. It requires investees to be certified replicate. As a result, Ant increased its loan portfolio from by the SMART campaign, which is a set of principles $0.5 billion to $4.0 billion over a four-year period. for responsible financial inclusion, or endorse relevant IFC also helps educate market participants about how to guidelines such as the ICCR guidelines. DFIs should deploy technological innovations responsibly. also monitor and evaluate projects to generate empirical evidence of how AI is contributing to financial inclusion • IFC partnered with the Mastercard Foundation in 2017 in different contexts. This includes understanding if to publish a handbook on how to apply data analytics the application of AI is reducing nonperforming loan to digital financial services, including how practitioners ratios and service costs, improving customer service, can use data to develop algorithm-based credit scoring and resolving KYC and AML risks. Finally, DFIs must models for financial inclusion. 24 continue to invest in the enabling infrastructure for digital • The World Bank Group, through the ICCR, has financial services—including telecommunications and developed guidelines on Credit Scoring Approaches energy infrastructure, and human capital skills—to ensure that include guidance on the use of AI in credit scoring. that the three billion people without access to or effective These guidelines will soon be published. 25 use of digital technologies are not left further behind as the benefits of AI spread elsewhere. 28 • IFC, together with private sector investors, has developed Guidelines for Responsible Investing in Digital ACKNOWLEDGMENTS Finance, which has been endorsed by over 100 investors and financial services providers, including Branch. 26 The The authors would like to thank the following colleagues for adoption of these practices will be important to financial their review and suggestions: Martin Holtmann, Manager, institutions to maintain consumer trust in digital Financial Inclusion, Financial Institutions Group, IFC; Mahesh financial services and to minimize the risk of harmful Uttamchandani, Practice Manager Finance, Competitiveness lending practices. and Innovation Global Practice, The World Bank Group; Davide Strusani, Principal Economist, Telecom, Media, Looking Forward Technology, Venture Capital and Funds, Economics and Private Sector Development, IFC; Matthew Saal, Principal Early applications of AI in the financial services sector are Industry Specialist, FinTech, Financial Institutions Group, IFC; helping to overcome obstacles that impede the extension Shalini Sankaranarayan, Senior Financial Sector Specialist, of financial services to underserved individuals and Finance, Competitiveness and Innovation Global Practice, businesses in emerging markets. These obstacles include The World Bank Group; Georges Houngbonon, Economist, the difficultly some individuals and businesses encounter in Telecom, Media, Technology, Venture Capital and Funds— accessing traditional forms of identification, collateral, or Sector Economics and Development Impact, Economics credit history needed to secure a loan, as well as the high and Private Sector Development, IFC; Baloko Makala, cost to FSPs of meeting their compliance and regulatory Consultant, Thought Leadership, Economics and Private obligations and in managing cyber and fraud risk using Sector Development, IFC; and Thomas Rehermann, Senior existing processes. Instead, AI technologies can analyze Economist, Thought Leadership, Economics and Private new and real-time data sources and further automate Sector Development, IFC. business processes to overcome these operational and cost 7 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. 1 Strusani, Davide and Georges Vivien Houngbonon. 2019. “The Role of Artificial Intelligence in Supporting Development in Emerging Markets.” EM Compass Note 69, IFC, July 2019, pp. 1–2. This definition is also guided by the AI100 Panel at Stanford University, which defined intelligence as “that quality that enables an entity to function appropriately and with foresight in its environment.” See “One Hundred Year Study on Artificial Intelligence (AI100).” 2016. Stanford University. https://ai100.stanford.edu/. 2 Niestadt, Maria, Ariane Debyser, Damiano Scordamaglia, and Marketa Pape. 2019. “Artificial Intelligence in Transport: Current and Future Developments, Opportunities and Challenges.” European Parliamentary Research Service (EPRS), PE 635.609, March 2019. 3 Cambridge Centre for Alternative Finance, World Economic Forum. 2020. “Transforming Paradigms—A Global AI in Financial Services Survey.” January 2020, p. 6. See also Chapter 2 in this report, “The Adoption of AI in Financial Services” pp. 25–36. 4 See its website https://msurvey.co/. 5 Cheung, KC. 2018. “MyBucks Using AI to Offer Loans via Social Media.” M-X Lab 3 January 2018. https://algorithmxlab.com/blog/mybucks-using-ai- offers-loans-via-social-media/. 6 MyBucks. 2018. “Annual Report.” https://downloads.ctfassets.net/9tkv8u9zei1u/1nv0SR36bDiBR5wqpxx4XL/040fc2767d82e8eb20091c699ea88d0d/ MyBucksAnnualReport2018.pdf. 7 Cambridge Centre for Alternative Finance, World Economic Forum. 2020. p. 25. 8 IFC. 2019. “Anti-Money-Laundering (AML) & Countering Financing of Terrorism (CFT) Risk Management in Emerging Market Banks—Good Practice Note.” https://www.ifc.org/wps/wcm/connect/e7e10e94-3cd8-4f4c-b6f8-1e14ea9eff80/45464_IFC_AML_Report. pdf?MOD=AJPERES&CVID=mKKNshy. 9 Barne, Donna and Pirelea, Florina, 2019. “Money Sent Home by Workers Now Largest Source of External Financing in Low-and-Middle-Income Countries (Excluding China),” World Bank Blog, 2 July, 2020. http://blogs.worldbank.org/opendata/money-sent-home-workers-now-largest-source- external-financing-low-and-middle-income. 10 Breslow, Stuart, Mikael Hagstoem, Daniel Mikkelsen and Kate Robu. 2017. “The New Frontier in Anti–Money Laundering,” McKinsey, November 2017. https://www.mckinsey.com/business-functions/risk/our-insights/the-new-frontier-in-anti-money-laundering. 11 Nduati, Hildah. 2018. “Cyber Security in Emerging Financial Markets.” CGAP (the consultative Group to Assist the Poor). https://www.findevgateway. org/sites/default/files/publication_files/cybersecurity_in_emerging_markets_06-30_0.pdf. 12 ADB. 2019. “ADB Briefs: 2019 Trade Finance Gaps, Growth, and Jobs Survey.” https://www.adb.org/sites/default/files/publication/521096/adb-brief- 113-2019-trade-finance-survey.pdf. 13 Megta, Sunita and Sam Permutt. 2019. “Old Processes -New Tools: Goes Technology hold the Solution to More Accessible Financing and Cheaper Cost of Debt for SMEs?” TXF article, 5 August 2019. https://www.txfnews.com/News/Article/6815/Old-processes-new-tools. 14 Santiago, Chito. 2019. “Why the Global Trade Finance Gap is Going to Get Worse.” ESG Forum, the Asset, 9 September 2019. https://esg.theasset.com/ ESG/38618/why-the-global-trade-finance-gap-is-going-to-get-worse. 15 Juniper Research. 2019. “Bank Cost Savings Via Chatbots To Reach $7.3 Billion by 2023, As Automated Customer Experience Evolves.” Juniper Research, February 2019. https://www.juniperresearch.com/press/press-releases/bank-cost-savings-via-chatbots-reach-7-3bn-2023. 16 IBM Watson. 2019. “How a Brazilian Bank Pays Personal Attention to Each of their 65 Million Customers.” https://www.ibm.com/watson/stories/ bradesco/. 17 ArthaYantra, 2019. “ARTHOS: An Advanced Platform, Yet Simple to Use.” https://www.arthayantra.com/how-financial-services-works#our-tech. 18 Hight, Jim. 2019. “Using Risk Transfer to Achieve Climate Change Resilience.” Nephila Climate, June 2019. https://static1.squarespace.com/ static/5a38079c90bcceedee1bba93/t/5d07ceaddac3c80001210057/1560792750823/%40Risk+Transfer+for+Climate+Resilience.pdf. 19 Bird, Jane. 2018. “Smart Insurance Helps Poor Farmers to Cut Risk.” Financial Times 4 December 2018. https://www.ft.com/content/3a8c7746-d886- 11e8-aa22-36538487e3d0. 20 Lloyd’s. 2018. “A World at Risk: Closing the Insurance Gap.” Joint research report by Lloyd’s and CEBR. https://www.lloyds.com/news-and-risk-insight/ risk-reports/library/understanding-risk/a-world-at-risk. 21 ICCR (International Committee on Credit Reporting). 2018. “Use of Alternative Data to Enhance Credit Reporting to Enable Access to Digital Financial Services by Individuals and SMEs operating in the Informal Economy: Guidance Note.” Report published by the Global Partnership on Financial Inclusion 28 June 2018, https://www.gpfi.org/sites/gpfi/files/documents/Use_of_Alternative_Data_to_Enhance_Credit_Reporting_to_Enable_Access_ to_Digital_Financial_Services_ICCR.pdf. 22 Bull, Greta. 2019. “Great Expectations: Fintech and the Poor.” CGAP Blog 29 January 2019. https://www.cgap.org/blog/great-expectations-fintech-and- poor. 23 Using the Internet to Expand Microfinance in China: https://www.ifc.org/wps/wcm/connect/news_ext_content/ifc_external_corporate_site/ news+and+events/news/ifc%2C+ant+financial+use+internet+technologies+to+help+scale-up+microlending+for+chinese+msmes-+women. 24 IFC and the Mastercard Foundation. 2017. “Data Analytics and Digital Financial Services: Handbook.” https://www.ifc.org/ wps/wcm/connect/369c10de-1703-4497-876f-9cdf0367a4d4/IFC+Data+Analytics+and+Digital+Financial+Services+Handbook. pdf?MOD=AJPERES&CVID=lRrkzEd. 25 ICCR (International Committee on Credit Reporting). 2019 (Forthcoming). “Credit Scoring Approaches: Guideline.” 26 Biallas, Margarete, Momina Aijazuddin, and Lory Camba Opem. 2019. “The Case for Responsible Investing in Digital Financial Services.” EM Compass Note, No. 67, IFC. https://www.ifc.org/wps/wcm/connect/a89771c3-b1c3-41fe-a1c5-848cc0d920b9/EMCompass-Note-67-Responsible- Investing-DFS.pdf?MOD=AJPERES. 27 Biallas, Margarete, Momina Aijazuddin and Lory Campa. “The Case for Responsible Investing in Digital Financial Services.” EM Compass Note 67, IFC, April 2019. 28 Pathways to Prosperity Commission. 2018. “Digital Lives: Meaningful Connections for the Next 3 Billion.” https://pathwayscommission.bsg.ox.ac.uk/ sites/default/files/2018-11/digital_lives_report.pdf. 8 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.