Digital Lending – The Strategy Game

A study by Goldman Sachs revealed that “Financial institutions stand to earn $30 billion in the next eight years (driven by a $600 billion loan market) by utilizing digital lending technology “.  A Credit Suisse report further stands to forecast the growth of this ‘600 billion worth loan market’ to grow up to $3 trillion in the next 10 years in a report published as of July 2016.

While the numbers may distract, the essence lies in the phrase ‘utilizing digital lending technology’. Is digital really the secret sauce to the pie?

And what is digital lending technology, anyway? Are we talking products, platforms or solutions a bank can go with when it comes to digital l ending?  How does it really differentiate?

To quickly box the definition of Digital Lending, it is characterized by a focus on making cheaper loans available through completely digital means but it does not end there , digital lending also focuses on automating the entire processing of this low cost asset. Typically, characterized by the following :

  • Newer Segments – Targets a completely new set of borrowers (typically small loans for the traditionally undeserved)
  • New Data Sources – Identification of new markets through new data points social media clues , behavioral analytics , data analytics
  • Intelligent Algorithms – Higher reliance on data technology and complex algorithms on aggregated data
  • Automation Technology Solid reliance on automation technology in the automated Underwriting Process with a no rip/replace method
  • Continuous Digital Dialogue The lender , borrower and guarantor/aggregator are in continuous sync through real-time notifications and event based alerts making it seamless

While banks offer traditional loan products, the market is strife with new and aggressive fintech players who are upping the game by introducing new business models.When banks explore ‘digital’ to boost the lending business, what is it that really that they seek ?

A New Sport , To Stay in the Game or Play the Champion’s Game ?

While the use of technology can help the bank modernize, the success and proof of value can be realized only by the forces that drive.  It is important to ascertain the end goals of digitization – is it an experimental and exploratory decision for the business or a definitive conscious one aligning to the larger business goals which could be anything from creating a new space in the market, achieving scale or building better customer stickiness.

Choosing the Right Fit or Fitting In

Alternative lending models such as P2P lending and B2B marketplace lending make cheap loans available to an audience that is either cash strapped or traditionally deemed credit-invisible. The fintechs may have uncovered a new segment and created a new ecosystem but banks are still key for the entire lending cycle to thrive and grow.

The P2P lending is the most popular alternative lending model that is attributed to have democratized lending. Individual borrowers take unsecured loans from other individuals who have surplus cash and are willing to lend for return on value.  Similar to P2P , B2B lending enables and empowers SMEs to receive necessary funds from a set of online direct individual and institutional investor. However, the bank still plays a key role in the entire chain.

Another model is Crowd Funding which is typically used to raised funds for a common interest and against reward or equity. Individuals and businesses have used this to raise early-stage investment, presell products, obtain market validation as well as to crowdsource ideas, engage customers, secure partnerships and build loyal communities.

On the other hand , equity based crowdfunding enables entrepreneurs/startups to raise early-stage capital in an online marketplace directly from individual investors, angel investors and VCs in barter for equity in the company. The bank relegates into the oblivion in such a model.

Collaborate and Compete

It is important that the battles are chosen wisely the in digital lending . With emerging models , competition from disruptors the bank needs to strategize on its feet. Here are some options to choose from basis the competency of the bank in terms of current integration capability , investment at hand and time to market.

digital lending strategy for banks
Banks can opt for any of these strategic options to achieve scale in the Digital Lending

Digital Build or Digital Buy? Unbundle and Latch On!

Typically, technology decision makers often take calls in terms of whether the solution should be completely developed in-house or ready off-the-shelf solutions be used. However, in the age of digitization, this no longer remains a build-buy-partner decision restricted to tech anymore. It is increasingly becoming a business call.

Today, there are ecosystems in place which bring the borrower, lender and guarantor together into a single marketplace. With no investment for lead generation or customer acquisition, or even creating a differential customer experience, the FI can simply focus on creating the best product offering on such platforms. Latching on to the ecosystem can be the simplest digital lending strategy to be adopted. This however, is not the prescribed leader’s game. It can only be one of the levers.

The most effective approach for a FI that looks at latching on to multiple lending ecosystems as an option and is open to offer products through diverse digital channel partners and aggregators is that of the ‘Unbundling’. Looking at each business process in isolation to create the most effective stack of APIs available to the marketplace is an approach that most banks are aggressively adopting.

Product , Platform or Technology

Well , digital lending is a combination of all of the above . Attempting to jot down a few of the key technology investments that are required in the cycle

The Adoption of digital technologies in the various stages can be visualized as below (Degree of Adoption Represented as Green – High , Blue – Moderate , Grey – Low)

digital technology adoption in lending
Adoption of Digital Technolgies in the Lending Landscape

End Note : Digital Lending is going to be the future as it operates with an intelligent understanding of customer’s need for funds and reaches out to ensure pro-active lending thus making loans quick and accessible and offers a bouquet of cheap and practical products. The right strategy at the right time is the key differentiator – biz models and technologies are only levers.

The Hidden Treasures of Semantic Fingerprinting

The futurists of the world are besotted by the  science that simulates the engagement , decision and the discovery process of the human brain  -in other words Artificial Intelligence(AI) or Cognitive Sciences. Cognitive systems quickly identify patterns similar to a way our brain identifies situations and responds to it. Within AI , there are a number of different technologies and they lie at different points in the curve of adoption with virtual assistants and robo-advisors riding the crest of adoption. This post talks about the merits of the lesser known yet immensely intriguing Semantic Technology  .

Semantic Technology , as Forrester predicts is still a good  5 to 10 years away from mainstream adoption but some companies such as the Vienna based Cortical IO and  Cambridge Semantics  are few of those who are braving the first wave.

Pic Source : Forbes

Semantic Maps – Vocabulary Building Tools

Semantic Maps or Networks have been around for more than a decade where words are understood under the context they are used , very similar to how the human brain understands and processes words, phrases or a language. This technique is used as a teaching method for children learning a language  and has been seen as a very powerful vocabulary building tool.

The same terminology takes on a far more complex computing context .Semantic Mapping can be used for dimensionality reduction of a set of multi-dimensional vectors to retain main data characteristics. The original properties are clustered to generate an extracted feature. This technology has typically been explored for text mining and information retrieval.


Semantic Fingerprinting

Semantic fingerprinting is a new method whose manner of processing text is modeled on that of the human neocortex . Semantic fingerprinting has the potential to be more powerful in document comparisons than are word list-based analyses. The leading proponent of semantic fingerprinting is Cortical ( technique to identify similarities , identify context and also arrive conclusions.

Since semantics are heavily dependent on context , the fingerprint for ‘Apple’ would have strong connections with the computer brand.    Words, Sentences and whole texts amounting to terabytes can be compared against each other.

Source : Cortical.IO

Big Data + Semantics = Endless Possibilities

Big Data Semantics is where the technology is applied to reduce the stream of unstructured data to understand , predict , categorize or sort just as much as one would with any form of data once reduced to an understandable , identifiable form.

The system ingests data in any unstructured form – emails , faxes , documents , sms , social media or data from internal systems and then runs it through a semantic engine.   The semantic engine generates smart binary vectors with minimal memory footprint represented in boolean. This helps to compare aggregated or atomic representations of words .

Some ways in which it is being leveraged are seen in the table below

Semantic Table

With the rise of  AI startups receiving funding across the globe ,  disruptive technology and their use cases are only to rise . We can only wait to watch the treasures it unlocks.

If you have an interesting use case you would like to share , please leave a comment or get in touch with us, 

Is India ready to be the AI Capital ?

In a country of 1.32 billion people where 48.26 million are unemployed , any kind of automation till recently , was unwelcome .With the government , taking a strong stance on digital technologies , a floodgate of investments seem to have opened up.  AI is the latest muse for investors , large corporates and fintechs alike. A few weeks ago a list of the top 100  AI Startups was compiled and published by the very credible CB Insights and to our utmost dismay  , not a single Indian startup figured in the list.

In a very recent research by Quartz ,  the talent crunch in India’s AI circuit was revealed as extreme. Only 4% of the talent pool has actually worked on deep learning or neural networks. Belong Research quotes that only 17 % of AI talent works around Financial Services and the demand to supply ratio for AI talent has a severe shortfall as seen below. 


This week spelled some good news for a lot of companies toying with a roadmap around AI , startups as well as the developer community  as two very big corporates announced investments to boost AI in India.

Intel bets on India for AI 

Intel announced heavy investment to groom local AI talent and Intel South Asia Managing Director Praksh Mallya said  “Our developer education program will educate 15,000 scientists, developers, analysts and engineers on AI technologies, including Deep Learning and Machine Learning in India.Our collaboration with the industry and the academia will help democratize AI, by reducing entry barriers for developers, data scientists and students,”

Through 60 programs in a year, the initiatives will empower the community with the know-how for AI adoption with ready-to-deploy platforms and tools for solution development. “As India undergoes digital transformation, the data centre and the intelligence behind the data collected will enable the government and industry to make quick decisions based on algorithms,” said Mallya on the margins of Intel’s ‘AI Day 2017′. The company’s Indian subsidiary is collaborating with Hewlett Packard Enterprise, Wipro, Julia Computing and Calligo Technologies for using AI in the country.

EY announces its first AI Center in India

Ernst and Young announced its plans to enhance its suite of automation and artificial intelligence offerings with the opening of its first Artificial Intelligence (AI) Center in India. Artificial intelligence is already being deployed across industries such as automotive, telecom and technology . EY’s AI Center will bring together teams of multi-disciplinary practitioners, combining expertise in AI, Robotics etc. along with domain experience in sectors.

“The launch of the AI Center,aims to lead the next step of this transformation journey by helping enterprises combine AI’s autonomous reasoning with systemized learning opportunities.” said Milan Sheth, Partner – Advisory Services and Technology Sector Leader, EY India.

Intel’s  developer education program will educate 15,000 scientists, developers, analysts and engineers on AI technologies, including Deep Learning and Machine Learning in India.

The Curious Case of the Camouflaged Rule Engines

Banks are under tremendous pressure to create tangible digital transformation use cases that aim to have an overnight impact on its brand perception or customer experience. There is also a fair bit of smart camouflage happening to the bank’s existing systems to ensure investment protection while adopting disruptive technology and staying in the game . The Rules Engine is the current muse of many an architect giving a smart makeover to the digital landscape of the bank. In its new avatar , it powers intent classification , interprets signals from smart channels and even acts as a reasoning engine while doing its classical role of decisioning , computation and comparison.

Most ‘smart’ banks aiming to ride the digital transformation wave are incing towards their 100 day goals of AI Adoption .Deliberations while evaluating strategic vs tactical goals and trade-offs between cost and time-to-market of AI enabled systems zero in on the rule-engine as a powerhouse of pre-built logic. If only the rule engine scales up to become cognitive adaptive , a number of interfacing systems may be able to draw from its intelligence. However , here is a look at how the rule engine and the smart systems are sharing space in the evolving digital space.

Intent Identification for the Workflow Bot

While the user interacts through his channels of choice, an integrator service such as the one embedded in the Microsoft Bot Framework, accepts the message and routes it towards the bank’s bot . The Bot then registers the request , passes on the request to its Chat Interpreter or the NLP engine eg LUIS. The NLP Engine can perform a preliminary intent identification based on heuristics .The Rule Engine accepts this as an input and traverses through predefined interaction patterns and zeroes in on the most accurate intent by applying an additional set of rules routing it to the most appropriate workflow or point of entry(state) in the identified workflow.

Decisioning for the Robo Advisor

Going by a similar interaction model as mentioned above, advisory services use the rule engine’s decisioning abilities to generate responses which include the following

· Calculation based on parameters (such as eligibility checks)

· Comparison across generated inputs (recommend credit cards or local offers)

· Response Selection (to advise , insist or warn — selection of tone , selection of products , selection of actions)

Interpretation for the ChatBot

Not always but sometimes the interpretation can be powered by a rule engine as well. If helps train your bot with stories that are of the nature ‘If A says this .. then B says this ‘ it can be abstracted to create if-else rules of the nature ‘If A .. then B ..’ .The rule engine cannot take on the task of training the bot but can ease decisioning or accelerate basis the nature or complexity of statement.

The rule engine’s forward and backward chaining capabilities help it to act like an inference engine as well helping build responses based on past interactions

Message Transformation for the Smart Peripherals

The future of Fintech and IoT is interlinked. From paper to computer to phone to watch — the instruction to the bank is now probably going to be at the tap of a coffee mug. With different devices and no solid standardization rules yet , the rule engine steps in leveraging its embedded library of transformation rules to enable any to any conversion messages.

It will be interesting to see how the rule engine evolves over time . Will it be merely camouflaged to power smart systems or will it emerge as the cerebrum of the AI ecosystem that helps reuse interaction patterns , evolves and self learns to leverage across product processors , smart devices and bots. In other words , will the adaptive cognitive rule engine be a contender for future investments in the AI strategic roadmap . Lets , wait and watch !

The Year of the Banker Bots

With digital banking taking new age dimensions and moving beyond the standard automation and user experience , there is a notable shift in the rise of investments in intelligent systems across the globe . 2017 is predicted to be the real game-changer where the bots are going to greet , guide and operate the bank.

Geared up for Botsification

With 11,000 bots live on Facebook messenger alone and 23,000 developers having signed up for building bots this year , get ready to chat with your service providers.As per Gartner 38% of consumers have already used a virtual assistant for services on their mobile phone in 2016  . The bots in fintech are typically either goal-based dialog agents or chatbots. This is the year of finbots

Over the last 2 years the Bot builders have reached a certain state of momentum if not maturity and therefore today ‘botsification’ of processes is no longer sounding far-fetched or experimental. Google , Facebook , Microsoft , Apple and every company that matters has invested heavily in bots and now have bot building frameworks ready for mainstream adoption.

Is Banking Ready for ‘Botsification’?

Banking as an industry is not a leader in the adoption of chatbots but there is a conscious decisioning if not adoption process underway in most banks with a careful and definitive digital strategy.  Healthcare in the US , has seen the maximum advancement in mainstream operations , with bots playing a major role in the transcription process. Most of these conversational interactions with a bot are now HIPAA compliant with mainstream service providers stepping up on the compliance front. Banking is yet to enforce Regtech in to the ‘botsification’ arena and is toying with dialog agents and customer service chatbots as its step one into the world of adoption of AI .  However , the use cases or area of engagement in most banks is clear and evolving.

Banks that have Invested Already

Santander UK introduced the Santander Smart Bank app where customers could speak to their mobile phone and ask queries related to their accounts . Future versions promised to have voice enabled payments and reporting of lost or stolen cards. Very ambitious! However, I have my reservations on bot enabled payments because artificial intelligence as a technology relies greatly on historic data and pattern recognition. Unless there is security to match up this may be a step to premature.

In April last year , Swedbank adopted Nuance’s Nina as its intelligent virtual assistant based on NLU technology. Nina offers a conversational experience to Swedbank’s mobile customers . Nina offers its chatbot primarily as a healthcare transcription engine . We are yet to measure the nature of success Nina can offer Swedbank .

Erica , Bank of America’s chatbot due for launch in 2017 is a shift away from the conversational bots where the conversation is initiated by the consumer . Instead , Erica can drive conversations . BoA promises ‘Questions , Answers, Insights’. Insights is what is intriguing the community. Erica is going to be able to guide a user on spending habits and also track credit score. Experts such as Chris Gledhill think Erica might be the ‘coolest’ entrant if industry rumors are to be believed.

Closer to home , HDFC launched it OnChat platform through’s  chatbot integration last week. Niki is a great platform to start building your services but its NLU capabilities cannot yet be commented on.Measurable goals for such initiatives in terms of business and engagement are yet to be ascertained as experimental or mainstream.

The DBS Digibank’s Mykai powered by Kasisto is perhaps one of the most conversationally mature bots to have entered the arena this year. Not surprising as Kasisto is a spin-off from SRI International that powers Apple’s SIRI.  Mykai is currently available on app stores in the US.

Bank / FI BOT Tech  Vendor Phase
Santander UK SmartBank Nuance and built in assistane with Santander Universities Live
Swedbank NINA Nuance Live
Bank of America Erica Homegrown To go live in 2017
HDFC OnChat Niki.AI Live
DBS , Digibank MyKai/Kai Kasisito spun off SRI International(makers of Apple’s SIRI) Live
RBS Luvo IBM watson Was to launch in December with 10% of Scotland customers
Axis Bank Not known Active Intelligence PTE


Announced its award of contract to the tech vendor
ABSA Not known Not known Announced intent
Bank of Tokyo Mitsubishi Nao. This  is not a chatbot. It is a  pint sized interactive robot Aldebaran Robotic (Softbank Robotics) Early 2015


Mizuho Pepper. Aldebaran Robotic (Softbank Robotics)
Yes Bank Payjo To Launch in 2017
RBL Payjo To Launch in 2017

The Bots will Grow Up Over Time

It is going to be interesting to see how the bots grow up over time. The investment in these bots is also with the intent that artificial intelligence is a continuously learning technology. The more you talk to your bot, the more intelligent it gets . Well , the historic embarrassment that TAY caused Microsoft merely 16 hours post its twitter launch will ensure banks invest in technology to monitor and control bot activity . While budgeting for the investment in bots , it is equally important to invest in companies who are working on technology that empowers and secures the bot.

Generative , Hybrid or Heuristic ?

While the tech providers today are using primarily the ‘Retrieval Based Model’ , it is foreseen the Chatbots build on the ‘Generative Models’ will survive the test of time.

A Generative Model takes as input the user message and the previous message to generate responses. It urges you to talk more to it and divulge more information , before it responds.

On the other hand  , the Retrieval Based Model relies primarily on a database of responses – it assimiliate context , user message and retrieves the closes pre-defined response through syntactical algorithms .Context may include a position in the dialog tree. In case it does not use context , it will respond only basis the last message ensuring stateless retrieval.

Alternately , there is also the Pattern Based Heuristic Model which engineering responses . The chatbot traverses through multiple patterns prior to responding.

Finally , a hybrid model that uses a combination of generative for intent classification and retrieval based for entity categorization , pass them through a heuristic model (or rule engine ) for response

Usability drives Adoption

The ultimate test which is goes beyond the Turing Test is that of a user’s acceptance of the offered conversational experience. Will it be dismissed as an annoying addition to the mobile phone or a trusted personal assistant , only time and technology will tell.