How SmartCoin is disbursing loans and credit services utilizing AI and machine learning

Even before the pandemic, the demand for small business loans through official channels have been on an increase, and this trend just got amplified and intensified during the pandemic. Additionally, a number of alternative financing methods from microfinance to digital lending emerged in recent years. They are certainly bringing about a transitory shift with the help of technologies like AI, machine learning, and big data, which are developing a robust digital lending ecosystem to bridge the credit gap for micro-entrepreneurs. Embellishing such a scenario, SmartCoin, a micro-lending NBFC company overcomes the barriers that these small businesses face, in their financing process, by offering personalized loan offers. Such technological solutions play a significant role in enhancing efficiency in the financial lending space, and simultaneously providing an immaculate customer experience. SmartCoin has disbursed more than 300,000 loans to its 100,000+ users worth INR 1000 cr. Mr. Amit Chandel, Co-Founder and CTO, SmartCoin share more insights on the same


  1. Give us a brief overview on how your company leverages AI.

SmartCoin Financials leverages Artificial Intelligence and ML algorithms to extend a variety of small-ticket loans through an immaculate business model that ensures credit access to the various underbanked and underserved sections of society. The breakthrough enterprise utilizes the power of AI for digitizing and streamlining various operations such as Delinquency prediction, Fraud Detection, Income Prediction, SMS/Bank Parsing and categorization, Collection Scorecard, Assisted customer support (chat / onboarding) and Cost Optimizations.

Our systems leverage alternate data from mobile devices and digital footprints for user onboarding and authentication. For example, we use facial recognition algorithms to verify the prospective borrower’s KYC documents. Having served over 95% districts spread across the entire nation, SmartCoin’s capacity to corroborate real-time user data weeds out fraudulent profiles early on at the application level itself.

Take For example – Manjunath, a 27-year-old employee in a small factory needed Rs.6, 000 to pay for his son’s school fees. He could not avail loans from any bank because of his blue-collar work profile and the absence of a credit history.  After a quick online search, he found SmartCoin, a Bangalore-based fintech company, which enables credit access for these underserved lower and middle-income segments through an innovative mobile-first approach within minutes. Manjunath had to simply download and register on the app by filling out a few personal details. That too without needing any collateral, physical visits, long waiting queues or references. A win-win situation, indeed.

This became possible as our proprietary AI/ML models scans over billions of alternate data points spanning transactional and behavioral attributes while transcending traditional sources to anticipate fraud and default risk to predict the best loan terms suitable to the user’s risk profile. Our credit scoring has consistently outperformed standard bureau scores in terms of risk assessment for various cohorts of users, especially those with little or no credit history. Furthermore, our text extraction and language and image processing engines enact a crucial role in building a 360-degree profile of users which is impossible by traditional form-filling approaches. All thanks to our connected design, integrated governance and democratic decision-making.

Our app also takes the user through a gamified credit ladder journey with audio-visual cues and vernacular support, thereby, allowing us to create a spectrum of personalized products. We have re-engineered the entire loan lifecycle to make it 100% automated, digitized and paperless reducing our operating costs drastically vs. banks. This makes all kinds of loan customizations sustainable and the model highly scalable across segments and geographies.  Our In-house Tech Systems consists of Full-stack servicing technologies that have been developed internally to not only ensure swift, seamless experience in availing credit but also support the scaling of our financial services with minimum opex. Having received the license from RBI (India’s central bank) to expand the portfolio of our financial services, we are well positioned to become the first “neo-bank” for Emerging India.


2. Elaborate on some of AI governance methods, techniques, or frameworks used within your organization to ensure that your products/solutions provide the best possible experience to the users with real-life examples or use cases.

Our AI/ML Governance has two guiding principles;

  • Decisions made by AI/ML models should be EXPLAINABLE, TRANSPARENT & FAIR
  • AI systems should be HUMAN-CENTRIC

Furthermore, it focuses on the following five key areas:

  • Holistic and integrated system for data, features, policies and decision making to help teams use the centralized information together and leverage it to build the models, strategies and take decisions independently.
  • Ability to evaluate the constructed strategy, models and decisions so as to compare the performances on a near real time basis.
  • Dynamic review and approval workflow and democratic decision-making under an integrated staging and a production environment.
  • Validation and monitoring the results – Near real time monitoring of actual performances of models, strategies and decisions with defined metrics for each development.
  • Auditability – We have a built-in traceability framework and real time automated alerts for full audit of the systems and exceptions.

When extending credit across the credit lending system, the two most important questions that need to be addressed beforehand are the prioritization/scoring/ranking of users and allocation/capacity. Both of these are addressed majorly based on financial and behavioral characteristics of users. And similar information gets utilized in answering both. Our integrated and centralized systems help users to access a single source of information to avoid discrepancies and a connected machine learning platform that creates independent models for the same. We leverage AI/ML to get patterns within the data and provide prioritization scores as well as capacity scores.  Furthermore, we have established benchmarking KPIs to evaluate the models post each iteration. Post review of such models we follow our approval framework and push these models to the staging environment and finally deploy it to production. Post any deployment we are continuously monitoring the relative performance of the models, score distributions, capacity distributions, rate of rejections, reason of rejections and many more such relevant KPIs on a real time basis.  In case a user is denied credit we have logs stored for every decision step for audit purposes. Following one such use case we have been able to provide appropriate credit denial reasons to our users. As prioritization is linked to pricing, our users are getting the most optimal pricing and our customer retention rate is the validation. Given the completely integrated and automated framework, we have been able to timely update our models based on changes in market conditions, economy strain or during the pandemic. Our designs help us in implementing new policy, decision model and framework changes very efficiently and offer almost the best TAT in the fintech industry.


3. Why are tech leaders talking about AI ethics, responsibility, and fairness in the present era? Why is it the need of the hour? How can it be ensured at scale?

 AI actors should respect the rule of law, human rights and democratic values, throughout the AI system lifecycle. These include freedom, dignity and autonomy, privacy and data protection, non-discrimination and equality, diversity, fairness, social justice. And it is fundamental to warrant that any decision-making that involves or depends on AI is not privy to these factors.

In the recent past there have been several instances where violation of such basic rights were reported. Take for instance, a robotic soap dispenser that failed to distinguish dark-skinned hands. It only gave soap to persons with pale skin. In another example, AI used in American courts have denied bail to people just by looking at their pictures which was often biased by color.

Even though Artificial intelligence can create its own programs, it follows a certain framework ordained by humans and uses data sets collated by humans. If there is any possibility that the data is compromised or sabotaged, the machines will obviously perpetrate bias and prejudice. Therefore, it’s the need of the hour to ensure AI ethics, responsibility, and fairness across AI-based frameworks.

We deploy every necessary measure within our AI frameworks to implement AI ethics and achieve fairness where Biases are eradicated by looking at the distribution across different variables and real time monitoring of such influences. It is a tech driven integrated system to scale over billions of users and to explain the decisions. We honor transparency and accountability by empowering the users with all the requisite information even in cases where credit is denied with proper reasoning over appropriate communication channels. Moreover, we have built-in traceability and lineage that allows for full audit at each decision step for the users.


4. How are you ensuring that your data science and AI/ML teams are aligned with the company’s AI governance policies, or best practices? Give an example.

 The team conducts specific data checks and assesses the data to be used by the model in relation to consistency, completeness, transparency and data accountability. These include requirements covering the use of internal data over external data where possible, data ownership and testing of critical data elements. There are defined metrics for data quality checks and alert metrics regarding the same.

Model developer performs several statistical checks to ensure that the data samples are unbiased (e g. Information values, Correlations, Binning, Trend); and performs validations (e.g. Out of time validation) on the output of the model to rule out any bias in the results. Post this there is model evaluation and benchmarking w.r.t performance of the model prior to and during its actual implementation.  After the model validation there is a manual review with senior management before deployment to ensure adherence to the governance.

There is close monitoring of the models during and post its deployment to ensure that it worked within the pre-defined parameters. There is a process to track the performance of the models on a periodic basis to confirm the relevance and adequacy of the models.


5. Research has identified nearly 200 biases that influence human decision making. How do you avoid those biases from being introduced into your AI algorithms?

 As explained earlier, biases are eradicated by looking at the distribution across different variables and real time monitoring of such influences. We do not feed the characteristics and variables that define the identity of a user to our algorithms in any form.


6. Do you have a due diligence process in place to ensure that data is collected ethically, especially while using third party plug and play data sets or models?

 We have a comprehensive data collation process where due diligence is ensured at every step. We take explicit consent from users with contextual details about how their data will be used. We ensure that every bit of data is encrypted while in transit and while at rest. We make sure that CISA and ISO-27001 guidelines are followed to the last letter for confirming data privacy and security. We also have an End-to-End tracking of permissible data uses in our teams and a secured VPN tunnel with standardized approval process to ensure minimal access privileges within the organization.


7. How do you systematically feed ethical principles related to AI and AI applications into your platform?

 While we strive towards innovating and scaling our ML platform for the next billion users, we continuously ensure that the platform is following all the ethical principles (inclusive growth, privacy, transparency and fairness) in a 100% digital format where no human decision making is involved. Biases are eliminated and fairness is ensured by analyzing the distributions across different variables/data points independent of the identity of a user. Prolonged continuous experimentation over various micro-segments to achieve inclusive growth. We are completely transparent towards our users with the data collection process and with all the requisite information even in the cases where credit is denied with proper reasoning. We have built in traceability and lineage that allows for full audit at each decision node for our users.


8. How does your company ensure the protection of consumer data privacy?

 We ensure that every bit of data is encrypted while in transit and while at rest. We make sure that CISA and ISO-27001 guidelines are followed for confirming data privacy and security. We also have an End-to-End tracking of permissible data uses in our teams and a secured VPN tunnel with standardized approval process to ensure minimal access privileges within the organization.


9. What are your efforts in helping brands foster a trusted, transparent relationship with consumers?

 We believe the key to gain user trust and confidence is in providing top-notch customer experience, affordable and optimal pricing, being always reachable to them, accepting their feedback, presenting reviews and testimonials of existing customers as a proof and moreover honesty of our representatives. As mentioned earlier we have a gamified ladder in our app to enrich customer experience, AI driven model for optimal pricing and loan amounts, constant customer care support and dedicated channels to collect their feedback and testimonials. We would emphasize that transparency is in communicating with them effectively. We have multiple appropriate channels to communicate with them about credit denial, approval, outstanding amounts, due dates, and new product features etc.


10. Did you come across any biases, or ethical concerns/issues lately within your organization/industry/product? If yes, how did you address them?

During the initial phases of model development, some models would prioritize the customers paying before due-dates relative to the ones which paid on time. We addressed this issue by adding relevant variables with appropriate weights to tune our models. Our AI governance framework has helped us in detecting the issues in the sandbox environment and resolving them before it could add this bias in the live environment.

Leave a Response