Assessing risk is an integral part of how financial institutions operate. Daily, much money saved lent out and invested. Hence, they need to accurately predict how their credit decisions will pan out. A necessity if they want to remain sustainable. But this isn’t an easy process. Assessing risk is no easy task in any economy. The current pandemic has only made it harder. In the wake of spiking job losses, how can banks minimise the risk of default when giving out loans? By leveraging the power of big data analytics and artificial intelligence, Sri Lanka-based startup Algoredge aims to tackle this issue.
The startup is an alumnus of the HatchX virtual fintech accelerator. The programme was the first of its kind in Sri Lanka run by Hatch in partnership with the Lankan Angel Network and funding from the Ford Foundation. The programme was focused towards connecting local fintech startups with resources, mentorship and business support to grow and expand across the South Asia region. Over the course of the programme, Algoredge pivoted to adapt to the circumstances of the pandemic.
Piecing together the puzzle
“This isn’t just a problem in the current context,” says Urmila Chandrasekaram, Co-Founder of Algoredge, a firm that provides data analytics services to financial institutions. “Even earlier, banks either lend money to someone who doesn’t honour their financial obligations or refuse credit to a creditworthy applicant, because of insufficient historical data available about that person.”
Pointing out that lenders have traditionally looked to customers to provide risk assurances, such as guarantors, credit histories and similar information, Urmila noted that in today’s connected world, vast amounts of data is generated everyday. This offers firms far more data points they can tap into. Yet, they don’t typically capture such information. This is where Algoredge fits into the picture. As a firm that uses machine learning and data analytics to assess and minimise credit risk and fraud, it uses the full range of data available to make more accurate predictions of customer behaviour.
“We go beyond the information banks usually collect, and use other relevant information available about the customer, to find a correlation between these data points so we can assess the creditworthiness of the individual” noted Urmila. But what are these additional data points Algoredge taps into? Urmila’s fellow Co-Founder, Kukaraj Tharmasegaram, noted this could include consumers’ psychometric data, their digital footprint, and other relevant unstructured data available.
“It depends on the demography the financial institution caters to; it could be a traditional bank, a mobile wallet, or a peer-to-peer lending system. We look at what type of data will help them. We specialise in bringing all this data into one place and understanding how all of it fits together and makes sense through data analysis,” Kukaraj explained. “At its core, Algoredge is a problem-solver. Firms come to us with a problem – we help them solve it and grow their revenue.”
Faster, smarter, safer
The technology-assisted system leads to better credit decisions, while also making the process a lot faster, stated Kukaraj. “Right now, when a small business or an individual approaches the bank for a loan, it can take a while for the loan to be approved. There are manual processes involved and things are not unified. At Algoredge, when we start working with a financial institution, we bring all the relevant information together into what is called a Data Lake. Once this is done, any solution we build on top of it can bring instant results – loans can even be approved on the spot.”
Something that financial firms should take note of when implementing credit-scoring models with non-traditional data points, as Kukaraj pointed out, is whether these models are accurate to their environment and the segment of the market which the company might be serving. In order to maximize the potential of non-traditional data points, companies need to build their own custom models utilizing the data form the market they have been serving. This is why Algoredge instead focuses on building a custom Machine Learning model, based on MAML (Mind Augmented Machine Learning), for its clients. Thereby ensuring each solution is tailored to fit unique conditions that may apply.
How to build a unique machine learning model
The process starts by identifying the exact business problem that the company wants to solve. Explaining the process, Kukaraj said, “We start with the business problem because at the end of the day any cutting edge solution you provide if it doesn’t help an organization’s bottom line, the solution is useless. So we start the journey with any organization by first understanding their most pressing business problem which they want to solve. Then with our expertise, we change the business problem into technical requirements, this is one of the most crucial steps that determine the overall success of the project.”
Following this, the existing data in the organization is analyzed to understand underlying patterns and correlations. If required Algoredge also provides the necessary plugins to start collecting additional data, which the company might not have been collecting thus far that can be used in the future for improved business outcomes. A model is then built to run through this information and tested with past data to identify the optimal model. This model is then deployed either on-premise or in the cloud-based on the company requirement. This also includes the necessary infrastructure such as a data lake or data warehouse.
A crucial step in building any machine learning model is choosing the right data. This is what will ultimately decide whether such a system succeeds or fails. In the case of financial institutions, a faulty dataset could result in the wrong customers being approved for credit and the right ones being rejected. To guarantee its accuracy, the model is then put into use through a phased approach. Working in parallel with the client’s existing credit assessment method for an extended period. This allows for further fine-tuning of the system through continuous performance monitoring, making sure the system is mature enough to make accurate predictions.
Kukaraj says “With a platform like this, you also need data in the correct form. Although there are such platforms abroad, it’s not as simple as doing the same in a developing country like Sri Lanka. Here the data needs to be cleaned up into a usable form and brought to one place. We had to develop a lot of tools and expertise for just that stage of the project. If you want to analyse data, the company has to first gather it in the right form, but in Sri Lanka, we’ve had to help firms bridge that gap and digitise that data properly before we could move forward.”
Algoredge’s systems aren’t only capable of credit assessment – they can also be used to detect credit fraud. However, as the methods of fraud vary by region, and are evolving with technology, they are most effective when given specific types of fraud or malpractice to be on the lookout for. Kukaraj recounted how one of their clients had identified a type of fraud that was unique to Sri Lanka, and through the custom-made model Algoredge developed were able to prevent it from occurring. This would not have been possible with a pre-built system – it was only because they were looking to solve this particular problem that the solution worked so effectively.
From idea to execution through HatchX
The entire system of tailoring solutions to meet specific problems is actually how Algoredge itself was born. Urmila recalls how, in 2019, the duo wanted to start a ‘buy now, pay later’ online payment platform, so online shoppers could split their payments into interest-free instalments. This idea materialized in the form of mintpay.lk, Sri Lanka’s first ‘buy now, pay later’ payment facility.
“But this kind of business model comes with a fairly large amount of risk. To mitigate this, you need to develop an algorithm that assesses the creditworthiness of an individual by harvesting their digital footprint. So there was a credit decisioning tool that we built earlier,” she recounts. “But then early this year, COVID hit, and we understood that this system could be put to use for banks; the process could be tweaked and enlarged to assess the creditworthiness of potential customers.”
Switching tactics on an operation like theirs would not have been easy, but Urmila shares that they had the right support through the HatchX programme. “It came at a time when the world went into a lockdown. Despite the difficult situation we were facing, they managed to get all the relevant stakeholders, like the Central Bank, the other banks, and similar contacts, which is what a startup needs. I wouldn’t know how to gain visibility with such institutions; basically, HatchX is what opened the doors for us,” she says. Kukaraj agrees, saying, “It brought all the necessary contacts needed for a FinTech into one place. It’s difficult to create the level of credibility with institutions like banks during the initial phase, but being a part of HatchX really helped to position ourselves in front of these”
Batting on a larger pitch
Currently Algoredge’s founders shared that they are in talks to implement their system with several commercial banks. All of whom are at varying stages of their digitisation processes. However, things are changing – while the rise of more tech firms in Sri Lanka has seen its digital ecosystem grow healthily, the pandemic also pushed both companies and consumers to adapt to online transactions, effectively forcing them to digitise activities.
On the surface, the major benefit this brings is efficiency in financial institutions and opens doors to new solutions like Mintpay. Data analysis systems can cut out hassle on both sides of the fence. Consumers will avoid the tedious rush to gather and submit a series of documents to prove their creditworthiness. On the other side financial institutions can save time on assessing applicants to gauge the likelihood of non-payment or fraud.
We can see this efficiency bring cascading benefits to many more stakeholders further down the line. For instance, entrepreneurs and legitimate businesses will find it easier to apply for credit, while import/export and supply-related transactions will be processed more smoothly; and we can only imagine how much could be streamlined if these systems are implemented at Government-level. Ultimately, this is what the ambitious duo at Algoredge aims to achieve as the company grows in scale: strengthen the country’s financial systems, empower the Sri Lankan economy, and help every citizen reap the benefits.