fraud detection

Fintech companies are using AI and ML models to help with fraud detection (Photo by Punit PARANJPE / AFP)

StashFin is improving fraud detection in India with machine learning and AI

Fraud detection is something the financial services industry can’t afford to take lightly. From retail to consumer banking, to fintech and insurance services, fraud detection is often the first step in protecting their customers.

Today, advancements in artificial intelligence and machine learning have enabled fraud detection to be automated, allowing industries to not only reduce the number of cases but also detect fraud before it happens.

In India, financial services are poised for explosive growth, and digital lending is estimated to be a US$1 trillion opportunity over the next 5 years. However, fraud in India is still a big problem, with the Reserve Bank of India reporting 7,400 bank fraud cases in the financial year, a slight drop compared to the previous year.

At the same time, the South Asian nation is also witnessing growth in fintech. Digital lending, whereby loans are applied, disbursed, and managed through digital channels is seeing significant opportunities. India has over 2,000 fintech companies with the market projected to grow to US$84 billion by 2025.

Fintech and fraud detection 

StashFin, an Indian fintech company focusing mostly on digital lending, is disrupting traditional lending in India. The team builds their own applications to manage all transactions from over a million customers.

Parikshit Chitalkar, StashFin’s Co-Founder (IMG/StashFin)

According to co-founder Parikshit Chitalkar, there are a few lending model options for customers today in India — each with its own fraud detection techniques. One of them is lead advocators, where customers go onto websites looking for credit, and the sites will send them to the most appropriate lender based on a selection matching process.

Next, there are Non-Banking Financial Companies (NBFCs) that offer credit based on the applicant’s capital and balance sheets. The third model is a hybrid model where companies like StashFin have their own NBFC facilities and partner with other NFBCs.

“We are unique because we are the only platform in the country that can serve the customer throughout their entire lifecycle.

For example, a fresh grad out of college working as a barista will have a new profile and be able to get funding based on their capabilities. As the fresh grad’s career progresses, their credit score will be improved, allowing them to have access to more lending options.

“We are the only platform in the country that actively reduces borrowers’ rates as their credit profile improves,” said Chitalkar.

And this is where machine learning (ML) and artificial intelligence (AI) come into play. Chitalkar explained that a lot of people feel ML and AI tools are only used for risk management. However, the same tools can also be used for applications on the operating side of the business as well. 

AI and ML in fraud detection 

When it comes to fraud detection and fraud management, AI and ML allow computer vision, image matching, facial detection, and more — StashFin builds its models around fraud detection. For example, in facial recognition, when someone takes a selfie for an application, the histogram algorithms can detect if the person is real or not as synthetic fraud is high.

Once a customer goes through these filters, several ML filters process the data and figure out where the customer lands in the risk level. From there, the customer’s score will be produced and compared with other data in the registry as well.


“Apart from fraud detection, we also have several AI and ML models built for risk positioning. On the operating side, we use AI and ML for loan processing. We receive about 150 to 200 loans per hour — which is humanly impossible to process. But this is all done via ML and AI models. For payment collection, we don’t send any reminders. We use ML models to scan the customer’s payment habits and remind them based on that, instead of calling them every other day,” explained Chitalkar.

While StashFin’s models detect about 45 to 50 fraudulent activities a day, Chitalkar mentioned that it’s still a cat and mouse game, especially with false positives being a constant issue. Whenever deploying a new model, there are strong chances of false positives being detected.

For fintech companies like StashFin, there is always a risk involved when it comes to lending and rejecting a customer. Hence, they will run fraud detection models and look at some rejected cases manually to ensure the false positive ratio is below an acceptable level.

“Wherever we feel the thresholds are borderline, we will push the case to a manual check for some human intervention to see how we can make it better,” added Chitalkar.

At the same time, as StashFin relies mostly on AI and ML models to run its entire business, Chitalkar highlighted that they run the entire company with a small agile team. With India being a software service economy, StashFin also builds its models. Having the right tech team with the right mindset that understands the technology and its capabilities is key.

As fraudsters continue to find weaknesses in systems to scam financial services, fintech companies will only continue to develop more AI and ML models that can detect and prevent fraudsters from having their way. For companies like StashFin, the market opportunity in India is huge and fraud detection can be reduced, ensuring everyone has a chance to finance.