Relevance of artificial intelligence in BFSI: Dr Jayaram K Iyer, Co-founder and CEO, DeepQuanty

One of the challenges faced when it comes to deploying artificial intelligence in BFSI is the lack of skilled employees in AI and advanced analytics

Supriya Rai
New Update
Artificial intelligence in BFSI

Automation is the key to achieve efficiency in businesses in the current contemporary world, and artificial intelligence is one of the technologies that is widely being applied in order to do so. However, in a sector like BFSI, which has been used to the manual workforce for a long time now, the application of new-age technologies faces its own set of challenges. In an interview DataQuest, Dr Jayaram K Iyer, co-founder and CEO, DeepQuanty Artificial Intelligence Labs Pvt Ltd talks about the importance of artificial intelligence in BFSI, challenges faced while implementing the same, and the issue of skills shortage in India.


Importance of artificial intelligence in the BFSI sector

From intellectually demanding actions such as computing risk scores to mundanely routine tasks such as transcribing data, the BFSI sector has it all. Improving any of those processes means superior customer experience, leading to improved market share; or can mean lower turn around time or lower costs, leading to improved profits. AI undoubtedly can help so improve. In the times of high competition in the BFSI and even higher costs (say due to NPAs), even small increments in market share or small savings in costs can be a matter of even existence.

Consider just a couple of use-cases. A loan application consists of an application form, past bank statements, statements of accounts, KYC documents and such. It takes time to not just transcribe the data from an application docket, but also to analyse the contents for its creditworthiness. It may take several days. Computer Vision developed basis deep learning Artificial Intelligence can read such forms and statements. It also performs cognitively heavy functions of an analyst - e.g. calculate a credit score. The result is not just lower costs and accuracy but quick turn-around. Another use-case is clearing cheques, especially the handwritten ones. Computer Vision products can read date and suggest if the date is acceptable. It can also read and compare the amount in figures and words. Lastly, it can compare the signature on the cheque with the core banking version.


Are there enough skilled employees in artificial intelligence and advanced analytics in India?

No. There is indeed a humongous shortage of quality talent for AI and Advanced Analytics in India and it is so even in the world. An interplay of three reasons explain why this shortage has come up.

(a) Firms are realizing that there are needs that hitherto could not be resolved or difficult to resolve can be resolved now. (b) Technological capability in analytics that can cater to such stated and unstated needs is exploding. The rise and democratisation of Big Data, Cloud Computing, Graphical Processing Units, Python and its array of analytical capabilities. (c) However, the supply of quality data scientists is lagging both in number and quality. This is because, the ecosystem of analytics is vast, complex and fragmented. Data scientists have to learn new languages, algorithms, frameworks and new tools by the hour. If the end delivery in on diverse (edge-) products (e.g. say an app in mobile), then they have to additionally customize and optimize for each type of CPU. A close analogy to this conundrum is the ever-increasing rise of personal transport (cars including Uber) and travel needs but an acute shortage of drivers.


Only people passion for analytics, continuous learning and never-say-die attitude can excel in this field.

Challenges faced while integrating new-age technologies in BFSI

A big challenge in BFSI is regulations - especially with respect to the privacy of data. Several applications in Computer Vision based on AI/ML thrive on cloud-based delivery. Banks demand on-prem and it is a challenge. Another problem is that just a few of them warm up to technologies that are nearly 100% open source. Geographic differences in handwriting post some unique challenges. Lastly, Karma. For e.g., consider in cheque processing, signatures matching. In legacy systems, card-signatures are typically stored in very low-resolution noisy images. They are sufficient for human-eye processing and help in fast transmission over net. However, when it comes to Automatic Signature Verification, past images pose problems.

How artificial intelligence is being used in solutions offered by DeepQuanty

At the heart of all DeepQuanty products are Deep Learning and Artificial Neural Network techniques. Products are developed basis learning from thousands of real life samples. Consider for e.g. SnapChek that automates reading and processing handwritten and printed cheques. Thousands of mock cheques were carefully designed and prepared for training. The mock cheques were digitized, processed, and converted to data using patent pending algorithms. Or consider ZapSkore that automates reading and processing bank statements. Post training on synthetic data, the images of bank statements are processed and converted to data using neural network algorithms. Further, on the extracted data, machine learning algorithms work to produce credit risk scores. Likewise, FormEasy is another AI driven product that extracts data from application forms such as Bank Savings / Current Account, Credit Card, etc.