One of the most popular trends on social media these days is the “before and after” physical transformation pictures of people. Everything from 10-year challenges and body fitness to hairstyles and ‘beardo’ looks are lapped up with great delight by netizens. But these same transformations are a nightmare for BFSI institutions and service providers when they need to verify user identity by comparing an ‘after’ pic with a ‘before’ pic. 

Consider this example – every Indian knows that Ankit Sharma and Ankita Sharma are two completely different people of opposite genders. Ask a computer to make this differentiation and suddenly you will need AI to do the job.

Banks and financial institutions carry the onus of verifying identity, address and other details at every step, but this process is highly error-prone and can lead to further errors and even fraud. In addition, customers have a poor digital experience because financial institutions' backends lack end-to-end digital processing. These are the problems that Karza Technologies is solving using AI, to create a dependable, automated and secure system to alleviate the strain of lengthy verification processes to decrease the danger of bank fraud. 

Founded in 2015, Karza Technologies chose to develop anti-fraud solutions for KYC applications. With 200+ clients across BFSI, including the top private sector banks, NBFCs, Fintech, and payment firms, they are revolutionizing identity authentication and risk mitigation with quick and secure digital solutions that enhance transactions at each stage - from screening to monitoring. 

They provide the most comprehensive suite of microservices for digital onboarding with 2 best-in-class solutions:

• Computer Vision for Optical Character Recognition (OCR), Face Match, Liveness, Tamper Checks, etc.
• Natural Language Processing algorithms for Name Match, Address Match, Related Party Identifications across millions of legal documents, etc.

While these can have multiple applications, Karza used it first to create the first-ever Unified Fraud Risk Score that considers identity, contactability, and employment risk assessment.

This solves a dual problem for the BFSI sector - frauds due to document falsification, and customers experiencing a broken digital experience due to multiple manual interventions. In India, data sources are as varied and scattered as its geography. And just like there is no single language natively common to all regions, there is no single identifier of customer details across sources, which complicates search and compilation and also renders “thin file” customers as unviable (due to high processing costs), resulting in their financial exclusion.

According to, Alok Kumar, Co-founder and CTO of Karza, “Automation attempts of these processes are bottlenecked by the fact that inputs are non-standardized and document quality is extremely poor, wherein OCR and face-match solutions struggle to deliver reliable accuracy.”

Powering Straight Through Processing of KYC Process with Advanced AI solutions

To fix these issues, Karza Technologies devised a process based on three primary solutions:

1. OCR Models – Karza's OCR text detection and text recognition algorithms are trained on hundreds of various font families and background sounds tailored for the BFSI industry, allowing for smooth text detection and identification on any Officially Valid Documents encountered during client onboardings. In addition, their card identification and parser models are retrained to accommodate different document formats; these parsers based on graph CNN make use of document topologies for better accuracy.

2. Face Based Models - They utilize stitching methods across government datasets to mine appropriate training datasets (lower resolution, Indian faces, worn-out document faces, etc.) in a scalable manner and use face detection. Face similarity combined with a face attribute model provides a second line of defence in challenging circumstances – typically like the ‘before-after example.

3. Name Analytics & Similarity - The Gender Classifier uses encoding methods to make training and generalization easier. At the same time, a combination of IR approaches and special features gives previously unavailable token-wise uniqueness, as well as a given name and surname score that can be rolled up or drilled down demographically. Custom Matching is also made possible – remember ‘Ankit and Ankita’?

AI and ML models get sharper with increasing volumes of data that can train them. 

“We are using government data sources in real-time for the verification process. For name analytics and similarity mapping, we currently use a database of 200+ million individuals from across the country. We have our own proprietary encoding and embedding techniques, and a unique feedback loop between the rule-based engine and deep learning model, which makes the overall generalization much better than other solutions in the market today. Another important feature is that we have made these solutions mobile-friendly to cater to remote areas where internet access cannot be taken for granted,” explained Alok Kumar.

Take the example of a Payday Lender that wanted to reinvent client onboarding for Personal Loans. Karza’s solution was a single interface for authenticating identity, contactability, and the use of multiple APIs to help with end-to-end customer journey automation.

The result was incredible, as it sped up verification of loan approval applications, reducing approval time to less than 5 minutes per file while allowing for real-time filtering of fraudulent applications through digital screening of 100% of applications. This increased monthly application volume from 0.5 million to 2 million without sacrificing quality.

So what does Karza do differently?

• Hundreds of publicly available government data sources are automatically stitched together 
• Adaptation across 200+ leading BFSI players ensures reliability at scale and diversity
• Data-driven preparation of training and evaluation datasets that minimize sampling bias
• Multiple models yielding interpretable features by name and address match like correctly spaced name or address string, given-middle-surname tokens, uniqueness score of each token and address buckets like house no, locality, building, landmarks, etc.
• Human-like cognitive skills to interpret Indian names, addresses, officially valid documents, faces, and liveness
• Data-driven approach taken to learn thousands of demographic quirks in a scalable manner continuously

The Karza Impact:

With technology and AI comes accountability, accountability for optimizing resources, and securing data. Karza has made a mark by preventing fraud worth INR 25 billion by weeding out bad applications even before they enter the system, saving 70% of processing time by digitizing processes realizing a 7x increase in business. Karza's technology can work not just with BFSI, NBFCs, Fintech, and Payment Companies, but also with Gaming, Hr Services, Fleet Aggregators, and other industries.

Interpreting and integrating data from an ecosystem as complex as India to drive digital inclusion through great consumer experiences is truly the holy grail that Karza Technologies is helping India Inc gun for.