Data-driven, intelligent decision making is one of the most common applications of AI in recruitment. The verification industry, which deals with humungous volumes of data, can dramatically increase the quality, speed, and efficiency of its solutions by combining AI and automation. 

Rampant fraud was visible in the employment market. The problem was so big that traditional manual checks were not a viable option and thus needed a technology-enabled solution. This is when IDfy got into this business and started doing employment background verifications, bringing the latest technologies to catch resume fraud, identity fraud, and misrepresentations at the time of hiring.  

With all these things in mind, we got in touch with Ashish Sahni, CTO of IDfy to understand how the AI-based verification startup is on a journey to prevent fraud and reduce risk.  

The beginning

Some times back, proper background checks on the drivers in the taxi aggregation business took off. Over the last 5-6 years, the prevention of fraud and illegal activities gain traction with the rapid opening of the financial services sector with digital payments, instant loans, and new banking solutions. 

“In Jan 2020, the RBI officially introduced Video KYC as a means to onboard customers remotely. By then, we already had our Video KYC platform ready, so were the first ones to hit the market with a 100% compliant solution. Over the last couple of years, we have become a household name for Video KYC services. Over the last couple of years, with the rise in e-commerce and on-demand economy, we have deployed our solutions to safely onboard delivery partners, customers, and sellers on these platforms,” said Ashish.

Coming to solutions

IDfy boasts several AI-based solutions across sectors for onboarding, verification and fraud detection.

• The startup enables seamless, fraud-free onboarding through AI-based solutions for detecting documents, extraction of relevant information, detecting liveness for individuals and more.
• The AI-based services also extend to the redaction of Aadhaar numbers on IDs to be compliant, detection of tampering on ID documents, and more.
• In addition, it has built AI-based solutions around the enablement of "branded" video calls.

Further, the Video Services platform is built using abstractions that are each designed to be scalable independently. Services are deployed on Kubernetes based nodes which can scale horizontally and vertically on demand. Based on concurrent video sessions, the company can scale up their media servers as part of the Video Service system. Overall the system is designed from the ground up to be scalable, available and resilient. 

But, what about data security

Data is completely segregated - “we have client level buckets to store profile captured artefacts.” Encryption is everywhere - from data at rest to data in transit with AES256-GCM based payload encryption. Moreover, in line with the government policy, data is localised- all artefacts and video recordings are stored in India.  

“Foolproof security is ingrained in our design - no third party can access IDfy’s APIs. Our Secure API Management Gateway manages endpoint authentication using API Key with Key Rotation Interface. This means clients must identify themselves every time, and keep updating its key regularly to further reduce the risk of unauthorised access,” said Ashish. 

Challenges at hand

The key challenges are with ensuring:

• Having the right data- The company has data pipelines that ensure that we have this covered
• Model retraining - Data will change over time and hence the AI models require updating from time to time. We have parameters at the service level which are monitored and ensure that this is reviewed periodically. We also have a strong support process in place that ensures we have early warnings for such data drift.
• Production readiness of new use cases – The company needs to be able to clearly define problems to solve, run experiments with defined failure modes and templatized applications with testing and security guardrails that enable fast and secure deployments. In fact, more than 90% of proof of concepts (POCs) make it to production, claims IDfy.

The road ahead

“We expect the industry to move to a trustworthy AI-driven verification and onboarding with only minimal human intervention”

Fraud detection being a critical pain point, the industry is expected to come together to pool data for fraud detection. Also, it is important to note that while a "customer" is often seen as the perpetrator of fraud, they are often the victims of fraud. Preventing bad actors from defrauding genuine users through social engineering will be a major dimension of AI-led fraud detection.

Last year, the company received Rs 86 crore in a Series D funding round. “We plan to use the newly infused funds from TransUnion and Blume Ventures to strengthen our product offerings. We are also expanding our business internationally partnering with firms in South East Asia as well as the Middle East to provide merchant onboarding services,” concluded Ashish. 

Image from IDfy blogs