We must be careful how GenAI is used: PhonePe CTO Rahul Chari | Technology news

We’ve been quite open about the fact that in a fully interoperable network where there are no barriers to entry… if you take the number of applications that are out there and that are released, there is an option that has been built over time and in the back part. That’s why consumers choose based on what is reliable.

Since late 2019 we stopped cash back and incentive programs for users, so it’s the consumer’s choice in that environment. For companies that have invested time, effort and money along with the network to build this great digital payments ecosystem, I think the limit does not make sense.

Any type of limit is discouraging for any segment. The question then is why investment, innovation and entrepreneurship will occur in the future. Setting a limit is not an adequate model to resolve systemic risk. One methodology could be the model of having stricter standards based on the scale. We invest heavily in capacity planning… so we can ensure that performance, reliability and availability are always our first focus.

But regulators are not wrong either. Market concentration is a risk

The regulator’s concern is fair. The question is whether the solution is correct. Concerns can be addressed in multiple ways. The fact that new players are continually entering and becoming relevant (large) players shows that there is still appetite to enter and that there are opportunities to build new use cases. Having something like a limit might actually discourage this; may go against the most important reason.

The issue of commercial discount rate (MDR) is being debated lately, but nothing has happened from the regulators.

We have been very transparent about this. Zero MDR on UPI played a role in scaling up UPI, especially on the acceptance side. I think at some point payments for the sake of paying should be self-sufficient and building a purely payments company should be encouraged. I think MDR should be done at some point so that the unit economics and the payment itself are positive.

How do you use PhonePe GenAI in your processes?

When it comes to GenAI we must be careful with its use, especially when it comes to the regulated space. We have been using ML (Machine Learning) extensively in stopping risks and frauds for the longest period of time. At our scale of over 270 million daily transactions, I cannot use established rules to begin to determine whether a transaction should go through or not. We have been using ML significantly in cohorting and onboarding.

GenAI: We have to be cautious because we have to have explainability when doing things like underwriting. Even in the case of code generation, I think you have to be very careful when it comes to money management or use cases in the financial space.

How is PhonePe’s technology prepared to handle scale?

A large part of our effort in payments continues to focus on scale and performance. More than 50-70 percent of the work goes into ongoing maintenance of our systems. It’s also about having an extremely good team; having good talent in large-scale distributed systems is essential. One of the things we had the luxury of doing (this is our third startup) we built a network of incredible talent throughout our journey with our first startup and with Flipkart. Some of those best engineers and architects are still with us. We designed a system that on the first day would be delivering 10 million transactions. That was our goal the first day. There are many decisions that we made from the beginning that have helped us. Some of the other choices we made were to go on-premises, so payments operations don’t run on any public cloud. This gives us the ability to utilize the hardware to its full potential and also design many of our systems to perform efficiently outside of regulatory requirements, such as data localization. Of course, we use the public cloud for some of the other operations.

Fraud and cyber threats on UPI are increasing. How is PhonePe preparing for this?

Many of our investments are made in risk and fraud detection platforms. That is one of our platforms that we have created as Guardian and now we are looking to see if we can outsource it to the ecosystem and other actors. Whether it’s in terms of fingerprinting devices to ensure there’s no cloning of devices, or being able to have a platform where we can run our own ML models and rule sets at scale… we also do a lot of anti- pattern matching to detect whether a transaction being performed matches the user’s pattern. This may be through social engineering or any other malicious takeover attempt; We try to block such transactions.

When it comes to cybersecurity, there is a multiple approach. We have two different roles: one team looks from the outside and are more like ethical hackers and the other information security team looks from an internal perspective with regular updates etc.

Given the magnitude of UPI growth, we have to make more efforts to curb it (cyber threats). This is crucial because the next 300 million users who join will be the ones who will need even more support and education to not be let down.

There is a debate about using more Indian-made software or systems than those from multinationals. What is your point of view?

We have built our systems entirely on open source software. We’ve been able to use some of the best work people have done around the world to build population-scale systems at the lowest cost without being tied to any proprietary solution. We just opened three of our solutions. PhonePe Github is now a public repository. All of our container orchestration solutions have been developed in-house and are open source. And we also opened our security framework… and hope to give more back to the community. I think that since we have used so much of the open source, we should also give it back to the community.

First published: October 13, 2024 | 23:36 IS