Vishal Verma discusses AI-driven fraud detection, mule accounts, and network intelligence for India's UPI ecosystem
Summary
Vishal Verma of LexisNexis Risk Solutions explains how the speed of digital payments creates a narrow window for fraud detection, necessitating a shift from transaction-by-transaction checks to improved risk assessment models. He highlights the critical role of 'mule accounts' opened by genuine individuals and advocates for 'perpetual KYC' to detect behavioral deviations. Verma argues that AI is a double-edged sword, used by both fraudsters and banks, and supports the RBI's draft framework for temporary transaction holds. He emphasizes that network intelligence across sectors like telcos and travel can increase fraud detection rates by over 50%, provided there is robust governance and data sharing.
Key takeaways
- 1 banking? Let's get telcos in, let's get insurance companies, let's get travel portals in. So wherever consumers today are having any digital interaction, that is your potential network, in my view. And we have scenarios, and I can talk about one of them where we see that we can use the intelligence from how somebody behaves on a digital platform for booking travel tickets. That information can be used to understand whether this current banking transaction is actually fraudulent or no.
Unidentified speaker (speaker_1) · 11:49
Episode sections
- 0:00 – 3:57
Introduction and the Speed of Fraud
The host introduces the topic of financial fraud and the challenge of detecting it within the sub-second timeframe of digital payments. Vishal Verma notes that while payment speed benefits consumers, it leaves banks with little time to investigate and stop fraudulent transfers, necessitating improved risk assessment rather than slower payments.
- 3:57 – 5:25
The Role of Mule Accounts in Fraud Chains
Verma explains that mule accounts are critical for fraudsters to hide proceeds of crime by transferring funds through a series of accounts, often owned by genuine individuals who have rented or sold their identities. These accounts act as a safe haven, allowing money to move away from the original bank's visibility before reaching the mastermind.
- 5:25 – 7:57
Strengthening Onboarding and Behavioral KYC
The discussion highlights that mule accounts are often opened by legitimate customers with real documents, making initial onboarding checks insufficient. Verma advocates for 'perpetual KYC' and analyzing the full behavioral context of transactions against declared profiles to detect when a genuine customer's behavior shifts to suspicious activity.
- 7:57 – 10:46
AI as a Double-Edged Sword in Fraud Detection
Verma states that AI is used by both fraudsters to create synthetic entities and commit fraud at scale, and by banks to analyze vast amounts of data in real-time. Banks use AI to generate risk-based scores for transactions, allowing them to pause high-risk activities or process low-risk ones efficiently, a task impossible for humans to perform manually at scale.
- 10:46 – 15:47
Barriers to Network Intelligence and Data Sharing
The host asks about practical barriers like privacy and competition that prevent institutions from sharing fraud signals. Verma defines network intelligence broadly, including non-financial sectors like telcos and travel portals, and emphasizes that trust, common definitions, reciprocity, and governance are essential for a workable data-sharing model in India.
- 15:47 – 19:20
RBI's Draft Framework and Cooling Periods
Verma supports the RBI's proposal for temporary holds on suspect transactions, citing examples from Singapore. He argues that with sufficient data collection (device, location, behavior) and AI analysis, banks can implement 'cooling periods' to investigate high-risk transactions without eroding customer trust, provided the process is explainable and transparent.
- 19:20 – 22:24
The Necessity of Human-in-the-Loop
Despite AI advancements, Verma asserts that humans remain essential in fraud detection to handle new modus operandi and provide explainability to consumers. He notes that bankers are not replacing humans but are upskilling them to manage complex cases and engage with customers, ensuring that fraud detection systems remain adaptive and trustworthy.
- 22:24 – 25:20
Future Resilience for India's Digital Payments
Looking ahead, Verma predicts UPI will reach one billion transactions per month. He identifies two key requirements for resilience: assessing fraud at every step of the customer journey and institutionalizing network intelligence through anonymized data sharing across the ecosystem to protect consumers and sustain growth.
Claims worth verifying
The speed of digital payments leaves very little time for banks to investigate or spot fraud and stop the money.
Unidentified speaker (speaker_0) ·0:00
How to verify: Verify by reviewing RBI guidelines on real-time payment settlement times and bank fraud response protocols; compare with industry reports on average fraud detection latency. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
Mule accounts are a set of accounts where proceeds of frauds or stolen funds are transferred through a series of accounts to hide the money from the original bank.
Unidentified speaker (speaker_1) ·4:11
How to verify: Verify by searching for RBI enforcement actions or news reports detailing mule account networks and their role in money laundering in India. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
Banks are deploying AI tools to assess more information about customers and analyze data from different channels in real-time to generate risk-based scores.
Unidentified speaker (UNKNOWN) ·10:01
How to verify: Verify by reviewing annual reports of major Indian banks (e.g., SBI, HDFC, ICICI) for disclosures on AI/ML investments in fraud detection. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
LexisNexis Risk Solutions analyzes almost 116 billion transactions on an annual basis.
Unidentified speaker (UNKNOWN) ·13:36
How to verify: Verify by checking LexisNexis Risk Solutions' official website, press releases, or annual reports for transaction volume statistics. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
Network intelligence can increase fraud detection rates by almost more than 50% for an institution.
Unidentified speaker (speaker_1) ·13:56
How to verify: Verify by requesting case studies or white papers from LexisNexis Risk Solutions or other financial crime data providers detailing detection uplift metrics. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
Credit Bureaus are properly licensed under the Reserve Bank of India.
Unidentified speaker (speaker_1) ·14:22
How to verify: Verify by checking the RBI's official list of authorized Credit Information Companies (CICs) in India. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
RBI's draft framework proposes temporary holds on suspect transactions to give banks time to act.
Unidentified speaker (speaker_0) ·15:47
How to verify: Verify by accessing the RBI's official website and searching for the 'Draft Framework on Prevention of Financial Fraud' or related circulars. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
UPI will touch a billion a day in terms of the number of transactions.
Unidentified speaker (UNKNOWN) ·22:43
How to verify: Verify by checking National Payments Corporation of India (NPCI) monthly transaction statistics for UPI volume trends. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
AI tools are used by fraudsters to create synthetic entities and open accounts for entities that do not exist.
Unidentified speaker (UNKNOWN) ·8:49
How to verify: Verify by reviewing law enforcement press releases or academic papers on synthetic identity fraud techniques in digital banking. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
Fraudsters can test compromised login credentials against millions of accounts in less than 60 seconds.
Unidentified speaker (UNKNOWN) ·9:28
How to verify: Verify by reviewing cybersecurity incident reports or threat intelligence briefings from firms like LexisNexis, Kaspersky, or FBI IC3 reports on credential stuffing attacks. For recent events, verify 2026-08-04 through 2026-11-04. Use another period only when the speaker explicitly names it.
Do your own research
- Review the RBI's Draft Framework on cooling periods for suspect transactions to understand regulatory expectations.
- Analyze LexisNexis Risk Solutions' business model and potential conflicts of interest in promoting data-sharing solutions.
- Verify claims regarding the 50% increase in fraud detection rates through anonymized data sharing by checking primary research or case studies.
- Check the current status of UPI transaction volumes and projections for reaching one billion transactions per month.
- Investigate the legal and privacy frameworks governing data sharing between banks, telcos, and travel portals in India.
- Review the definition and implementation of 'perpetual KYC' in the context of Indian banking regulations.
- Assess the technical capabilities of AI systems used by major Indian banks for real-time risk scoring.
- Look for counter-arguments regarding the privacy risks and potential false positives of implementing widespread network intelligence sharing.
Full transcriptexpand
- 0:00
- Unidentified speaker (speaker_0): How are criminals exploiting the speed of digital payment? The same speed which makes the payment so fast, the same speed leaves very little time for the banks to investigate or spot a fraud and then stop the money. We don't have to slow down the the pace of digital payment. I think we have to improve the way we do our risk
- 0:19
- Unidentified speaker (speaker_0): assessment. I think that's where the catch
- 0:21
- Unidentified speaker (UNKNOWN): is.
- 0:26
- Unidentified speaker (UNKNOWN): Hello
- 0:27
- Unidentified speaker (speaker_0): and a warm welcome to a very special podcast series Unusual Suspects. I'm your host, God of
- 0:38
- Unidentified speaker (speaker_0): disruptors yet not in your face. Their disruptions have consistently delivered on the purpose and the mission they've embarked on. Their The values are as strong as the mission. Unusual suspects suspect streams are new. on nearshot.in, Spotify.com, Apple Podcasts, YouTube, MoneyControl.com, and a host of other podcast streaming platforms
- 0:58
- Unidentified speaker (speaker_0): and directories.
- 1:00
- Unidentified speaker (UNKNOWN): Our
- 1:00
- Unidentified speaker (speaker_0): guest today will unpack an issue that has perhaps affected all of us at some point in time or the other,
- 1:07
- Unidentified speaker (UNKNOWN): or
- 1:07
- Unidentified speaker (speaker_0): someone just one degree removed from
- 1:10
- Unidentified speaker (UNKNOWN): us.
- 1:11
- Unidentified speaker (speaker_0): It's also climbing rapidly up the regular. regulators, primarily the Reserve Bank of India's agenda.
- 1:19
- Unidentified speaker (UNKNOWN): We
- 1:19
- Unidentified speaker (speaker_0): are talking about financial fraud. While a highly connected, interconnected payments ecosystem moves money quickly, it also clears the way for faster fraud. And that is a reality that we need to be alive to and alert to. Vishal Vurma, Verma, Senior Director APAC at Lexis next. Nexis
- 1:39
- Unidentified speaker (speaker_0): Risk Solutions, joins us to explore how India's fraud ecosystem is evolving, whether banks can move beyond investigating individual transactions to identifying the wider patterns, identities, and of course the relationships behind them. At LexisNexis Risk Solutions,
- 1:59
- Unidentified speaker (speaker_0): Vishal which are leads strategy and business growth across financial. crime compliance, fraud prevention, and digital identity solutions. With more than 20 years of experience spanning Asia Pacific and the Middle East, he has worked with financial institutions to strengthen risk management capabilities across corporate
- 2:19
- Unidentified speaker (speaker_0): banking, credit information, advanced analytics, and of course digital transformation. Welcome to the show with Vishal. Delighted to have you.
- 2:28
- Unidentified speaker (UNKNOWN): Thank
- 2:28
- Unidentified speaker (speaker_1): you, Gorf. Glad to be here and thanks for the kind introduction.
- 2:31
- Unidentified speaker (speaker_0): You know Vishal, let's it's a very this is a very basic question that we can start off with, but it also sets the context and also clears our understanding on something that affects us but we know very little about. How are criminals exploiting the speed of digital payments? I ask you this question because you were right there. You are you're solving this. You have
- 2:51
- Unidentified speaker (speaker_0): you have a front row sit and you are kind of shepherding many changes so
- 2:57
- Unidentified speaker (UNKNOWN): that
- 2:57
- Unidentified speaker (speaker_0): we remain ahead of the criminals criminals rather than the other way around. Bishal?
- 3:02
- Unidentified speaker (speaker_1): Sure, I think it's a good start and I think the answer lies in the question. The same speed which makes the payment so fast, the same speed leaves very little time for the banks to investigate
- 3:14
- Unidentified speaker (UNKNOWN): or
- 3:14
- Unidentified speaker (speaker_1): spot a fraud and then stop the money. Today we talk about money, you know, moving the accounts instantly, subsecond. And the same thing doesn't have a luxury with the bank to start the investigation in a spot where the patterns don't look nice and then stop the money transfer. So it's the speed which is which is kind of a double-edged double edged sword
- 3:33
- Unidentified speaker (speaker_1): sometimes. We don't have to slow down the the pace of digital payment. I think we have to improve the way we do our risk assessment. I think that's where the catch is. So criminals definitely look at exploding that speed which everybody's offering in the market, which is the right thing to do for consumers. And this becomes sometimes the gap where majority of the cases today are
- 3:53
- Unidentified speaker (speaker_1): happening because of criminals just tapping that speed.
- 3:56
- Unidentified speaker (speaker_0): Right.
- 3:57
- Unidentified speaker (UNKNOWN): You
- 3:57
- Unidentified speaker (speaker_0): know, Vishal, there's this whole concept and if I may so avail ava the a vehicle uh called the Mule Accounts, which is uh why are mule accounts so critical to the fraud chain, Vishal?
- 4:11
- Unidentified speaker (speaker_1): So yeah, this is interesting, and as you know, sometimes we say in our industry that you stop the mules and you stop the frauds. Because fraud is like if if there are no mule accounts, fraudsters have no place to hide their money in the digital world, as the entire payment. ecosystem is getting digitized, where the money would go. You have to transfer the money
- 4:31
- Unidentified speaker (speaker_1): somewhere. So Mule accounts essentially are a set of accounts where you transfer the proceeds of frauds or stolen funds, and then you transfer through a series of accounts, which eventually goes to the mastermind. It's very rare that the direct Mule account is actually the mastermind.
- 4:48
- Unidentified speaker (UNKNOWN): So
- 4:48
- Unidentified speaker (speaker_1): if I if I kind of defraud you, I take the money into X account, X probably wouldn't even know that money is getting defrauded from somebody. And that money moves from X through series of accounts through small, small split portions. And that's how it sometimes, or actually most of the times, it lose the sight of the original bank.
- 5:06
- Unidentified speaker (UNKNOWN): And
- 5:06
- Unidentified speaker (speaker_1): that's why mules are so critical for fraudsters. They they nurture them, they farm them over a period of time. There's a series of rented accounts, and sometimes you put people under a lot of pressure to you know let go of their accounts. So it is
- 5:19
- Unidentified speaker (UNKNOWN): a
- 5:19
- Unidentified speaker (speaker_1): safe haven, so to say, for all the criminals and fraudsters to park that money and eventually take it
- 5:24
- Unidentified speaker (UNKNOWN): out.
- 5:25
- Unidentified speaker (speaker_0): So essentially, you know, mule accounts have to be opened before they can be used. Uh that is the necessary, you know, condition, if I may say so. Uh and these are they the they they're frequently opened with uh gel. genuine uh as you said they're these are genuine individuals uh but they also have
- 5:45
- Unidentified speaker (speaker_0): rented identities and these uh functions are probably on the ca uh follow the characteristics of what we may call as a borrowed bank account. Uh how can banks make the onboarding process itself more robust? So in if we do that, we nip it in the bud, butt, isn't it, Bishal?
- 6:02
- Unidentified speaker (speaker_1): Yes. and that's a very important point because you know the way you started the question, because most of these accounts are actually legitimate customers.
- 6:10
- Unidentified speaker (UNKNOWN): The
- 6:11
- Unidentified speaker (speaker_1): customer is real, the person is real, the documents are real,
- 6:14
- Unidentified speaker (UNKNOWN): and
- 6:14
- Unidentified speaker (speaker_1): they're opening a bank account. You you can't refuse that as a banking institution. But if the same account is handed Hand it over to somebody else to operate, that's when the challenge comes in.
- 6:24
- Unidentified speaker (UNKNOWN): So
- 6:24
- Unidentified speaker (speaker_1): you got to have checks at different levels. When you onboard a customer, you have to do proper KYC. Today we are living in a perpetual KYC world, so you do it on a period, you
- 6:33
- Unidentified speaker (UNKNOWN): know,
- 6:33
- Unidentified speaker (speaker_1): periodic basis. And that's just a one-time check sometimes when you onboard customer. How they are behaving or how they are dealing with their banking service, which is an account, or let's say even loan, that should be in line with the profile they have declared to the bank at the time of onboarding relationship. And that's where it's important to understand the
- 6:53
- Unidentified speaker (speaker_1): context of every transaction. So if I'm
- 6:56
- Unidentified speaker (UNKNOWN): a
- 6:56
- Unidentified speaker (speaker_1): customer which opened an account and submitted some of let's say my income details at the time of account opening,
- 7:02
- Unidentified speaker (UNKNOWN): that
- 7:02
- Unidentified speaker (speaker_1): should tally with the behavior I'm doing by my banking transactions, right? It cannot be like really way off from my average banking transaction. And so yeah UIC EYC becomes very important. I mean documents are important but it's important to understand the full behavioral context, the transactional details every time you assess a customer journey.
- 7:22
- Unidentified speaker (speaker_1): And
- 7:22
- Unidentified speaker (UNKNOWN): I
- 7:22
- Unidentified speaker (speaker_1): think one point where, especially in India, all the banking institutions have done a really good job of assessing fraud at every customer touch point, which is onboarding, when you log into your banking app, when you add somebody as a beneficiary, or you make a payment, or you receive a payment. all those touch points are getting now screened or risk assessed for
- 7:42
- Unidentified speaker (speaker_1): potentially fraud. So that's where it's it's critical that you understand your customer at every touch point and then see whether they have the propensity
- 7:51
- Unidentified speaker (UNKNOWN): to
- 7:51
- Unidentified speaker (speaker_1): become mule from a normal customer. That that's what is right now happening in the SD coral.
- 7:56
- Unidentified speaker (UNKNOWN): Yeah,
- 7:57
- Unidentified speaker (speaker_0): you know, Vishhal, Michelle, doesn't AI if we are in the high noon of AI, and uh so isn't artificial intelligence a very, very potent tool in the hands of the regulators and the financial institutions and banks? Because uh can AI and n and network intelligence help identify suspicious activity early enough to stop the funds moving beyond
- 8:16
- Unidentified speaker (speaker_0): reach? What has been the experience so far? Uh do you think there's a lot of promise there, Michelle?
- 8:23
- Unidentified speaker (speaker_1): Yes, absolutely. And again, AI is is
- 8:25
- Unidentified speaker (UNKNOWN): is
- 8:25
- Unidentified speaker (speaker_1): everywhere, right? I mean and then it's with us, it's also with the fraudster. So if you take a step back, what AI does in any scenario, whether it's banking or manufacturing or media, it lets you do complex things at a scale.
- 8:40
- Unidentified speaker (UNKNOWN): And
- 8:40
- Unidentified speaker (speaker_1): you if you play that back in the fraud story, it lets you do complicated frauds at
- 8:45
- Unidentified speaker (UNKNOWN): It
- 8:45
- Unidentified speaker (speaker_1): is a scale, which wasn't really possible a few years ago.
- 8:49
- Unidentified speaker (UNKNOWN): All
- 8:49
- Unidentified speaker (speaker_1): the tools we have, which we use for our productive purpose, you know, to create multiple uh transcripts, for example, you know, media companies use it to kind of make the video editing better. All those tools are used by Frosters as well. So I can create synthetic entities using multiple isolated pieces of information and then have probably open an account for
- 9:09
- Unidentified speaker (speaker_1): an entity which doesn't even exist. Same AI tool is used by
- 9:14
- Unidentified speaker (UNKNOWN): the
- 9:14
- Unidentified speaker (speaker_1): banking industry to kind of minimize the fraud at different levels. Now, the scale with this happening, it sometimes becomes impossible for human to handle, to tackle. And that's where to your point, the banks are also deploying AI tools.
- 9:28
- Unidentified speaker (UNKNOWN): For
- 9:28
- Unidentified speaker (speaker_1): example, if you have water tax, bottatech can actually ensure that if I'm a fraudster, So I can I can and if I have access to some compromised login ID details, which I got through, let's say, through Dart. dark web or a data breach, I can fire that query into millions of accounts in probably less than 60 seconds.
- 9:46
- Unidentified speaker (UNKNOWN): And
- 9:47
- Unidentified speaker (speaker_1): even if I get five, 10 success, I'm happy I can take the money out and go away. It's impossible for a human being to do that. It was impossible a few days, a few years back. Now, what banks are doing, they're also, of course, deploying AI. And the way they look at AI is
- 10:01
- Unidentified speaker (UNKNOWN): they
- 10:01
- Unidentified speaker (speaker_1): are able to assess a lot more. information than they had before about the customer. And they can analyze all the data, which are coming from different channels, different patterns, on a real time to say whether this transaction or this activity by my consumer looks genuine or no.
- 10:17
- Unidentified speaker (UNKNOWN): So
- 10:17
- Unidentified speaker (speaker_1): then you have a risk-based score.
- 10:19
- Unidentified speaker (UNKNOWN): And
- 10:19
- Unidentified speaker (speaker_1): basically, that risk-based score, you can take an action. If it's a high-risky high-risk transaction, you can probably pause it, you know, call customer, ask a few questions and see. does it make sense? It's a medium risk, you can actually go ahead and process the transaction. Or if it's a low risk, you want to give a better customer experience. So you're right, AI is there for everybody to use. And I can see all around when I speak to
- 10:39
- Unidentified speaker (speaker_1): senior bankers, India or overseas, they all are using it at different levels. So it's definitely picking up Gorra. Barb.
- 10:46
- Unidentified speaker (speaker_0): You know, network intelligence w only works if institutions can see the same. patterns. Uh
- 10:53
- Unidentified speaker (UNKNOWN): but
- 10:53
- Unidentified speaker (speaker_0): there of course can be practical barriers, isn't it? What practical barriers such as privacy, consent, data standards, and competition stop banks, fintechs, and payment providers, and we're not just talking about banks, there's a whole universe of institutions there from sharing fraud signals today. And what would a workable model look
- 11:13
- Unidentified speaker (speaker_0): like in India? Because everything eventually will boil down to sharing information across even peer group and competitive competition companies on the currencies trust here, and how do you go about building that?
- 11:29
- Unidentified speaker (speaker_1): Yes, and I think I forgot to mention in your last question. So network intelligence is very important. And and you you mentioned two words which are critical. I think one is what is network? How do we define what a network is? Are we talking about a few banks coming together and sharing data? That could be one network, of course. Or are we talking about going beyond
- 11:49
- Unidentified speaker (speaker_1): banking? Let's get telcos in, let's get insurance companies, let's get travel portals in. So wherever consumers today are having any digital interaction, that is your potential network, in my view. And we have scenarios, and I can talk about one of them where we see that we can use the intelligence from how somebody behaves on a digital platform
- 12:09
- Unidentified speaker (speaker_1): for booking travel tickets. That information can be used to understand whether this current banking transaction is actually fraudulent or no. So there are examples happening. And second, which is I think a very powerful word With what you said, is trust. And I think for any network to happen or network to nurture or persist in a
- 12:29
- Unidentified speaker (speaker_1): market, you need a trusted operating model. Why would I share my data with anybody else? It's like in a personal relationship, also, we need trust to have any subnetwork, right? So coming back to the fraud parlance, you are absolutely right. A bank can only see what's happening within its ecosystem. And as I said in my first answer, in mostly
- 12:49
- Unidentified speaker (speaker_1): the fraud payment, the fraud proceeds. go through Mule accounts and these are Mule networks. And after one or two transactions, the second, third hop, as we call for mules, will go through outside the bank, the original bank from where the transaction is originated. So there is no impossible for one bank to figure out oh where the money actually has gone. The switch Which
- 13:09
- Unidentified speaker (speaker_1): can see if you are the country countries switch which you can see where the payment is moving, but that is mostly post post-factor. facto.
- 13:14
- Unidentified speaker (UNKNOWN): You
- 13:14
- Unidentified speaker (speaker_1): won't be able to stop that real time. So, network intelligence is one way that we share data, we share risk signals. We don't have to really create
- 13:24
- Unidentified speaker (UNKNOWN): a
- 13:24
- Unidentified speaker (speaker_1): huge database, you know, having all the details about customer. That's not needed to prevent fraud to happen in today's world. You have technology which can enable sharing of very specific risk signals,
- 13:36
- Unidentified speaker (UNKNOWN): which
- 13:36
- Unidentified speaker (speaker_1): we believe are important fraud, you know. for fraud propensity check. Those can be shared in a way, and to you rightly mentioned about taking care of things like privacy, data confidentiality. We can have anonymized data sharing models. So that's the network intelligence. And we like we analyze almost 116 billion transactions on an annual basis in our network, and we
- 13:56
- Unidentified speaker (speaker_1): have seen the network effect on the fraud detection rate is almost more than 50%.
- 14:01
- Unidentified speaker (UNKNOWN): So
- 14:02
- Unidentified speaker (speaker_1): in institution. an is able to detect 50% more frauds if they have access to the network data. And so your second question, second part of your question, Gar, was around the definition and what is the right model for India. We've seen few models. Memory Credit Bureau is a classic example, which is properly licensed under Reserve Bank of India. All the
- 14:22
- Unidentified speaker (speaker_1): banks come together and you know share data.
- 14:24
- Unidentified speaker (UNKNOWN): I
- 14:25
- Unidentified speaker (speaker_1): think there are a couple of points when we look at any trusted operating model. I would think one is the common definitions of what is getting ability to be shared. And they have to be common definitions. Okay, if we define this as a fraud or a suspicious, and we share that into the consortiums or the network. Second, I think
- 14:45
- Unidentified speaker (speaker_1): is which is very important, is the principle of reciprocity. Everybody shares in the network, not that just one or two large banks or large institutions come and share and the smaller ones don't. Everybody has to uh share so there is equal respect for what other banks are doing. And third, I think is the governance. How do you
- 15:04
- Unidentified speaker (speaker_1): increase or ensure there is enough governance, there is enough explainability to both consumers, to the respective banks, as well as to the regulators? You should be able to have proper training. trail of data that how you've taken a decision on a customer or a transaction using the network intelligence, from where the data came, how it came in what
- 15:24
- Unidentified speaker (speaker_1): form. You should be able to have a mechanism which you can decode it. We don't need to share consumer details. We don't need to share PII data. All that can be stored separately. And enough protocols are available today to ensure that data is you know not leaked and compromised. So I mean it's a big subject, but in in short, common principles need to be agreed. And
- 15:44
- Unidentified speaker (speaker_1): I think that's the way forward, Korov, I believe, for India.
- 15:47
- Unidentified speaker (speaker_0): You know, let's uh change uh keep the focus on trust on a on a complete complex uh from a completely different angle. You know the RBI's draft framework proposes temporary holds on suspect transactions to give banks time to act. And that's a very legitimate and value a valid premise. Uh on the technology side,
- 16:07
- Unidentified speaker (speaker_0): how can banks keep false positives low? Because that can also turn out to be a bit of a concern, uh, you know, through explainable risks, well-tuned thresholds, and human reviews. Because I ask you this question because at the end of the day, you banks will need to have and enjoy
- 16:26
- Unidentified speaker (UNKNOWN): the uh
- 16:27
- Unidentified speaker (speaker_0): you know the trust of the genuine customers. If the false positives continue to grow and reach a certain critical mass or beyond a certain threshold, you could actually end up eroding the genuine customers' trust. So how do you go about doing that? Because this is something where you need to find fine uh strike a fine balance and reach
- 16:47
- Unidentified speaker (speaker_0): equilibrium, isn't
- 16:48
- Unidentified speaker (UNKNOWN): it?
- 16:48
- Unidentified speaker (speaker_0): Michelle?
- 16:50
- Unidentified speaker (speaker_1): Yes. No and and I think it's a very good initiative from the RBN. We have seen it in few other markets like Singapore and you know other emerging economies around where the Reserve Bank is stipulating a cooling period for the transactions to be processed. And it's very much possible, Gorov, today with technology, again essentially what we've been talking, if you
- 17:10
- Unidentified speaker (speaker_1): have enough information about your customer and the transaction under investigation, if you're able to collate all the data points in near real time or real time,
- 17:20
- Unidentified speaker (UNKNOWN): and
- 17:21
- Unidentified speaker (speaker_1): able to use technologies like AI to assess that whether it's a genuine transaction or no. So, simple example: if I'm making a digital payment, let's say I'm sending money to you, whether it's a UPI or through bank-to-bank transfer, there are different elements involved. My login credentials,
- 17:37
- Unidentified speaker (UNKNOWN): the
- 17:37
- Unidentified speaker (speaker_1): device from which I'm making a payment. the Wi-Fi hotspot, the IP, the location through different triangulation of the metrics there, and my behavior. How do I normally transact with my mobile app versus my laptop? How do I hold the mobile? How do I press? And all those elements become part of your risk assessment.
- 17:54
- Unidentified speaker (UNKNOWN): If
- 17:55
- Unidentified speaker (speaker_1): you're able to collect all that information and then have fairly good confidence
- 17:59
- Unidentified speaker (UNKNOWN): that,
- 18:00
- Unidentified speaker (speaker_1): okay, this transaction looks risky or high risky.
- 18:03
- Unidentified speaker (UNKNOWN): You
- 18:03
- Unidentified speaker (speaker_1): can put it through a cooling period or that,
- 18:05
- Unidentified speaker (UNKNOWN): uh,
- 18:06
- Unidentified speaker (speaker_1): you know, before you process the final payment and then do your investigation. Maybe you need a couple of questions from customers or maybe other parties saying, hey, this doesn't look like a normal transaction. Is there something we need to know about this before we process? And I think the important point you said that they need to have the banks need to have trust of genuine customers. And I think when you
- 18:25
- Unidentified speaker (speaker_1): are transparent.
- 18:27
- Unidentified speaker (speaker_1): you are explainable in your justifications to the consumer, I think you can create and build trust. All of us are customers of different banks, right? If my bank comes and asks me five questions of a transaction, which they found suspicious, and if I found the questions legitimate and fair questions, I think I would be fine. Of course, I would like them to proceed faster and you know not stop me from
- 18:47
- Unidentified speaker (speaker_1): doing banking. It's all about the details. And to get details, you need to have details. That's why data collection becomes important, the network intelligence becomes important, and how best we can use all the data through technology, running models, whether algorithms or rules, and have a clear-cut response that okay,
- 19:06
- Unidentified speaker (UNKNOWN): a lot
- 19:06
- Unidentified speaker (speaker_1): of transactions are processed in the last one minute. These three are high risky, and we need a step up. We need more authentication from consumers.
- 19:13
- Unidentified speaker (UNKNOWN): As
- 19:13
- Unidentified speaker (speaker_1): long as we can explain. to the consumers and to the regulators, I think that measure from RBI is very much achievable today.
- 19:20
- Unidentified speaker (UNKNOWN): You know,
- 19:20
- Unidentified speaker (speaker_0): Vishal criminals are also using AI and uh from and you you can see the tools and you can um uh there they're they're dime a dozen and it's a plenty now from voice cloning to synthetic media in scams like digital error arrests and which we see many people falling uh you know, victim to
- 19:40
- Unidentified speaker (speaker_0): that. What has changed in the last year, and what does uh keeping a human in the loop concretely look like on a bank's fraud desk, Michelle?
- 19:51
- Unidentified speaker (speaker_1): Yes, I mean so
- 19:53
- Unidentified speaker (speaker_1): human in the loop is is a is a
- 19:55
- Unidentified speaker (UNKNOWN): I
- 19:55
- Unidentified speaker (speaker_1): mean a question. All the all the bankers I meet almost on a daily basis, they have this, and when they talk about designing their fraud programs for next year,
- 20:03
- Unidentified speaker (UNKNOWN): or
- 20:03
- Unidentified speaker (speaker_1): they're scaling up their operations and also getting technology.
- 20:07
- Unidentified speaker (UNKNOWN): I
- 20:07
- Unidentified speaker (speaker_1): think nobody's talking about doing away with the human in the loop.
- 20:11
- Unidentified speaker (UNKNOWN): But
- 20:11
- Unidentified speaker (speaker_1): human in the loop does not mean that for every transaction I should be asking my customers, hey, is it you making transactions?
- 20:17
- Unidentified speaker (UNKNOWN): Then
- 20:18
- Unidentified speaker (speaker_1): you will lose that customer, right? I mean, banks also work in a very competitive world, and as a consumer, I want the best service with the least friction if I'm genuine. So human in the loop. here is after having all that information which we spoke about about the transaction, about the channel, about the context, about the behavior, about the network, I'm able to collate all
- 20:38
- Unidentified speaker (speaker_1): the data into an iSchool machine learning algorithms or AI tools. After that, I'm sure there'll be transactions where you still want to verify further couple of things. And that's where human humor in the loop becomes very important. For a bank of a scale where they have millions. and millions of consumers, you would never be able to get a perfect score, or at least today, then you can just run
- 20:58
- Unidentified speaker (speaker_1): your fraud detection shop without having any humor in that process, right? Because the new patterns are coming. Machine learning or AI will tell you what it is right today or wrong today. Frouds Fraud sters are also evolving. They are using the same AI tool. Tomorrow we have a new MODAS Modus operandi Operandai coming into. fraud. Bottles have to change. So that's why human in the loop is very
- 21:18
- Unidentified speaker (speaker_1): important to spot something, hopefully, which machine is not able to spot. And then also have an element of engagement with the end consumer and saying, hey, we are calling from this bank. We've seen this thing. And explainability when it comes to a human, it's just more easy for a consumer to kind of you know put the trust back into systems. So I think again, this is my my discussion with all
- 21:38
- Unidentified speaker (speaker_1): the seniors. senior banks and and of course some of the regulators where we present on a regular basis, they all are talking about keeping that human in the loop very, very in-depth from the entire when they design their fraud framework. So I don't think Sagar that's going away anywhere. And I think people are going to more put more emphasis on creating more qualified human in the loop. do.
- 21:58
- Unidentified speaker (speaker_1): You know, with technology, we also have to continue to empower and educate people who are. managing fraud. Somebody who has a good 20 years of knowledge of managing fraud prevention, if he or she doesn't know what's happening in the last one year, probably your knowledge is obsolete because a lot has changed in the last 12 months. So it's important that those human in the loop are educated, up to
- 22:18
- Unidentified speaker (speaker_1): speed. That's when they can add value to not only to their institution, but also to the to the end consumer.
- 22:24
- Unidentified speaker (speaker_0): And finally, Vishal, Vishhal, uh what will it take to make India's next phase of digital financial growth more resilient to fraud, Vishhal?
- 22:33
- Unidentified speaker (UNKNOWN): Yeah,
- 22:33
- Unidentified speaker (speaker_1): interesting question. I mean, I was I was watching another broadcast last week, I guess, and and I heard that probably UPI will touch a billion a day in terms
- 22:43
- Unidentified speaker (UNKNOWN): of the number
- 22:43
- Unidentified speaker (speaker_1): of transactions, hopefully in a couple of months. So that's massive volume, right? We are talking about the volume which none of the other countries have seen. And I think it'll only grow. The growth rate might slow down, but it will only grow. My my wish list will be
- 22:57
- Unidentified speaker (UNKNOWN): two
- 22:57
- Unidentified speaker (speaker_1): things, and I keep it very simple, Goros. One,
- 23:00
- Unidentified speaker (UNKNOWN): which
- 23:00
- Unidentified speaker (speaker_1): I think a lot of banks are doing it, and a lot of non-financial institutions are also doing it.
- 23:05
- Unidentified speaker (UNKNOWN): See,
- 23:06
- Unidentified speaker (speaker_1): fraud doesn't have a start and a stop in a journey. So don't think that one leg of my consumer interaction is more risky than the other. leg.
- 23:15
- Unidentified speaker (UNKNOWN): As
- 23:16
- Unidentified speaker (speaker_1): I said earlier in the discussion with you, Gor, it's important that you assess every interaction you have with your consumer. Customer journey, consumer journey starting from onboarding till the time the money leaves the bank. Every step is important because we know customer is the most vulnerable element in the whole equation. So it's important that you assess every point.
- 23:36
- Unidentified speaker (speaker_1): So I think that will only increase
- 23:38
- Unidentified speaker (UNKNOWN): as
- 23:38
- Unidentified speaker (speaker_1): we have. more and more data. Number two on my wish list is definitely network intelligence in some form. There are pockets of development we can see where people are coming up with some sort of knowledge sharing. Data sharing, I think that will become that will get institutionalized, is my reading. We are helping a lot of banks in Singapore,
- 23:58
- Unidentified speaker (speaker_1): Malaysia, Hong Kong, you know, through our technology where we do this anonymized data sharing. We have better been quite quite optimistic that that will pick up big way in India. As
- 24:07
- Unidentified speaker (UNKNOWN): I
- 24:07
- Unidentified speaker (speaker_1): mean, if I'm banking with three banks and I have some challenge with my bank account, there is no way my bank can able to,
- 24:15
- Unidentified speaker (UNKNOWN): you
- 24:15
- Unidentified speaker (speaker_1): know, can figure out where the money is going and how do we stop that and protect the consumers, they will be much better equipped if they have access to the network intelligence. So these two things assess consumer. assess fraud at every step in consumer journey, and of course try and take benefit of network intelligence, which we need to create in some form in India. And I think
- 24:35
- Unidentified speaker (speaker_1): these two will sustain that 1 billion per month, which we expect UPI to hit hopefully in a few weeks. And next time we speak, we should see that number. So I I'm I'm pretty positive about that, Caro. Garo.
- 24:46
- Unidentified speaker (speaker_0): Absolutely. And thank you, Vishal, Vishhal, for taking the time out and dropping. by to have this conversation with us on such an important topic and subject, uh, which affects us all and which about which we need to ed continuously educate ourselves because we need to be one step ahead of the criminals who are
- 25:06
- Unidentified speaker (speaker_0): always lurking at all points in time. Thank you, Vishhal, for having this wonderful conversation with us.
- 25:12
- Unidentified speaker (UNKNOWN): Thank
- 25:12
- Unidentified speaker (speaker_1): you so much, Garo. Garab. Thanks a lot. Pleasure talking to you.
More from Moneycontrol Podcast
5323: Nvidia minus Jensen Huang, Sachin Bansal’s new AI startup and more
Episode discussion: Nvidia CEO exit, Sachin Bansal's AI investment, and Microsoft outage discussed
Speakers’ claims: The episode covers three main topics: Jensen Huang's departure from Nvidia and employee petitions for his return, Sachin Bansal's $20-30 million investment in a new agentic commerce startup, and a global Microsoft 365 outage affecting over 70 countries. Speakers note concerns about Nvidia's future direction under a new CEO, particularly regarding its Indian startup ecosystem investments. They also highlight the scale of the Microsoft outage, which disrupted services like Outlook and Teams without evidence of a cyberattack. Bansal's investment in Manav Garg's venture is presented as a strategic move into agentic commerce, with Garg aiming to raise $100 million total. The discussion frames these events as significant developments in the AI and tech sectors.
5322: Tech3 | India AI reaction to Jensen Huang affairs, Meta Muse coming to India next year and more
Episode discussion: Norwest raises $4B fund, India AI chip scrutiny, Meta Muse launch, BankBazaar stake sale
Speakers’ claims: Speakers discuss Norwest's record $4 billion fund focused on Indian industrial sectors, highlighting its strategic concentration in India. They analyze potential regulatory risks for India's AI chip imports following personal conduct issues involving Nvidia's CEO, emphasizing compliance burdens under US export rules. Meta's expansion into Indian payments via WhatsApp and the upcoming launch of its Muse AI model are covered. Finally, the episode notes BankBazaar's exploration of a majority stake sale at a discount to its last valuation, citing its revenue growth trajectory.