loading
loading
AI is all set to take centre stage at the Global Fintech Fest (GFF) in Mumbai this week, with product launches, live demos and industry conversations on how this emerging tech is reshaping financial services. But beyond the buzz, the bigger story is how deeply AI-powered systems are entering the Indian fintech stack, from customer onboarding and credit decisions to fraud detection, servicing and payments. To understand where the industry stands and where it is headed, Inc42 spoke with senior executives from Zeta, HyperVerge, TruCommerce and RevRag AI, who are participating in the seventh edition of GFF. The experts believe Indian fintechs have moved beyond the proof-of-concept phase: voice agents are handling customer interactions, AI is accelerating KYC and verification, fraud systems are becoming more sophisticated, and in-app assistants are beginning to execute tasks rather than simply answer questions. The next challenge is figuring out where these applications can deliver measurable value, and how far fintechs can take them without compromising trust, compliance and accountability. So, where is AI delivering the most immediate value for fintechs today? Let's unravel in this edition of The AI Shift… Fintech’s First AI Moves Are Now Operational Fintech companies are using AI to make existing processes faster and more efficient rather than letting it make financial decisions on its own. Kedar Parikh, the CPO of HyperVerge, an AI-powered identity verification and KYC platform, divides the sector’s adoption of AI into three broad phases. Digital onboarding: This includes KYC, face matching and liveness checks. Customers across banking, financial services, insurance and telecom can now be onboarded in minutes instead of days Fraud prevention: Faster onboarding also creates more opportunities for fake documents, synthetic identities and deepfakes to enter the financial system. This is pushing regulated entities to adopt AI for document forgery detection, biometric checks and deepfake prevention, although adoption remains a work in progress. AI-assisted decision-making: This phase is still at an early stage. AI can collect data, spot patterns and present useful inferences, but lending, insurance and other high-value decisions continue to require human judgement. This pattern is visible in customer service as well. According to Sivaram Kowta, the president of digital banking at Zeta India, a digital banking and card-issuing technology company, customer support, fraud management and underwriting are among the most active areas for AI adoption. Zeta’s AI agent resolves around 80% of customer-service calls for US-based subprime credit-card fintech Sparrow, Kowta said. Ashutosh Prakash Singh, the cofounder and CEO of RevRag AI, an AI voice and chat agent startup focused on banking and financial services, said fintechs have moved from testing voice AI and in-app agents to scaling them and measuring their return on investment. Singh said the next wave of deployment will go beyond contact centres, with fintechs, banks and non-banks such as PhonePe and Bajaj Financial Services exploring AI across in-app journeys, back-end workflows, branches, kiosks and other customer touchpoints. He expects these use cases to move from early deployments to broader scale across the BFSI sector by the end of this year and into next year. AI As A Paper Pusher The closer AI gets to a financial decision, the more difficult deployment becomes. In lending, for example, an AI system can collect customers’ financial history, classify information and apply existing rules but cannot become the final authority to approve a loan. Kowta said the RBI has made it clear that regulated entities cannot shift responsibility for credit decisions to an AI model. A bank must understand the decision and remain accountable for it. As a result, banks and NBFCs are using AI in underwriting while retaining a human at the final decision point. “AI is not doing something fundamentally new. It is a paper pusher, just an extremely capable or efficient paper pusher,” Kowta said. If every other step in the lending process is automated but the final credit review remains manual, the overall process may become faster, but it does not become fully autonomous. The compliance challenge is more complex. These constraints are not stopping adoption, but shaping the ecosystem. Currently, the main regulatory and compliance issues faced by Indian fintechs include: The DPDP compliance: AI systems must account for consent, data use and the right to erase personal information under the Digital Personal Data Protection framework. Data localisation: Banks and fintechs need clarity on whether financial information sent to overseas AI models remains in India. User-level authentication: Agents need permissions linked to the authenticated customer, not broad access to an institution’s systems. Payment authorisation: Agent-initiated payments in India require additional safeguards, including OTP-based authentication in several use cases. Human accountability: Regulated entities remain responsible for lending, underwriting and other critical decisions, even when AI performs the supporting work. According to Viren Inaniyan, cofounder and CEO of TruCommerce, a commerce infrastructure platform for AI agents, basic building blocks for agent-led commerce are already falling into place. These include AI platforms, commerce protocols, payment networks, gateways, merchant catalogues and checkout systems. India’s payment ecosystem also has safeguards in place to make transactions more secure. For instance, card details need to be encrypted and cannot be exposed to or stored by an AI agent. However, the bigger challenge is that very few people are using AI agents to make payments today. Agentic commerce accounts for an estimated 100 Mn transactions, or just 0.04% of UPI’s total transaction volume. This dataset is still too small for payment companies and merchants to understand how consumers behave when an AI agent is making purchases on their behalf. The Shift From Features To Foundations The conversations ahead of GFF point to a fintech ecosystem that is increasingly focused on scale, but not yet equally prepared for the cost of scaling. Kowta said less than 10% of bank budgets currently go towards AI, with spending often taking place project by project. However, he expects the approach to change over the next one year. Smaller banks may slow down their AI plans, while larger banks could commit significantly bigger budgets, potentially investing ₹500 Cr to ₹1,000 Cr at one time to build shared infrastructure, governance and AI capabilities rather than funding disconnected projects. For now, much of the industry is still in an AI computerisation phase. It is using the tech to perform existing tasks faster. The bigger change will come when banks redesign workflows around AI instead of adding an AI assistant to an old process. The message from the ecosystem is not that AI adoption is slow. It is moving quickly in customer service, onboarding and fraud prevention. But the next jump, allowing AI to influence decisions involving money, will depend on whether Indian fintechs can build the compliance and accountability layers that make those systems trustworthy. Top Stories From India & Around The World KRAFTON To Invest $250 Mn In Indian Startups: The South Korean gaming giant is committing an additional $250 Mn (₹3,300 Cr) to Indian startups across AI, robotics and deeptech over the next three to four years, taking its total planned India investment to over $1 Bn. Chairman Byung Gyu Chang announced the plan after a meeting with PM Narendra Modi. Navana.ai Raises ₹40 Cr: The Mumbai-based voice AI startup has raised ₹40 Cr ($4.2 Mn) in a Series A round led by existing investor and upGrad cofounder Ronnie Screwvala, with Antler India, Neysa.ai founder Sharad Sanghi and WestBridge's Sandeep Singhal also participating. The company will deploy the capital to deepen its Indic-language speech models and AI contact centre. OpenAI Launches GPT-6 Astra: OpenAI has released GPT-6 Astra, with a 1 Mn-token context and a headline API price of $10-$50 per Mn tokens. Because it is the first OpenAI model rated critical on cybersecurity, the cyber-sensitive capabilities are gated behind a trusted-access programme, and it is rolling out to enterprise customers first before wider ChatGPT availability. Nvidia’s PAIR Turns Idle PCs Into Personal AI Data Centres: Announced at IFA 2026, Nvidia's free, open-source Personal AI Router (PAIR) pools idle laptops, desktops and Macs on a home network to run local AI models for agentic workloads, using existing Ollama or LM Studio installs. The tool routes agent subtask requests to whichever machine is free, letting developers run multi-agent setups locally without cloud fees or privacy trade-offs. The Weekly Buzz: An Alien Mind OpenAI chief scientist Jakub Pachocki published a striking essay titled “An Alien Mind,” warning that increasingly capable systems may develop goals misaligned with human values. He argues that no lab has solved alignment and monitoring well enough to continue scaling at maximum speed for much longer. Pachocki calls for voluntary slowdowns, stronger third-party auditing, and international coordination, describing advanced AI as something that has ‘grown more than designed’. The essay has intensified debate inside and outside the lab. Supporters see it as a rare moment of candour from a frontier institution; critics note the tension between the warning and OpenAI’s continued rapid deployment. The contrast with Jensen Huang’s simultaneous “AGI has arrived” statement has only sharpened the discussion. Coming amid a week of cascading model releases, the essay reframes the central question of 2026: not only how capable the systems are becoming, but whether the institutions building them are prepared for the consequences of their own progress. Startup In The Spotlight: Antimattr Founded in 2025 by Sridipto Ghosh and Sirsho Chakraborty, Bengaluru-based Antimattr is building a voice-first computing system designed around memory and productivity. AI assistants are growing increasingly capable, but using them in the real world still means pulling out a phone or sitting in front of a screen. The problem surfaced while the founders were building voice agents: the AI worked, but noisy environments such as cafés, airports and open offices made voice interactions unreliable. Instead of putting a screen between the user and the machine, Antimattr pitches voice as the primary interface to AI. Its Project Mnemosyne combines a smart ring with a pair of voice-isolation earphones: the ring captures thoughts, records and summarises conversations and controls AI agents, while the earphones use six microphones and proprietary voice-isolation technology to enable clearer conversations and dictation even in noise. The startup counts Y Combinator, Nikhil Kamath's WTFund and gener8tor among its backers, with YC investing $500K. Antimattr operates in India's AI wearable market, projected to reach $19.1 Bn by 2033. As the company bets on a screenless AI future, its longer-term roadmap extends beyond Mnemosyne to a privacy-focused AI computer and, eventually, an AI-native vehicle. Prompt Of The Week What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work? Here’s the prompt Ram Khizamboor, COO at Indium, uses to run a pre-mortem on a client engagement, essentially forcing the AI to imagine failure: "Assume it's twelve months from now and this client engagement has failed. The contract wasn't renewed, the relationship soured, and leadership is asking hard questions about what went wrong. Write the post-mortem for it, answering these three questions in detail: What were the three most likely causes of the failure Which early signals did we ignore in the first 90 days of the engagement What would the client's CIO say, in hindsight, we should have seen coming Be specific and unsparing. Don't hedge or soften the findings just because this is a hypothetical exercise. Treat it the way a real post-mortem would be treated internally, with names, timelines, and decisions called out where relevant." Editor’s Note: Some prompts may need to be adjusted by users for best results or may not work as intended for certain users.
AI is all set to take centre stage at the Global Fintech Fest (GFF) in Mumbai this week, with product launches, live demos and industry conversations on how this emerging tech is reshaping financial services. But beyond the buzz, the bigger story is how deeply AI-powered systems are entering the Indian fintech stack, from customer onboarding and credit decisions to fraud detection, servicing and payments. To understand where the industry stands and where it is headed, Inc42 spoke with senior executives from Zeta, HyperVerge, TruCommerce and RevRag AI, who are participating in the seventh edition of GFF. The experts believe Indian fintechs have moved beyond the proof-of-concept phase: voice agents are handling customer interactions, AI is accelerating KYC and verification, fraud systems are becoming more sophisticated, and in-app assistants are beginning to execute tasks rather than simply answer questions. The next challenge is figuring out where these applications can deliver measurable value, and how far fintechs can take them without compromising trust, compliance and accountability. So, where is AI delivering the most immediate value for fintechs today? Let's unravel in this edition of The AI Shift… Fintech’s First AI Moves Are Now Operational Fintech companies are using AI to make existing processes faster and more efficient rather than letting it make financial decisions on its own. Kedar Parikh, the CPO of HyperVerge, an AI-powered identity verification and KYC platform, divides the sector’s adoption of AI into three broad phases. Digital onboarding: This includes KYC, face matching and liveness checks. Customers across banking, financial services, insurance and telecom can now be onboarded in minutes instead of days Fraud prevention: Faster onboarding also creates more opportunities for fake documents, synthetic identities and deepfakes to enter the financial system. This is pushing regulated entities to adopt AI for document forgery detection, biometric checks and deepfake prevention, although adoption remains a work in progress. AI-assisted decision-making: This phase is still at an early stage. AI can collect data, spot patterns and present useful inferences, but lending, insurance and other high-value decisions continue to require human judgement. This pattern is visible in customer service as well. According to Sivaram Kowta, the president of digital banking at Zeta India, a digital banking and card-issuing technology company, customer support, fraud management and underwriting are among the most active areas for AI adoption. Zeta’s AI agent resolves around 80% of customer-service calls for US-based subprime credit-card fintech Sparrow, Kowta said. Ashutosh Prakash Singh, the cofounder and CEO of RevRag AI, an AI voice and chat agent startup focused on banking and financial services, said fintechs have moved from testing voice AI and in-app agents to scaling them and measuring their return on investment. Singh said the next wave of deployment will go beyond contact centres, with fintechs, banks and non-banks such as PhonePe and Bajaj Financial Services exploring AI across in-app journeys, back-end workflows, branches, kiosks and other customer touchpoints. He expects these use cases to move from early deployments to broader scale across the BFSI sector by the end of this year and into next year. AI As A Paper Pusher The closer AI gets to a financial decision, the more difficult deployment becomes. In lending, for example, an AI system can collect customers’ financial history, classify information and apply existing rules but cannot become the final authority to approve a loan. Kowta said the RBI has made it clear that regulated entities cannot shift responsibility for credit decisions to an AI model. A bank must understand the decision and remain accountable for it. As a result, banks and NBFCs are using AI in underwriting while retaining a human at the final decision point. “AI is not doing something fundamentally new. It is a paper pusher, just an extremely capable or efficient paper pusher,” Kowta said. If every other step in the lending process is automated but the final credit review remains manual, the overall process may become faster, but it does not become fully autonomous. The compliance challenge is more complex. These constraints are not stopping adoption, but shaping the ecosystem. Currently, the main regulatory and compliance issues faced by Indian fintechs include: The DPDP compliance: AI systems must account for consent, data use and the right to erase personal information under the Digital Personal Data Protection framework. Data localisation: Banks and fintechs need clarity on whether financial information sent to overseas AI models remains in India. User-level authentication: Agents need permissions linked to the authenticated customer, not broad access to an institution’s systems. Payment authorisation: Agent-initiated payments in India require additional safeguards, including OTP-based authentication in several use cases. Human accountability: Regulated entities remain responsible for lending, underwriting and other critical decisions, even when AI performs the supporting work. According to Viren Inaniyan, cofounder and CEO of TruCommerce, a commerce infrastructure platform for AI agents, basic building blocks for agent-led commerce are already falling into place. These include AI platforms, commerce protocols, payment networks, gateways, merchant catalogues and checkout systems. India’s payment ecosystem also has safeguards in place to make transactions more secure. For instance, card details need to be encrypted and cannot be exposed to or stored by an AI agent. However, the bigger challenge is that very few people are using AI agents to make payments today. Agentic commerce accounts for an estimated 100 Mn transactions, or just 0.04% of UPI’s total transaction volume. This dataset is still too small for payment companies and merchants to understand how consumers behave when an AI agent is making purchases on their behalf. The Shift From Features To Foundations The conversations ahead of GFF point to a fintech ecosystem that is increasingly focused on scale, but not yet equally prepared for the cost of scaling. Kowta said less than 10% of bank budgets currently go towards AI, with spending often taking place project by project. However, he expects the approach to change over the next one year. Smaller banks may slow down their AI plans, while larger banks could commit significantly bigger budgets, potentially investing ₹500 Cr to ₹1,000 Cr at one time to build shared infrastructure, governance and AI capabilities rather than funding disconnected projects. For now, much of the industry is still in an AI computerisation phase. It is using the tech to perform existing tasks faster. The bigger change will come when banks redesign workflows around AI instead of adding an AI assistant to an old process. The message from the ecosystem is not that AI adoption is slow. It is moving quickly in customer service, onboarding and fraud prevention. But the next jump, allowing AI to influence decisions involving money, will depend on whether Indian fintechs can build the compliance and accountability layers that make those systems trustworthy. Top Stories From India & Around The World KRAFTON To Invest $250 Mn In Indian Startups: The South Korean gaming giant is committing an additional $250 Mn (₹3,300 Cr) to Indian startups across AI, robotics and deeptech over the next three to four years, taking its total planned India investment to over $1 Bn. Chairman Byung Gyu Chang announced the plan after a meeting with PM Narendra Modi. Navana.ai Raises ₹40 Cr: The Mumbai-based voice AI startup has raised ₹40 Cr ($4.2 Mn) in a Series A round led by existing investor and upGrad cofounder Ronnie Screwvala, with Antler India, Neysa.ai founder Sharad Sanghi and WestBridge's Sandeep Singhal also participating. The company will deploy the capital to deepen its Indic-language speech models and AI contact centre. OpenAI Launches GPT-6 Astra: OpenAI has released GPT-6 Astra, with a 1 Mn-token context and a headline API price of $10-$50 per Mn tokens. Because it is the first OpenAI model rated critical on cybersecurity, the cyber-sensitive capabilities are gated behind a trusted-access programme, and it is rolling out to enterprise customers first before wider ChatGPT availability. Nvidia’s PAIR Turns Idle PCs Into Personal AI Data Centres: Announced at IFA 2026, Nvidia's free, open-source Personal AI Router (PAIR) pools idle laptops, desktops and Macs on a home network to run local AI models for agentic workloads, using existing Ollama or LM Studio installs. The tool routes agent subtask requests to whichever machine is free, letting developers run multi-agent setups locally without cloud fees or privacy trade-offs. The Weekly Buzz: An Alien Mind OpenAI chief scientist Jakub Pachocki published a striking essay titled “An Alien Mind,” warning that increasingly capable systems may develop goals misaligned with human values. He argues that no lab has solved alignment and monitoring well enough to continue scaling at maximum speed for much longer. Pachocki calls for voluntary slowdowns, stronger third-party auditing, and international coordination, describing advanced AI as something that has ‘grown more than designed’. The essay has intensified debate inside and outside the lab. Supporters see it as a rare moment of candour from a frontier institution; critics note the tension between the warning and OpenAI’s continued rapid deployment. The contrast with Jensen Huang’s simultaneous “AGI has arrived” statement has only sharpened the discussion. Coming amid a week of cascading model releases, the essay reframes the central question of 2026: not only how capable the systems are becoming, but whether the institutions building them are prepared for the consequences of their own progress. Startup In The Spotlight: Antimattr Founded in 2025 by Sridipto Ghosh and Sirsho Chakraborty, Bengaluru-based Antimattr is building a voice-first computing system designed around memory and productivity. AI assistants are growing increasingly capable, but using them in the real world still means pulling out a phone or sitting in front of a screen. The problem surfaced while the founders were building voice agents: the AI worked, but noisy environments such as cafés, airports and open offices made voice interactions unreliable. Instead of putting a screen between the user and the machine, Antimattr pitches voice as the primary interface to AI. Its Project Mnemosyne combines a smart ring with a pair of voice-isolation earphones: the ring captures thoughts, records and summarises conversations and controls AI agents, while the earphones use six microphones and proprietary voice-isolation technology to enable clearer conversations and dictation even in noise. The startup counts Y Combinator, Nikhil Kamath's WTFund and gener8tor among its backers, with YC investing $500K. Antimattr operates in India's AI wearable market, projected to reach $19.1 Bn by 2033. As the company bets on a screenless AI future, its longer-term roadmap extends beyond Mnemosyne to a privacy-focused AI computer and, eventually, an AI-native vehicle. Prompt Of The Week What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work? Here’s the prompt Ram Khizamboor, COO at Indium, uses to run a pre-mortem on a client engagement, essentially forcing the AI to imagine failure: "Assume it's twelve months from now and this client engagement has failed. The contract wasn't renewed, the relationship soured, and leadership is asking hard questions about what went wrong. Write the post-mortem for it, answering these three questions in detail: What were the three most likely causes of the failure Which early signals did we ignore in the first 90 days of the engagement What would the client's CIO say, in
document fetched 2026-09-10 02:58 · canonical https://inc42.com/features/indian-fintech-moves-beyond-ai-demos