AI was everywhere at GFF 2026. Full automation was not

The Global Fintech Fest (GFF) 2026 concluded in Mumbai on September 11, bringing together financial institutions, fintech companies, regulators and technology providers from more than 70 countries. In keeping with this year’s theme, agentic AI dominated conversations on and off the stage.

By the time the conference ended, the conversation had spilled beyond the venue. At a post-event mixer, a restaurant ticker screen flashed: “ye kya AI AI laga rakha hai”. It was a joke, but it captured the mood. AI was everywhere.

Beyond the buzz, however, conversations with founders and senior executives across the tightly regulated banking, financial services and insurance (BFSI) sector were increasingly practical. The debate was less about whether AI would be used and more about where an agent should stop, which model fits a task, and what adoption means beyond applications and data centres.


100% automation not on the cards

‘Humans are not going out of the loop’ was one of the biggest takeaways from GFF 2026. Several discussions prompted the question of whether agentic AI will facilitate end-to-end transactions and decision-making. According to Perfios Group CEO Nitin Chugh, the answer is no. Parts of the workflow, such as low-level basic tasks (code generation) that are repetitive, could get automated, but humans will always make the final call.

However, the role of an underwriter will change. Earlier, if they could process 10 loan applications, AI can enable them to process significantly more applications, Chugh suggested.

Also Read: BFSI sector in no hurry to adopt complete automation: Perfios Group CEO

Meanwhile, HyperVerge’s business-loan underwriting suite shows what that division of work can look like. Its agents pull information from bank statements, goods and services tax filings and income-tax returns, flag gaps, run background checks and assemble credit notes.

“Today, a credit underwriter examines bank statements, GST filings and income-tax returns, and then prepares a note for the team. If preparing that note takes two hours, AI can create it in about a minute,” HyperVerge cofounder and CEO Kedar Kulkarni said. The lending decision, however, remains with the credit manager.

The industry is also increasingly aware that compute power is not free. It is imperative to understand which tasks need to be automated and which can be done manually to save both time and money. Companies such as KFintech, Zeta, and Perfios are adopting this route by only adopting agentic workflows in processes where absolutely mandatory, to minimise their AI spend as well.

Jobs change, but hiring continues

‘Will AI take our jobs?’ is another question that you will find as a part of every conversation on the subject. At GFF, the answer was fairly and surprisingly optimistic.

Revolut India’s CEO Paroma Chatterjee said ‘making people redundant due to AI is a no-win’. The company plans to hire about 1,500 people in India in 2026. She said it is critical to create AI-targeted jobs and build workflows that have AI embedded in the design thinking.

However, roles built around repetitive tasks are likely to see the biggest disruption, while employees in more senior positions will increasingly be expected to handle higher-value work rather than routine processing.

Zeta's APAC CEO Ramki Gaddipati told us that the firm's coding teams have moved away from basic code-writing towards reviewing AI-generated code, with AI agents handling the first level of coding.

Smaller, specialised models find their moment

For financial companies operating under strict regulatory and data-security requirements, the attraction of AI is increasingly about using the right model for the right job rather than simply deploying the biggest available large language model (LLM).

This is creating interest in smaller, specialised and domain-specific models that can be deployed for particular use cases, while giving companies greater control over sensitive data and costs.

For instance, KFintech, the country’s largest Registrar and Transfer Agent (RTA), has used open-weight models from Alibaba’s Qwen to build a more specialised AI system for its operations. Since the company deals with large volumes of sensitive data, it believes that having its own SLM is more beneficial.

Mastercard's AI Garage, meanwhile, has built a large tabular model (LTM) for transaction data, allowing the company to work with characteristics specific to its payments data.

Revolut India has also developed PRAGMA, a family of transformer-based models designed to analyse customers' financial histories as temporal signals, using NVIDIA's Llama-Nemotron-Embed-1B-v2.

Voice AI finds its way into finance

Voice AI was another prominent theme at GFF, but the conversation had moved beyond scripted sales calls. Companies are increasingly looking at voice as an interface through which customers can ask questions, buy financial products, and seek assistance without navigating complex applications.

One example came from ONDC. The network’s chief business officer for financial services, Hrushikesh Mehta, said a network participant is testing a voice assistant that helps autorickshaw drivers in Bengaluru invest as little as Rs 20 in mutual funds.

Arrowhead AI cofounder and CEO Devyani Gupta described contextual voice agents being used in lending, while Policybazaar CTO Bibhu Krishna said the insurance platform is using voice systems for outbound calls and piloting an open-ended inbound bot for motor insurance.

Also Read: Voice AI startup Arrowhead eyes fresh funding to accelerate international expansion

The interface could lower the cost of reaching customers, especially in smaller towns and across local languages.

But the technology also brings a new set of risks. If AI can convincingly recreate a person’s voice, voice-based verification becomes less reliable.

Krishna pointed to attempts involving proxy voices and images during life-insurance video KYC. He also said fully autonomous policy issuance is not yet on the table, with people still needed for nuanced questions, including those involving claims ratios. Voice may become the entry point to a financial workflow, but identity checks and complex decisions will still need human escalation.

Adoption is high. Depth is another question

BCG’s flagship GFF 2026 report, Balance: Thriving in the Age of AI, points to India’s high workplace adoption. BCG’s AI at Work 2026 survey found that 70% of Indian frontline AI users said the tools saved them at least one full workday a week, the highest share among the markets covered, while 96% said AI had changed the skills required in their roles.

But widespread use does not necessarily mean India is building capabilities across the technology stack.

Also Read: India’s banks could turn legacy tech into an AI advantage: BCG

Conversations with fintech executives at GFF suggested that much of the adoption is happening either at the application layer, where companies are racing to build workflows and services on established models, or at the infrastructure layer, where billions of dollars are being committed to data centres and computing capacity.

This raises two problems: India has a relatively small number of companies building India-focused foundation models, including Sarvam and BharatGen, while the country's semiconductor manufacturing ecosystem is still developing.

For the BFSI sector, however, the immediate AI opportunity may not depend on India building a frontier model of its own. The more immediate question is how effectively financial companies can use existing models, while keeping data secure, controlling costs, and retaining human oversight over the decisions that matter.