Access Proceedings Report
Yutadhi joined hands with Thoughtworks India to jointly host an in-person Roundtable on Agentic AI in Payments & Financial Services for India, on January 21st 2026. We are pleased that this roundtable has been approved by the Ministry of Electronics & Information Technology (MeITY) as an official Pre-Summit Event, affiliated under the India – AI Impact Summit 2026 (held in February 2026).
The roundtable was organized into three segments. The first segment saw short invited staging presentations by Bharani Subramaniam, CTO IME, Thoughtworks, Charu Dutt Sharma, Chief Strategy & Sales Officer, AbleCredit, Aneesha Meka, Founder's Office, Gnani AI, Dr. Bijoy Jose, Principal Consultant, Abilytics and Achyutha Sharma, User Research India. These were following by two discussion segments covering the themes of Customer Engagement & Personalization, and Risk Governance, respectively.
Thoughtworks1 provided a sharp and noteworthy articulation of why the new AI-era is different from past eras and therefore ought to be understood differently (See table below).
Eg: a shopping cart
Eg: big data, cloud, IoT
Eg: Having data that provides a 360-degree view on the customer implies 'gigabytes' or 'terabytes' of such data, but the decision-maker, say the head of retail lending, tracks dashboards that will consume only 'megabytes' of data.
Eg: Agent to agent interactions
Eg: An individual's life-time data on spends is analysed to pick up a pattern on his/her unique behavior, and this is used to arrive at a better-priced loan to the individual (who is willing to wait an additional few seconds).
A majority of participants in the roundtable were from entities that were not directly risk-holding, ie, they were not banks, NBFCs or insurance companies. Hence, the discussions overwhelmingly captured views that did not reflect risk-holder entities, for the most part. Hence, this proceedings report captures the views from their perspective. They noted the following:
These are elaborated in the attached report that discusses the background for the roundtable and a summary of the proceedings that transpired.
The participants also came up with use cases where Agentic AI can be a potential gamechanger –
A. Use cases where there are coordination-failures
- Real-time fraud risk-scoring & actioning
- Getting fuller credit histories of borrowers:
B. Reimagining branch banking and Digital Banking Units with context-aware & conversational AI agents
C. Transforming financial literacy programs
The Solutions Space
A. Home-grown Small Language Model (SLM) capabilities: There was strong support for the need to develop SLMs (that can even be run on a CPU) to counter the significantly higher costs of using LLMs, which in turn has led to strategic decision-making that only aims for short-term and low-impact wins (For instance, bringing in AI agents to aim only for efficiencies that human agents currently achieve).
B. Build own tech-stacks that are context-specific to your industry: A few participants emphasized this as an important step needed to unlock value from AI systems.
C. Use 'user-centric' 'problem-first' approaches: Bring systematic and comprehensive focus to user research, both in, a) deeply understanding customer intent and goals that are being satisfied by AI product / process design, and b) establishing how strong or fragile customer trust is in existing products, to then predict how an AI interface can rupture existing levels of trust / distrust.
D. Certification Regimes: Mandatory certifications by say, CERT-In2, provided to AI solutions within regulated entities, and to third-party vendors can bring down the costs incurred by both types of parties in repeatedly evaluating solutions, the time spent and the frequency of such evaluations.
E. In addition to third-party certifications, there is a need to develop 'continuous' monitoring mechanisms for a bunch of metrics that can signal the robustness of the AI models via publicly available dashboards. For instance, the regulator can come up with 'red', 'amber', 'green' colour coding based on the potential impact of the risks in each model, and require every model to go through this monitoring process.
F. Risk-proportionate approaches to regulating and supervising AI in BFSI: 'How much oversight should the regulator have', should depend on the potential impact the AI application can have on every user, and not just on the risk-management systems or balance sheets of the provider (the more conventional micro-prudential approaches). User-impact must be a component to consider in determining the degree of risk (from both probability of risk and the severity of impact perspectives).
To conclude, the roundtable was able to capture a shared, forward-looking understanding of how AI agents, agentic AI and autonomous systems are being viewed by actors in the BFSI ecosystem in India, and how these should be responsibly designed, governed, and deployed within the Indian BFSI industry. The timing for this is apt given that we are already seeing some early deployments in banking, payments and in insurance.