[Download this note]
The Paradox at the Heart of BFSI
BFSI is the infrastructure on which almost every other industry runs, but it is also one of the most layered and operationally opaque systems built. Even as technology has advanced and created a positive outcome by reducing the drudgery and manual intervention in banking operations, the complexity in BFSI technology has increased multifold. The industry has now evolved into a digital-first and real-time ecosystem with multi-channel customer journeys, stringent regulatory intensity, and constant threat of cyber risk.
At Mela, we have spent the last several years investing at the intersection of technology and this structural complexity. Two of our investments from Fund-I anchor this thesis:
- VuNet, which provides AI-driven observability and real-time analytics across BFSI digital infrastructure
- iTuring, whose platform delivers predictive models and autonomous agents which are BFSI native
What we have learned from building alongside them shapes our thesis for investing in AI-BFSI.
Complexity and Scale: Two Sides of the Same Problem:
The underlying technology systems in BFSI have only become more complex with time. They are layered with decades of legacy infrastructure and fragmented workflows, further complicated by regulatory overlays over time. Any financial interaction often passes through multiple systems, teams, and checkpoints before completion. This complexity is inefficient, expensive and error prone.
The scale is mind-boggling. Billions of transactions run continuously alongside fraud signals, credit decisions, and compliance checks with zero tolerance for failure. A few minutes of system downtime creates regulatory exposure and reputational damage. The scale also reflects in the technology architecture; the systems are built with low latency, redundancy, dynamic scaling, and regulatory compliance in mind.
Together, they create the conditions which make AI a boon for BFSI. AI can be the abstraction layer capable of operating coherently across fragmented systems and delivering real-time intelligence at the volume BFSI demands.
VuNet makes this concrete – Its platform processes over 20 billion transactions per month, more than $900 billion in transaction value while delivering real-time observability across 100+ business journeys inside large banks. Operating reliably at this scale is not a feature. It is the moat.
Investment Lens: From AI Tools to AI Infrastructure
A significant portion of investments in AI for BFSI has followed use cases such as fraud detection, KYC automation, credit scoring, customer service bots, etc. These are real and valuable, but they represent AI as a series of point solutions. We are investing in AI platform plays where AI can become the core infrastructure and be woven into the operating model of financial institutions.
We segment this infrastructure into three distinct layers as discussed below. Our preferred potential startups ideally stitch all the three layers as part of their unified offering.
Layer One: AI as Operational Infrastructure
The first layer is visibility. BFSI institutions are sometimes flying blind on their own digital operations. They know when something has gone wrong because customers (internal and external) complain or transactions fail. But they lack real-time and root-cause intelligence that tells them why something failed, which business journey was affected, and what the revenue and regulatory impact is before it becomes a crisis.
AI systems that provide this layer such real-time observability and business impact intelligence tied directly to operational outcomes are becoming non-negotiable infrastructure. When these systems fail, the cost is technical, reputational and regulatory.
This category extends beyond observability. It includes:
- Voice-led platforms handling customer interactions at scale
- SRE automation ensuring system reliability and uptime
- AI-driven regulatory operations for real-time compliance and audit readiness
- Automated incident detection and resolution
- Fraud monitoring systems that act in real time rather than in batch
VuNet sits at the core of this layer – It’s solution stack provides for observability across transactions, applications, and infrastructure. For instance, a single OTP transaction completed by the customer in under a few seconds silently traverses 17+ discrete steps across five vendor layers and in a distributed architecture where each of these systems runs across independent nodes and data centres, the complexity compounds further.
VuNet is the system that stitches this journey into a single coherent view across architectures and vendors, helping you pinpoint the exact point of failure in real time. Vunet’s production deployment across the largest banks & mutual funds in India is evidence that this category has moved from pilot to infrastructure.
Layer Two: AI as Decision Fabric
If the first layer keeps institutions running, the second layer makes them intelligent. AI as the Decision Fabric refers to systems built within organisation’s workflows in assisting decisions as AI is continuously making, optimising, and executing them in real-time across customer acquisition, underwriting, renewals, and collections.
The next evolution is not just better models, but autonomous systems. The leading vertical AI players ride the model curve. As foundation models improve, these companies get better automatically. They compound proprietary workflow depth, data rights, and institutional integrations on top of models that keep improving beneath them. That compounding is the moat.
Platforms should move from building systems and workflows that recommend actions to powering autonomous agents that can act and close the decision loop – in other words, AI-centric platforms should build & deploy agents and exercise oversight of agentic execution while protecting enterprise data.
iTuring’s platform has moved beyond ML lifecycle automation into agentic AI. The company is building autonomous agents to solve last-mile challenges in customer acquisition, collections, underwriting, and fraud prevention, with compliance embedded in every layer rather than bolted on after.
Layer Three: AI for Trust and Governance
The first two layers depend entirely on the third. Banks will not deploy AI at scale if they cannot explain it to a regulator.
Both VuNet and iTuring solve a trust problem as much as a performance problem. Every deployment required passes through risk committees, model validation teams, and regulatory review. The institutions that bought from them did so in part because the AI could be audited, explained, and governed.
The white space here is significant. Model risk management, explainability tooling, and AI governance platforms built specifically for RBI, SEBI, and FCA-regulated environments remain underinvested relative to their importance. The moat potential is high precisely because the bar for institutional trust is so difficult to clear.
The Founder Profile That We Believe Wins in BFSI
BFSI Needs Domain-First Founders, Not Just AI Talent – In BFSI, one must first understand the complexity to abstract it. Based on the VuNet and iTuring playbooks, we believe that the right founders for us share a specific profile:
- Domain depth first, AI second – Both Ashwin Ramachandran (VuNet) and Suman Singh (iTuring) come with 10-20+ years in financial services or enterprise software before founding. They have sold to the CRO, CTO, and CFO. They have seen production deployment in large and complex enterprise setups.
- Patience calibrated to the sales cycle – These are not Product-Led Growth businesses. The right teams have already closed 2-3 marquee logos before raising a Series A. They understand that procurement in BFSI takes 12 to 18 months.
- Complementary teams built for distribution – The founder teams should have engaged with the BFSI community, can speak their language and earn the institutional trust in the process. Succeeding here requires AI expertise but perhaps more importantly, it demands systems thinking, architectural depth, and strong domain understanding. Great technology needs strong distribution and focussed execution to scale. Founder teams who can bring complementary skillsets in product, operations and distribution (access) set themselves up to win more consistently than teams skewed heavily towards technology and tools.
BFSI is not just another vertical for AI. It is where AI will be stress tested, regulated and scaled to its limits. The institutions that get this right will be structurally more defensible.
At Mela, we are looking for founding teams who want to address these complex problems at scale. We value ideas that can develop compounding moats over time and generate sustainable profits. Teams can operate across any part of the BFSI stack, as long as they combine strong execution, deep problem understanding, systems-level thinking to build for scale and trust.
Author:
Suhani Doshi, VP – Investments, Mela Ventures
