As the year winds down, it’s natural to jump straight into what’s next. But before we get caught up in predictions for 2026, I want to look at the ground we’ve covered in the last twelve months.  

AI has officially and authoritatively moved from our newsfeeds to our daily workflows. When you consider that over 18 billion messages are now being sent to ChatGPT every week, it’s clear that this technology has grown into an operational necessity.  

We’ve moved past the ‘what if’ stage and into a reality where AI is solving tangible, real-world problems. While experimentation continues in the background, AI is already proving to be a vital engine in driving a leaner, faster, and much more predictive financial ecosystem.  

Here is how this shift is playing out:  

From manual ‘reading’ to contextual understanding 

Earlier automation would often be limited to data extraction, still requiring teams to manually sort, verify, and cross-reference information. Intelligent Document Processing (IDP) goes beyond just digitising text.  

By contextually validating data against internal records and flagging discrepancies in real-time, IDP turns messy, unstructured files into clean, actionable intelligence. For underwriting, IDP feeds structured intelligence directly into the decision engine. It turns a week-long manual review of unstructured PDFs into a faster process, allowing lenders to move from applications to qualified offers before the customer has the chance to look elsewhere.  

Launching credit products in days 

Typically launching a new credit product involves a gruelling 3-to-6-month engineering marathon. Even minor updates often get buried in multi-team coordination, delaying time-to-market. 

We are now seeing a shift toward Gen-AI powered, no-code platforms that allow business and product teams to design and deploy these products using natural language and visual configurations. This removes the heavy lifting of backend builds and API stitching, allowing teams to: 

• Launch new products in a few days, not months 

• Deploy a single configuration across mobile, web, and partner environments  

• Use built-in A/B testing and behavioural analytics to fix user drop offs instantly 

The result? A lender launching an education loan in under three weeks, achieving over ₹300 crores in disbursals within just three months. 

Stability > speed 

The 2010 "Flash Crash" was a stark reminder that speed without intelligence is a liability. It took only minutes for $1 trillion to vanish, driven by rigid scripts that couldn't detect or adapt to data errors.  

Unlike static code, AI models constantly learn from market data and adapt to circumstances in real time. By processing global news, historical patterns, and sentiments simultaneously, these AI-driven models ensure that speed doesn’t come at the cost of stability. 

When fraud stops looking like fraud 

As loan volumes surge, so does the sophistication of fraud. Modern AI-driven systems are meeting this challenge by moving beyond ‘if/then’ logic. By layering machine learning models with dynamic credit scoring and velocity monitoring, these systems analyse vast datasets in real time to identify patterns invisible to the human eye. Blocking known threats is just the start; they use predictive logic to deter outliers and segment customer behaviour during the onboarding stage.  

This allows lenders to automatically approve, review, or reject applications with precision, stopping fraudulent activities before a single rupee is disbursed.  

Compliance from the get-go 

Legacy compliance isn't just slow; it’s a massive bottleneck. Every time a new regulation drops, the whole system halts while teams scramble to catch up. We’re finally seeing AI fix this by: 

• Performing large-scale data comparisons across disparate platforms to ensure consistency, 

• Automating complex regularity reporting in seconds by utilising generative prompts aligned with specific mandates, and 

• Proactively testing data to identify anomalies and outliers based on business rules before they escalate into violations. 

Beyond customer service chatbots 

McKinsey puts the potential value of AI in banking at a staggering $1 trillion, and a huge chunk of that comes down to one thing: personalised, predictive, human-like customer service. 

We are moving beyond basic chatbots into an era of Agentic Service. Onboarding stops being a series of forms and becomes a conversation where the AI handles the KYC and document checks in background. Beyond the first touchpoint, AI uses deep behavioural data to move from generic support to hyper-personalised financial coaching, predicting a user’s needs before they even ask. 

Next up is Agentic Commerce, where AI transacts on the user's behalf. This is where AI stops just making suggestions and starts actually making moves –– transacting on your behalf. We’re seeing the rise of secure spaces where an AI agent can handle the entire shopping trip, from finding the right item to hitting "buy." Because these agents live inside the payment networks we already trust, they can take over the boring stuff like checkouts and price comparisons. 

Looking forward 

For all the ways AI is making things better, it’s also making it a lot easier for things to go wrong. We’ve already seen how easily rigid systems break, and how tough it is to catch fraud when it’s designed to look flawless.  The reality is that the tools for creating friction and fraud are simply moving faster than the systems meant to stop them.  

The question for 2026 is simple: Are you investing an AI-driven system that can see what you can’t, or are you still waiting for a crisis to force your hand? 

I will see you in the next year.  

Cheers,  

Srijan Nagar
Co-founder  
FinBox