Why Context, Not Just Data, Will Define the Future of AI in Finance

Artificial intelligence may be getting faster, sharper, and more powerful by the day, but power without perspective can be dangerous. In the world of personal finance, where trust and accuracy define livelihoods, raw data alone is not enough. What really matters is context, the ability to understand not just the numbers but the story behind them.

In this conversation with CIOL, Jayanth Saimani, Distinguished Software Engineer and Director, Intuit, reflects on why domain expertise and human judgment must be woven into AI’s very fabric. From rethinking the role of engineers in the GenAI era to placing humans at the heart of machine learning, he argues that the future of responsible AI will be shaped less by scale and more by wisdom. Excerpts.

AI is becoming smarter, but is it becoming wiser? What’s the cost of building powerful models without sectoral context, especially in something as sensitive as personal finance? Is that what you mean when you say, 'context at the core of AI'?

Raw intelligence in AI and its ability to crunch numbers and process data is only one part of the equation. What it fundamentally lacks is wisdom, which comes from context. In areas like personal finance, building powerful models with deep domain knowledge is critical. The challenges range from misinterpretation of data to regulatory oversights that directly affect value for customers. That’s why at Intuit, we put "context at the core of AI."

This means moving beyond generic datasets to build specialised Financial Large Language Models (LLMs) trained on decades of anonymised financial expertise. It's about understanding the interconnected journey of our customers across our ecosystem—from the freelancer managing invoices in QuickBooks to that same individual filing taxes with TurboTax, to them monitoring their financial health on Credit Karma. This deep integration of context transforms our AI from a reactive calculator into a proactive financial assistant. For example, in our work with QuickBooks' financial data Q&A, we found that training on generic datasets was insufficient as it often lacked subtle accounting nuances.

When we integrated deep domain knowledge, defining terms like COGS, revenue, or net profit margin, the system's accuracy and reliability improved dramatically. Our AI doesn't just see a transaction; it understands the "why" behind a customer's financial goals, helping them make smarter decisions to power their prosperity. For us, context transforms AI from being just "smart" to being genuinely useful to the customers we serve.

You speak the language of code and the language of context. Which is harder to teach AI? As someone deeply involved in building AI-native platforms like GenOS, where have you seen the biggest friction—technical constraints or the challenges of encoding human, domain-specific judgment into machines?

Intuit is in a unique position to deliver exceptional value to our customers by combining human intelligence and artificial intelligence. We leverage our vast data assets, extensive network of human experts, and advanced AI capabilities to create AI-native "Done-for-you" experiences.

Our AI agents automate complex financial tasks, working alongside human experts to provide real-time insights and improve cash flow. This synergy of human and machine intelligence helps businesses grow by handling workflows across accounting, payments, and financial analysis, saving them up to 12 hours a month. This approach is highly effective, with 78% of customers saying our AI makes it easier to run their business.

Teaching code to AI is relatively straightforward. Machines can quickly master syntax, process massive datasets, and automate workflows with remarkable efficiency. These technical challenges, while complex, are largely solvable through engineering innovation and computational power. The real difficulty, however, is teaching AI the language of context. This is the language of human experience, where the greatest friction lies in encoding deep, domain-specific judgment into a machine. Unlike code, context isn't binary; it requires experience, interpretation, and the ability to navigate ambiguity in ways that AI struggles to emulate.

This is possible because we focus on teaching AI the language of context, not just code. While coding is straightforward for machines, encoding human judgment and navigating ambiguity is much harder. Our AI-native platform, GenOS, was built to solve this. It serves as a bridge between machine intelligence and human expertise, allowing our domain experts to embed their real-world understanding directly into the AI’s core logic. This ensures our AI is not only smart but also practical, responsible, and genuinely trustworthy.

We often hear about the ‘human in the loop,’ but is that loop truly closed? For instance, when building automated systems in regulated industries, how do you ensure that domain experts don’t just second the outputs, but actively shape and challenge the system’s logic and values?

At Intuit, we're evolving the concept of a 'human in the loop' into a 'human at the heart,' fostering a true partnership between human and machine. This vision transforms passive oversight into active collaboration, where our domain experts are engaged as co-creators from the very beginning. They're central to building responsible, context-aware solutions by embedding their deep financial wisdom into the AI system's core.

These experts are integrated throughout the entire lifecycle of our AI-powered financial tools. They actively guide model training, fine-tune predictions, and stress-test algorithms against sector-specific scenarios in accounting, payments, and cash flow management. This approach ensures our AI isn't just a predictive tool but a customer-centric solution you can trust.
By combining engineering innovation with this profound domain expertise, we create a dynamic cycle of human judgment and machine intelligence. This strategy allows us to deliver exceptional value and recently led to our groundbreaking virtual team of AI agents designed to fuel growth for businesses.

Our teams leverage a suite of AI monitoring and control mechanisms that enable real-time intervention and continuous learning from expert guidance. This constant dialogue ensures every human interaction makes the entire system smarter, more reliable, and more attuned to customer needs. This is how we ensure our AI systems uphold the highest standards of trust, reliability, and regulatory compliance, delivering practical, high-value outcomes for enterprises.

In the age of GenAI, what does craftsmanship in engineering look like? With tools like GenStudio and GenUX abstracting model development for domain experts, how is the role of engineers evolving? What does ‘good engineering’ mean when much of the heavy lifting is automated?

In the age of GenAI, craftsmanship in engineering is being redefined. It's no longer just about writing every line of code or building models from scratch, but about architecting robust, extensible systems that empower others to innovate. The very soul of engineering is transcending code to become the art of architecture. The measure of excellence is no longer found in the meticulous construction of every model, but in the visionary design of systems that empower domain experts to innovate.

With tools like GenStudio and GenUX abstracting complexity, the engineer’s role isn't diminished but elevated. They evolve from builders of applications to architects of innovation ecosystems. Their modern craft lies in forging the essential guardrails and scalable foundations that make widespread creativity possible.

"Good engineering" today is about more than technical efficiency. It's about anticipating complexity, and creating frameworks that balance speed of delivery with trust, compliance, and long-term sustainability, enabling safe experimentation. Ultimately, today's master engineers are the enablers of a new generation of creators, shaping a future where AI is not only powerful but also profoundly responsible and accessible to all.

If you had to choose—more data or deeper context? Particularly when scaling AI in fintech, what ultimately delivers more long-term value: massive datasets or tightly integrated domain understanding? And are they always complementary?

If I had to choose, deeper context wins every time. More data doesn’t guarantee better outcomes if the system doesn’t understand the meaning behind it. In fintech, a trillion rows of transactions mean little without the domain expertise to interpret what constitutes revenue, risk, or fraud. Our QuickBooks journey proved this—integrating accounting context turned raw data into accurate, actionable insights. That said, context and data are not adversaries, they’re complementary. Large datasets provide breadth, while domain understanding gives depth. It’s the intersection of both that delivers AI systems customers can trust and rely on.