India's First Purpose-Built GenAI Model for Financial Statements: Meet Bindu FRS 1.0

Mumbai (Maharashtra) [India], August 14: Indian enterprises have stopped asking "does this AI work?" and started asking "how fast can it work?" According to the EY India C-suite GenAI Survey 2025, 91 per cent of enterprise leaders now rank speed of deployment as the single biggest factor in their AI buy-vs-build decisions, and nearly half of Indian organisations already have multiple GenAI use cases running live in production. Pilots and proofs-of-concept are no longer the endpoint - production is. Recognising this shift, Decimal Point Analytics (DPA), a Mumbai-headquartered digital transformation company, has built India's first GenAI model purpose-built for a single, unglamorous but high-stakes task: converting unstructured financial disclosures into standardised, machine-readable data. Bindu FRS 1.0, developed and trained entirely in-house, does this one thing only - and it's built on a bet that in Indian enterprise AI, narrow may now matter as much as broad.

Shailesh Dhuri, CoFounder and CEO of Decimal Point Analytics

Why are financial statements a hard problem for general-purpose AI?

That shift matters for a specific, unglamorous but high-stakes corner of enterprise AI: financial statement analysis. Financial data aggregators, rating agencies and research providers process tens of thousands of company filings every year - annual reports, interim filings, prospectuses - each structured a little differently, each prone to new or non-standard line items that general-purpose AI models tend to get wrong. At that volume, even a small error rate compounds into bad data, compliance exposure and hours of manual correction. It's exactly the kind of narrow, repetitive, high-precision task that broad, do-everything language models were never optimised for.

Bindu FRS 1.0 is distinct from horizontal Indian foundation models, BFSI workflow copilots and payments-operations tools, none of which were designed for this specific task.

"India has built strong credibility in horizontal AI. The next frontier is vertical AI - models built deep into a single domain, with proprietary data as the moat and continuous improvement as the engine," says Shailesh Dhuri, Co-Founder and CEO, Decimal Point Analytics."Frontier commercial LLMs are extraordinary general-purpose tools, but for high-volume, high-accuracy, closed-domain work - like mapping the financial disclosures of tens of thousands of companies - general-purpose is the wrong architecture."

In Decimal Point Analytics -run internal testing - not independently verified - across 100 financial filings from UK and Canadian issuers, Bindu FRS 1.0 achieved 98.79 per cent first-pass accuracy on financial statement mapping, a 77 per cent reduction in first-pass errors compared with a frontier commercial LLM tested on the same task, with markedly fewer hallucinations on newly reported data points.

Commercial LLMs hallucinate most of the data points that matter most - the newly reported, non-standard line items in real company filings. Bindu FRS 1.0 removes that failure mode by narrowing the model's world to a single well-bounded task and teaching it from corrections, not from scale. The accuracy curve keeps improving; the unit cost stays flat.

How is it deployed?

Beyond domain-specific training, it points to a continuous-learning loop - analyst corrections feed directly back into the model, something static commercial APIs don't offer - and deployment on Decimal Point Analytics's own infrastructure, which removes the per-query costs that come with high-volume commercial LLM use. The model already lives within Decimal Point Analytics's managed data services, supporting aggregators, rating agencies and research providers.

Shailesh Dhuri says, "Bindu FRS 1.0 is meant to be the first of several Bindu FRS model built for other specialised financial domains - a bet that in India's next phase of AI adoption, narrow models built for one job may matter as much as the broad ones built for every job. Whether that edge holds as filing formats and reporting standards keep evolving remains to be seen."

Whether that edge holds as filing formats and reporting standards keep evolving remains to be seen - but for a market that has stopped rewarding novelty for its own sake and started rewarding tools that deploy fast and run reliably at scale, DPA is betting the answer is yes.