How Dezerv built post-call analytics for wealth management with Sarvam
How Dezerv built post-call analytics for wealth management with Sarvam
A Relationship Manager conversation is where trust is built or lost. For years the most important thing in our business was also the thing we could see the least of. Sarvam turned thousands of those conversations into something we can learn from.
Sandeep Jethwani
Co-founder, Dezerv
Background
Wealth in India is growing faster than the expertise available to manage it. Dezerv was founded in 2021 to close that gap. The firm designs customised investment portfolios for its clients, offered through its SEBI-registered Portfolio Management Services starting at ₹50 lakhs, along with alternative investment funds for those seeking exposure beyond traditional markets. As of 30th June 2026, Dezerv manages over ₹17,000cr+ across PMS, AIF and distribution with a team of 650+ across five cities.
At the centre of the model is the relationship manager, and every client is assigned one. RMs explain portfolio decisions, hold clients steady through volatile markets, and carry the conversations on which long-term trust is built. Those conversations happen over video conferencing, in whichever language the client is most comfortable using.
The Problem
Dezerv's RM conversations were the company's richest source of customer insight and its least accessible. Every call revealed what clients were worried about, which explanations resonated, and which topics RMs needed more coaching on. With thousands of calls every month, these conversations created a constantly evolving picture of client needs and business trends. Yet all of that insight remained locked inside audio recordings.
The previous pipeline was outsourced end to end, which meant Dezerv could not audit its own quality, and when the team ran its own comparisons, the transcripts fell short, especially on Indic languages. Nothing was queryable across the full call base either.
The transcription itself was the hard part. Dezerv's clients and RMs speak the way urban India speaks: English, Hindi and regional languages mixed within a single sentence, dense with fund names and numbers, carried over video conferencing. In a portfolio conversation, getting the numbers right is a crucial aspect.
Why Sarvam
Dezerv didn't choose Sarvam based on product claims alone. The team benchmarked multiple leading Speech-to-Text platforms using open-source datasets, including AI4Bharat's Svarah, measuring both Word Error Rate (WER) and Character Error Rate (CER) across English, Hindi, Marathi, Telugu, and Tamil.
Sarvam consistently delivered the strongest performance, particularly on Indic languages.
Three capabilities made the difference.
Accuracy on real Indian speech: Sarvam's Speech-to-Text models are trained on over a million hours of Indian audio, including code-mixed conversations, regional accents and telephony recordings. These are the same conditions Dezerv's calls present.
Diarization built-in: Sarvam's Batch Speech-to-Text API includes built-in speaker diarization, automatically separating different speakers throughout a conversation. In wealth management, knowing who said something is just as important as knowing what was said.
Scale without operational complexity: Dezerv submits large volumes of conversations asynchronously and receives structured, speaker-separated transcripts back, without building or operating any transcription systems of its own.
Execution speed reinforced the decision. When the pilot began, Sarvam didn't yet support video conferencing transcription. The capability was built within weeks. Later, when Dezerv needed support for calls longer than an hour, the required infrastructure changes were delivered in under a week.
The Solution
Dezerv integrated Sarvam through its SDK and today runs three production use cases: transcription of video meetings, transcription of phone calls, and live note-taking for Relationship Managers (RMs) during client conversations.
With client consent, recordings are processed through Sarvam's Speech-to-Text (STT) API in batch mode with speaker diarization. Every conversation is converted into a structured transcript that captures who spoke, what they said, and when they said it. Client data is managed under Dezerv's data governance framework, and Zero-Retention AI Processing ensures that no client-identifying information is ever stored with Sarvam.
From there, each conversation becomes a searchable record of the client relationship. Clients no longer have to repeat themselves. RMs can instantly revisit previous discussions without replaying hours of audio. Every interaction begins with the full context of earlier conversations: what the client asked, which topics needed further explanation, and the goals and concerns that matter most.
The Impact
Every call now becomes a searchable, speaker-separated transcript within hours. Dezerv processed close to 3.6 lakh minutes of speech through Sarvam last month: around 3.2 lakh minutes of video meetings and 40,000 minutes of phone calls.
Compared with its previous setup, transcription quality improved significantly, with Word Error Rate reduced by around 20%. Because Dezerv now controls the pipeline, it can verify that quality rather than take it on trust. Faster turnaround also allowed Dezerv to bring its insights module in-house.
Looking Ahead
Dezerv is building the wealth management firm India's professionals were never offered: expert-led, technology-first, accountable for outcomes. Learning from every client conversation is central to that vision. Every discussion now improves not only the relationship it belongs to, but also the firm's understanding of what clients need, where advisors can improve, and how better advice can be delivered at scale. As conversation volumes continue to grow and Sarvam advances its speech models for Indian languages, the value of that institutional knowledge will only compound.
Disclaimer: The statements contained herein express the views and opinions of Dezerv, based solely on Dezerv's internal data and personal experience utilizing Sarvam AI's Platform.