When a customer calls a bank to ask why their loan application was rejected, the person on the other end of the line is usually having one of the hardest conversations in financial services. Arrowhead AI, a Bengaluru startup, has taught a machine to hold that conversation — in the customer’s own language, at any hour of the night, without losing the thread.
The company builds enterprise conversational voice agents for banks, NBFCs and fintech firms: AI voice bots that handle inbound customer support and outbound sales calls end to end, sustaining natural conversations of up to 20 minutes. Its agents switch languages mid-call — English, Hindi, Marathi, Tamil, Malayalam, Kannada, Telugu, Gujarati, Bengali and more — and integrate with enterprise CRMs and diallers so customers never have to repeat themselves.
The real proof is already on the phone lines. Digital lending platform Kissht put Arrowhead’s agents on inbound support calls through its Ring app earlier this year, and within two months the bots were handling 40% of all inbound customer support volume, resolving around 88% of calls without any human involvement at an average handle time roughly 30% lower than human specialists — around the clock, 24/7. The company says the bots have delivered up to 45% higher conversion rates than human agents in sales deployments, and that every proof-of-concept it has run has converted to a live deployment.
What makes the technology remarkable is how much of it Arrowhead builds itself. Instead of layering an application on top of someone else’s models, the team built its core voice stack in-house: the orchestrator, the speech-to-text engine, a small language model tuned specifically for banking and financial use cases, and its own text-to-speech trained on thousands of hours of voice-artist recordings to sound colloquial and natural to Indian ears. The company says its speech recognition achieves a 10% lower word error rate than leading models, particularly in regional languages — and that its specialised small model cut conversational latency from about a second to roughly 500 milliseconds, which is what makes a phone call feel human instead of robotic.
The agents are engineered for the gravity of financial conversations too. Guardrails are built into the language model and the agent’s instructions: in the loan-rejection and NOC-cancellation categories it handles for Kissht, a mention of the regulator hands the call straight to a human specialist. The EMI and repayment agent now in development runs on an instruction the company describes as “money facts, absolute” — the bot may never estimate or round up a financial figure.
Arrowhead was founded in 2022 by Devyani Gupta (CEO, previously with Boston Consulting Group, Wharton-educated) and Vengadanathan Srinivasan (CTO, who was among the first five employees of AWS’s Elasticsearch team and previously built at Airbnb, Uber and Rippling). The team has grown past 40 people across Bengaluru, Mumbai, Delhi and Chennai, counts more than 50 large customers — including Kissht, Paytm, Aditya Birla Capital, Bank of Baroda Cards, Tata 1mg, UpGrad, Indiamart, InsuranceDekho, RBL Bank and BharatPe — and was named to LinkedIn’s 2026 Top Startups list for India this October, shortly after winning the L&T Finance Raise’25 competition.
More Read
From a conversation-analytics startup to a voice-AI workforce for India’s banks, Arrowhead is building the technology layer where India’s languages, its financial-services scale, and genuinely low-latency AI meet — and every phone call is a chance to prove it.
Keep reading: Thinking in Speech: Inside Chariot, the Delhi Lab Building Frontier AI for 10 Indian Languages · Hands That Feel: Inside Perceptyne, the Hyderabad Startup Teaching Humanoids to Work on Assembly Lines · More Industrial Automation & Technology stories
Like stories like this? Add Exceptional Technology as a preferred source on Google to see more of our reporting in your search results and Discover feed.
