Kissht (Ring), an Indian digital lender, put Arrowhead's voice AI agent on its inbound customer support line, starting with the queue its own customer experience team considered hardest to get right: customers calling after a loan application was rejected, and customers asking to cancel a mandate or obtain a no-objection certificate (NOC). These are emotionally charged conversations close to regulatory exposure, where answers have to be correct and consistent every time. According to Arrowhead, almost every voice AI deployment in India has started on outbound sales; Kissht started with inbound support.
The constraint was capacity and consistency. Senior specialists answered these calls well, but only during a nine-hour working day and only as many at a time as were on shift, across roughly 30,000 conversations a month. Calls outside working hours, about 20% of inbound volume arriving between 8 PM and 8 AM, waited in a queue for the next working day.
The scope was drawn narrow and enforced strictly. The agent, which speaks Hindi and English, handles only the two call types. If a customer mentions the Reserve Bank of India, the call is handed to a human within the same conversational turn, with no attempt to answer. The agent is integrated with Kissht's dialer and CRM, so the customer's record and application status are loaded before the call connects and customers do not have to identify themselves or repeat their story. The main technical constraint was latency: Arrowhead runs the agent on its own small language model, which brought conversational turn latency down from about one second to about 500 milliseconds.
The rollout was run jointly. The teams held daily stand-ups from day one, Arrowhead's prompt engineers listened in on live Kissht customer calls and worked alongside Kissht's team, and Kissht had direct access to Arrowhead's engineers and founders outside business hours instead of a ticket queue. "We had daily standups with the Arrowhead team from day one. Not a weekly review, an actual working rhythm," said Shadab, Senior Product Manager at Kissht.
"Within the first two months, the bot was handling forty percent of our entire inbound volume," said Suraj Shetty, Head of Customer Experience at Kissht (Ring). The 40% is measured against total inbound calls across all queues, not a test cohort. About 88% of those calls are closed without any human involvement, and average handle time is about 30% lower than for human specialists on the same queues over the same period, which Arrowhead puts at roughly a 30% saving on every call the agent resolves end to end, with customer experience holding steady. The after-hours volume is now answered 24/7 under the same scope and guardrails as during the day.
Kissht and Arrowhead are now building a second agent for EMI and repayment queries, covering payment confirmations, auto-debit outcomes, double deductions, late fee status, foreclosure, and NOC and CIBIL timelines. "We started with a queue where the risk was contained. And now we are moving into EMI and repayment journeys, where the stakes are much higher. We would not be doing that if the first one had not held up," Shetty said.