How voice AI is transforming banking and insurance: Desible.ai founders on the business behind the technology

An agentic workflow orchestration platform coordinates conversations, business rules, enterprise data and next steps. It does not merely make an AI agent sound human. It helps the agent complete work inside a company’s existing processes. That matters because voice technology is becoming widely available and cheaper, while deep workflow integration is harder to copy.

A voice-only agent might answer a question or collect information. A workflow-oriented agent can retrieve customer records, verify information, follow compliance rules and trigger servicing, claims, sales or renewal activity. Desible’s example includes identifying a medicine in a medical questionnaire, checking it against a database and asking the correct follow-up question.

The article presents this distinction as Desible’s central strategy. The founders found that automating conversations alone was not sufficiently differentiated after testing projects with 12 to 13 enterprises. They focused on insurance because customer calls connect closely to important operational workflows. Production success therefore depends on reliable integration, consistency and measurable business outcomes.

Voice AI’s cost has dropped sharply as the technology has matured. Traditional human contact-centre operations cost approximately Rs 20–Rs 30 per minute. Early voice AI solutions cost about Rs 5–Rs 7 per minute. The article describes current pricing at approximately Rs 1.5–under Rs 2 per minute.

This means current voice AI can cost roughly 90 percent less than human handling at the lower end of the human range. Compared with early AI pricing, the reduction is about two-thirds or more. The lower price changes the economics of calling customers, especially for repetitive tasks such as checks, questionnaires and verification.

Cost alone does not guarantee adoption. Banks and insurers still demand secure infrastructure, compliance, reliable multilingual handling and integration with existing systems. Desible’s founders argue that the larger opportunity is therefore not simply cheap conversation. It is using lower-cost voice automation to complete regulated enterprise workflows consistently and at scale.

The supplied article does not provide a daily call volume for Desible. It therefore cannot support a reliable answer to how many calls the company reportedly handles each day. It also does not give a specific before-and-after cost for executing each individual call.

The available figures describe the broader economics instead. Human contact-centre operations cost approximately Rs 20–Rs 30 per minute. Early voice AI cost Rs 5–Rs 7 per minute, while current pricing has fallen to approximately Rs 1.5–under Rs 2 per minute. These figures show why automated calls can become much cheaper, but they are not a daily call count.

The article does identify the mechanism behind lower costs: replacing routine human handling with voice AI and connecting conversations to automated enterprise workflows. It gives one earlier example in which a contact centre of approximately 20 to 25 people was reduced to three after AI-powered calls. No Desible daily volume is stated in the provided text.

A successful AI demonstration proves that a system can perform a task in controlled conditions. Production deployment requires much more. Banks and insurers must approve the technology across business teams, technology departments, information security, compliance functions and senior technology leadership. That extended review creates many opportunities for a pilot to stop.

The article describes Indian information-security reviews containing approximately 450 questions. These include data-residency and compute-infrastructure requirements. Production systems must also handle interruptions, multilingual conversations, enterprise data and context without breaking compliance. Insurance conversations demand consistent language rather than improvisation.

Uttam estimates that fewer than 35–40% of proofs of concept become production deployments across the companies he has observed. Desible used much of its early funding to build compliant infrastructure, not just voice capabilities. The lesson is that enterprise AI must be designed for security, integration and reliable operations from the beginning. A compelling demo is only the first step.

When voice AI can handle routine calls at a much lower cost, companies can reduce the amount of human labour needed for repetitive contact-centre work. The change is not only about cheaper minutes. It can alter how an operation is staffed and organised, especially when calls follow predictable steps.

The article gives a concrete example from a logistics-brokerage project. AI-powered calls checked trucker availability and completed verification. After implementation, the client restructured a contact centre of approximately 20 to 25 people to just three. That result shows how workflow automation can produce a large staffing effect when the task is repetitive and structured.

The article does not specify every role that remains or how workers are redeployed. It does show that humans remain important for regulated, complex or unusual situations, since production systems must manage interruptions, context and compliance. The broader implication is a contact centre built around automated routine work and smaller teams handling harder cases.

Insurance is demanding because customer conversations are tied to regulated decisions and important business processes. A voice agent cannot improvise freely. It must use consistent language, protect information, retrieve accurate records and follow the company’s rules every time. Errors can affect customer service, compliance and operational outcomes.

The article gives a medical-questionnaire example. If a customer interrupts before a question is complete, the agent must recognise the interruption and recover the missing context. If the customer names a medicine, the system must identify it, check a database and ask the correct follow-up question. These steps require more than natural-sounding speech.

Desible’s insurance focus connects voice interactions to renewals, persistency, claims, sales and servicing workflows. Enterprise deployment also requires information-security approval, data-residency controls and suitable compute infrastructure. Insurance is therefore a demanding test of agentic AI. The opportunity lies in reliably completing connected work, not merely holding a convincing conversation.

A company offering cheaper conversations competes mainly on a capability that is becoming widely available. Desible’s founders argue that voice itself is a commodity as prices fall. A workflow, however, sits inside an organisation’s processes. It connects the agent to systems, rules, approvals and measurable business activity.

For example, an insurance agent can ask medical questions, recognise an interruption, identify a medicine, verify it against a database and ask the right follow-up. That sequence combines conversation with enterprise information and operating rules. It can support claims, servicing, sales, renewals or persistency rather than simply producing spoken replies.

This deeper integration can create a stronger business position because replacing it requires changing more of the customer’s operations. The article also shows why it is difficult: vendors must pass reviews involving security, compliance, technology and senior leadership. Desible invested early funding in compliant infrastructure. Its strategy is to scale from an initial insurance use case toward broader enterprise workflow automation.