In this episode of the Prime Venture Partners Podcast, Gaurav Ranjan sits down with Uttam and Omkar to discuss what it takes to build and sell agentic AI in Indian BFSI, from product engineering and compliance to enterprise sales and scaling beyond the first use case.
From Rs 7 per minute to under Rs 2, voice AI is rapidly becoming cheaper. But Desible.ai co-founders Uttam Tiwari and Omkar Raikar believe the real opportunity lies beyond conversations, in automating the workflows that run large enterprises.
Building a voice AI agent that sounds human is no longer the hardest part of the problem. Getting that agent to operate reliably inside a bank or insurance company, navigate regulatory requirements, integrate with legacy systems and deliver measurable business outcomes is an entirely different challenge.
For Uttam Tiwari and Omkar Raikar, co-founders of Desible.ai, this distinction is shaping how they build their company. Desible is developing an agentic workflow orchestration platform for regulated enterprises, with an initial focus on insurance.
Speaking with Gaurav Ranjan, Principal at Prime Venture Partners, on the Prime Venture Partners Podcast, the founders discussed why AI pilots frequently fail to reach production, how enterprise buying behaviour is changing and why the economics of voice AI could fundamentally reshape the contact centre industry.
From 25 people to three: The moment that changed everything
Uttam and Omkar have known each other for 17 years. Their previous venture, Business Toys, was acquired in 2023, following which they worked with US enterprises, including Newsweek, Bank of America and Nestlé, through a software consulting and AI business.
The turning point came during a logistics brokerage project involving AI-powered calls to truckers for availability checks and verification. Following the implementation, the client restructured a contact centre of approximately 20 to 25 people to just three.
The experience demonstrated the potential of voice AI to change how enterprises operated. But as the founders explored the Indian market through proof-of-concept projects with 12 to 13 enterprises, they realised that automating conversations alone would not create a sufficiently differentiated business.
"Voice is a commodity. I do agree, but at the same time, workflows are not because workflow goes and sit deep inside organizational processes and that's where you get the moat," Uttam explained.
That insight led Desible to focus on insurance, where customer conversations are closely connected to renewals, persistency, claims, sales and servicing workflows.
Why AI demos succeed but production deployments fail
One of the most revealing discussions centred on the difference between demonstrating AI capabilities and deploying them inside regulated enterprises.
Uttam estimates that the conversion of proofs of concept into production deployments across the companies he has observed is below 35–40%. A successful demonstration, he argued, is only the beginning of a much more complicated journey.
Insurance companies require vendors to satisfy business teams, technology departments, information security teams, compliance functions and senior technology leadership before allowing systems into production.
Omkar described Indian information security reviews involving approximately 450 questions, including requirements around data residency and compute infrastructure. Even established international technology providers can struggle with these requirements.
Desible consequently invested much of its initial funding into building compliant infrastructure rather than concentrating exclusively on voice capabilities.
For Omkar, consistency is particularly important because insurance conversations leave little room for improvisation.
"You need to be speaking the same boring language every single time, staying compliant."
The technical challenges extend beyond compliance. Voice agents must manage interruptions, understand multilingual conversations, retrieve information from enterprise systems and respond appropriately without losing context.
In one example, Omkar explained how an AI agent conducting a medical questionnaire must recognise when a customer interrupts before hearing the complete question. In another, the system must identify a medicine mentioned during a conversation, verify it against a database and ask the appropriate follow-up question.
These capabilities distinguish a production-ready system from a conversational demonstration.
The economics of Voice AI are changing rapidly
The pricing of voice AI has undergone a dramatic shift. As Gaurav highlighted during the discussion, traditional human contact centre operations can cost approximately ₹20–₹30 per minute, while early voice AI solutions were priced at Rs 5–Rs 7 per minute.
Competitive pricing has since fallen to approximately Rs 1.5–Rs 3 per minute.
For startups building voice AI products, this creates an uncomfortable question: if the underlying technology keeps becoming cheaper, how can they capture a meaningful share of the value they create?
Omkar argued that per-minute pricing was never the natural economic model for contact centres, which historically charged based on seats or headcount. The emergence of voice AI introduced a pricing structure based on infrastructure and API consumption.
Desible is attempting to move beyond this model by charging for workflow execution rather than treating each call as an isolated transaction.
The company is already handling close to a million calls daily, according to Omkar. He also described how Desible has reduced its underlying call execution costs from approximately Rs 7–Rs 8 using earlier approaches to below Rs 1.5–Rs 2 through its orchestration capabilities.
However, the founders acknowledge that enterprise customers are still evaluating whether AI delivers sustainable returns.
Uttam shared an example of a customer that did not renew three small regional contact centres after Desible's solution had been running for approximately four and a half months.
The broader transition, he cautioned, remains gradual because enterprises need sustained operational evidence before reallocating budgets.
Enterprise buyers no longer care only about human-like voices
A year earlier, enterprise customers primarily evaluated voice AI through parameters such as latency, tone, language switching and how naturally an agent sounded.
That evaluation has changed. Today, Uttam explained, customers increasingly compare AI systems directly with human employees on contactability, conversion rates, lead qualification and the accuracy of information recorded after conversations.
The shift from technological novelty to measurable operational performance is also changing how Desible approaches enterprise sales.
Rather than pursuing every potential AI experiment, the company prioritises workflows that already exist within an organisation. Such processes have established owners, operating procedures, budgets and performance benchmarks, making it easier to demonstrate the value of automation.
"Your champion may not be your buyer," Uttam cautioned.
A business leader enthusiastic about an AI pilot may still lack the authority to approve deployment. Information security, technology, compliance and procurement teams can all influence the final decision.
For founders selling enterprise AI, identifying these stakeholders early can be as important as building the product itself.
In India, enterprise sales is as much about navigation as technology
Perhaps the most striking observation came when Gaurav asked whether enterprise sales was 40% product and 60% navigating internal organisational dynamics.
Omkar suggested that in India, the latter could account for as much as 80%.
Even a technically superior product, he argued, may fail without the ability to understand decision-making structures, secure internal support and sustain relationships across multiple stakeholders.
Desible's expansion strategy illustrates the approach. At one insurance customer, the company began with sales funnel management before expanding into welcome calls, persistency and inbound customer servicing.
Each successful workflow creates additional trust and reduces the effort required for subsequent integrations. Within insurance renewals alone, the company has identified more than 20 distinct workflows.
The founders believe that this combination of industry specialisation, compliance readiness and integration into existing enterprise processes creates a more durable advantage than voice quality alone.
Beyond voice: Building an operating system for insurance
Despite the competitive intensity surrounding voice AI, Omkar believes the market remains underexplored, particularly in India, where multiple languages, dialects, background noise and frequent language switching make conversational systems considerably more challenging.
He believes companies that solve these problems at scale could emerge as global leaders.
For Desible, however, the long-term ambition extends beyond handling calls. The founders envision an orchestration platform that supports insurance companies and intermediaries, including third-party administrators, brokers and health aggregators, across their operational workflows.
The ultimate measure of success will not be how convincingly an AI agent speaks, but whether it helps enterprises make better decisions, improve customer outcomes and operate more efficiently.
Uttam's assessment of where the industry currently stands captures the central lesson for AI founders:
"POC is overhyped but production is underhyped."
As enterprises move from experimenting with AI to demanding measurable results, the winners may not be the companies with the most impressive demonstrations, but those capable of turning AI into dependable infrastructure for everyday business operations.
Source: YourStory
