The Unit Economics Of India’s Voice AI Boom

India’s voice AI revolution has revolved around a few key pillars, such as speech recognition accuracy, multilingual capabilities, latency and human-like conversations. Engrossed in perfecting the tech, Indian entrepreneurs are realising that building it was only half the task, and the bigger challenge is building a profitable business around it.

As enterprises move beyond pilot deployments to rolling out AI-powered voice agents at scale, pricing strategies, infrastructure ownership and customer selection are proving just as critical as the quality of the AI itself in determining whether a voice AI startup can build a sustainable business.

The revelation made us curious enough to ponder: As India’s voice AI ecosystem matures, how are the startups powering it actually making money? Which pricing model is proving the most sustainable? And what kind of customers generate the scale needed to build a profitable voice AI business? Let’s talk about the unit economics behind voice AI.

Is Voice AI Making Money? 

Despite growing interest in outcome-based AI, the industry continues to revolve around one simple pricing model: pay for every minute an AI agent speaks.

For startups such as Bolna AI, which builds multilingual voice AI agents for enterprises, per-minute billing offers predictability for both customers and vendors. While outcome-based pricing is possible for narrowly defined workflows such as gig-worker recruitment or customer surveys, most enterprise deployments involve variables beyond the AI agent’s control, making usage-based pricing commercially simpler.

“Per minute works better for us and for the brand because factors that determine the outcome go much beyond just the voice AI call,” said Maitreya Wagh, the cofounder of Bolna AI.

This is also reflected in the company’s pricing. Self-serve customers typically pay around ₹5.5 per minute, while enterprises committing to significantly larger, long-term volumes can negotiate rates as low as ₹1.75-₹2 per minute. The discounts are central to the economics of voice AI.

Higher call volumes enable providers to negotiate better infrastructure costs, optimise model usage and improve caching efficiencies, ultimately lowering the cost of every conversation. Startups are also becoming increasingly selective about where they deploy expensive reasoning models, relying instead on lighter graph-based agents wherever possible to preserve margins.

The comparison with traditional call centres also makes the value proposition easier for enterprises to understand. Human calling operations can cost businesses around ₹7-₹8 per minute, excluding hiring, training and employee attrition. Voice AI significantly lowers those operating costs while offering round-the-clock availability.

However, founders believe the larger opportunity extends beyond replacing human agents. Voice AI is increasingly enabling entirely new customer interactions, from multilingual outreach to conversational form-filling, creating markets where businesses previously found customer engagement too expensive to justify.

The Margin Battle 

While pricing determines revenue, technology ownership increasingly determines profitability. Every voice AI interaction relies on multiple components, including telephony, speech recognition, orchestration, language models, denoising and text-to-speech systems. Many younger startups assemble these layers using third-party providers to accelerate product development. This lowers the barrier to entry but can significantly compress margins.

According to Ganesh Gopalan, the CEO of Gnani.ai, an enterprise conversational Voice AI platform, aggregators often struggle to make money because every layer of the voice AI stack carries an external cost.

Gnani.ai has built most of these capabilities in-house, including its orchestration engine, denoising models and turn-taking systems that determine when an AI agent should respond during conversations.

The company claims this enables gross margins exceeding 80%, highlighting the commercial advantage of owning more of the underlying technology stack rather than paying multiple providers for every conversation.

That distinction is becoming increasingly important as foundation models, speech models and telephony infrastructure continue to become cheaper.

According to Suman Gandham, the CEO of Vobiz.ai, infrastructure itself is gradually becoming commoditised. The real long-term value lies in helping enterprises manage Voice AI deployments through governance, compliance, observability, analytics and workflow optimisation.

Today, enterprises are not buying minutes of telephony or tokens of inference. Instead, they are paying for reliable customer interactions, compliance, operational visibility and measurable business outcomes. 

Infrastructure players are steadily shifting from selling telephony and inference to selling operational intelligence. The transition mirrors what cloud computing experienced over the past decade, where infrastructure providers gradually differentiated themselves through managed services rather than raw compute.

Who Is Buying Voice AI?

Voice AI may be attracting startups of every size, but its economics remain fundamentally a scale game. Businesses serving millions of customers derive the greatest value because usage-based pricing only becomes commercially attractive when conversations happen at high volumes.

One voice AI founder said deployments handling roughly 60,000-90,000 connected call minutes for a single use case are generally considered a healthy benchmark for the economics to begin working. That being said, the exact threshold depends on a company’s fixed costs and infrastructure investments. Below that, enterprises are often still evaluating pilots rather than running production-scale deployments.

This explains why banking, financial services, ecommerce, retail and collections have emerged as some of the earliest adopters.

Voice AI deployments are clustering around functions that directly influence revenue or cash flow. Customer acquisition, cross-selling and collections often receive budget approval faster than customer support because enterprises can directly measure their business impact. The opportunity is also expanding beyond cost optimisation.

Founders increasingly describe Voice AI as India’s interface for reaching India at scale. Businesses operating across multiple Indian languages often cannot economically employ enough customer service executives to engage every customer. AI agents capable of conducting thousands of multilingual conversations simultaneously allow companies to dramatically expand their reach without proportionally increasing operating costs.

Conversely, sectors dependent on highly personalised relationships, such as premium wealth management or private banking, are less likely to adopt Voice AI at scale because human interactions continue to remain central to their customer experience.

For now, per-minute pricing remains the industry’s dominant commercial model, while outcome-based pricing is gradually finding acceptance in narrowly defined workflows where business results can be measured more precisely.

But as enterprises become more comfortable handing larger workflows to AI agents, commercial models are likely to evolve beyond charging for conversation time alone.

Several founders believe the industry is likely to settle on hybrid commercial models, where enterprises pay a base usage fee alongside incentives tied to measurable business outcomes such as successful collections, loan disbursals or completed hiring workflows.

Top Stories From India & Around The World

• MeitY Pauses Frontier AI For Govt Cybersecurity: MeitY has asked central ministries to hold off on deploying OpenAI and Anthropic models for cybersecurity and related functions, citing concerns over the dual-use risks of frontier AI. The move underscores India’s push for AI sovereignty as the government weighs security, data control and reliance on foreign AI providers for sensitive functions.
• Unicorn Founders Go AI Native: India’s second-time founders are increasingly leaving operating roles to build AI-native startups, with recent entrants including Bhavin Turakhia (Neo), Rahul Sharma (AI robotics), Mukesh Bansal (Nurix AI) and Mukund Jha (Emergent). The trend signals a shift from consumer internet to enterprise AI, although founders now face far steeper capital and compute requirements.
• AI Push Triggers Layoffs At GoKwik: Ecommerce enablement startup GoKwik has reportedly laid off around 100-120 employees as part of an AI-led restructuring, with customer onboarding, implementation and tech teams among the worst affected. The move reflects a broader trend of Indian startups reshaping workforces as AI automates operational functions.
• OpenAI Expands Enterprise AI Play: OpenAI has launched GPT-5.6 and ChatGPT Work, an AI workspace combining coding, research, document creation and workflow automation to compete with Claude Cowork, Microsoft 365 Copilot and Gemini for Workspace. The launch highlights intensifying competition to become the default AI operating system for enterprise productivity.
  

The Weekly Buzz: Anthropic Goes Local For India

Anthropic has rolled out Rupee pricing for Claude subscriptions in India, its second-largest market after the US at 5.8% of global usage. Claude Pro now costs ₹2,399 a month, or ₹2,000 on annual billing. Max and Team plans are localised too, though UPI support is still missing, unlike ChatGPT’s India rollout.

The timing matters beyond convenience. In June, Anthropic suspended access to its Fable 5 and Mythos 5 models for non-US users, citing US Commerce Department export controls. Fable 5 access is back, Mythos 5 remains limited. That episode rattled Indian developers who had just started building on these models. Rupee pricing does not undo it, but it signals Anthropic still wants India’s business, even after a regulatory setback out of its own control.

There is a currency risk too. The Rupee has slid past ₹95 to the dollar this year, pressured by tariffs, FPI outflows, and oil volatility from the US-Iran conflict. Whether Anthropic holds this Rupee pricing or revises it as the rupee moves is worth watching.

The pricing itself is not cheap, roughly matching the dollar rate after GST. In a market where free users vastly outnumber paying ones, that gap remains the bigger challenge.

Startup In The Spotlight: OpenCFO

Founded in 2025 by Sankalp Singayapally and Prudhvi Rao Shedimbi, OpenCFO is building an AI-native financial operating system that helps enterprises automate accounts receivable, accounts payable and treasury operations from a single platform.

Rather than replacing existing ERP systems, the startup positions itself as an AI orchestration layer sitting above enterprise finance infrastructure. Its AI agents automate collections, payment reconciliation, invoice matching, liquidity monitoring and FX optimisation, while integrating with platforms such as Oracle NetSuite, Microsoft Dynamics 365, QuickBooks and Zoho Books.

OpenCFO claims its platform enables finance teams to improve working capital efficiency by reducing days sales outstanding (DSO), optimising days payable outstanding (DPO) and providing real-time visibility into cash flows, eliminating the need to manually reconcile data across multiple systems.

Looking ahead, the startup is betting that enterprise finance will increasingly shift towards autonomous, AI-driven operations, with treasury and cash management becoming continuously automated rather than manually coordinated.

Prompt Of The Week

What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work? 

Here’s the prompt Ganesh Shankar, CEO and Cofounder of Responsive, uses to stress-test important business decisions by turning AI into a virtual executive team that challenges assumptions before execution.

“Act as an executive team with different perspectives: CEO, CFO, CTO, Chief Customer Officer, and my toughest board member. 

Have each person argue for and against:

• the decision

• identify blind spots

• second-order effects

• execution risks
• unintended consequences. 

End with a recommendation, confidence level, assumptions that must hold, and a list of signals that would tell me I need to change course. ”

Editor’s Note: Some prompts may need to be adjusted by users for best results or may not work as intended for certain users.

[Edited by Shishir Prasher]
[Creatives by Varshita Srivastava]

The post The Unit Economics Of India’s Voice AI Boom appeared first on Inc42 Media.