Voice AI in Banking and Insurance: How Cost Per Minute Became the New Metric That Matters

# Voice AI in Banking and Insurance: How Cost Per Minute Became the New Metric That Matters

The unit economics of voice AI just crossed a threshold that changes how financial services companies think about headcount.

Two years ago, deploying a voice AI agent in a banking or insurance contact center cost Rs 5–7 per minute. Today, competitive pricing sits at Rs 1.5–3 per minute, with some providers pushing below Rs 2. That's not a marginal improvement. That's a pricing collapse that makes the human-versus-AI comparison unavoidable.

Desible.ai, a voice AI company building for banking and insurance in India, is operating at the center of this shift. Its founders recently laid out exactly how the economics work — and what the business model looks like when you're selling AI agents directly against human call center employees.

The number that replaced headcount: cost per minute

In contact centers, the traditional metric is cost per call or cost per resolution. Voice AI introduces a cleaner, more brutal comparison: cost per minute of conversation.

Human agents in Indian contact centers typically cost Rs 8–15 per minute when you factor in salary, training, attrition, floor space, and management overhead. Early voice AI at Rs 5–7 per minute was cheaper but not dramatically so — and the quality gap was wide enough to justify the premium for human staff.

At Rs 1.5–3 per minute, that calculus flips. The quality gap has narrowed. The cost gap has widened. And the availability gap — humans work shifts, take breaks, call in sick — has always favored AI.

Desible.ai's co-founder Uttam made the point directly: customers are now comparing AI systems head-to-head with human employees on contactability. Not just cost. Not just accuracy. Contactability — the percentage of customers you can actually reach and complete a conversation with — is becoming a primary KPI.

An AI agent that's live 24/7, handles simultaneous calls without queue delay, and maintains consistent tone across a million interactions per month scores differently on contactability than a 200-person call center floor. That's the conversation banks and insurers are now having internally.

Why banking and insurance are the right verticals

Voice AI has been applied across industries — retail, logistics, healthcare — but banking and insurance have specific structural properties that make them ideal deployment targets.

High call volume, low call variance. The majority of inbound banking calls fall into a narrow set of categories: balance inquiries, transaction disputes, EMI reminders, KYC updates, loan status checks. Insurers see similar concentration: policy queries, claims status, renewal reminders. Predictable call types mean AI training data is dense and the failure modes are manageable.

Regulatory pressure to document everything. Both sectors require call logging, compliance documentation, and audit trails. AI systems generate these automatically. Human-run contact centers require additional tooling and QA processes to achieve the same standard.

High stakes for non-contact. In lending and insurance, a missed call can mean a missed EMI, a lapsed policy, or a fraud event that wasn't flagged in time. Contactability isn't just an efficiency metric — it's a risk metric. That elevates the business case beyond cost savings.

The pricing model shift

The move from Rs 7 to under Rs 2 per minute didn't happen because underlying compute costs dropped uniformly. It happened because the market structure changed.

Early voice AI vendors were selling novelty. Pricing reflected low competition, high integration costs, and genuine uncertainty about performance. As more providers entered the market — and as large language models made voice AI dramatically easier to build and fine-tune — pricing pressure intensified fast.

Desible.ai's founders positioned their company at the intersection of pricing competitiveness and vertical depth. Rather than building a horizontal voice AI platform, they went narrow: deep integrations with banking core systems, insurance CRMs, and the specific compliance requirements of IRDAI and RBI-regulated entities.

Vertical depth justifies slightly higher pricing than pure commodity providers while still participating in the overall market compression. It's the same logic AI-native companies are using across sectors in 2026 — own a workflow, not just a feature.

What this means for companies watching the metrics

For anyone tracking revenue per employee at AI startups, voice AI deployments in financial services are producing some of the most striking numbers in the current cycle.

A voice AI system handling 50,000 calls per month at Rs 2 per minute, with average call duration of 3 minutes, generates Rs 300,000 in monthly recurring revenue. The team required to maintain and improve that system might be four to six people — engineers, a customer success lead, and a domain specialist. That's revenue-per-employee at a ratio that would be impossible in a traditional contact center business.

The companies building these systems aren't one-person operations, but the underlying logic is the same as the one-person unicorn model: AI handles the labor-intensive work, humans handle the judgment and relationships, and the resulting ratio of output per person becomes the actual competitive advantage.

The benchmark that's shifting under everyone's feet

Uttam's framing around contactability as the new benchmark is worth sitting with. It's not just a sales argument — it's a reflection of how buyers inside banks and insurance companies are starting to evaluate vendors.

The old framework: can your AI handle this call type accurately?

The new framework: what percentage of our customer base can we reach successfully, at what cost, and how does that compare to our current human capacity?

When the question shifts from accuracy to contactability at scale, the budget conversation moves from IT procurement to operations leadership to the CFO. That's a different sales motion with a different timeline and a different deal size.

Desible.ai is betting that companies who own that conversation in their verticals will build durable positions even as underlying model costs continue to fall. The margin on Rs 1.5 per minute is thin. The margin on being the vendor that reduced non-contact rates from 40% to 12% for a regional bank is something else entirely.

What founders building in this space should take from this

If you're building AI agents for high-volume, regulated industries, three things from the Desible.ai model are worth applying directly:

1. Lead with a metric the buyer already tracks. Contactability, cost per resolution, non-contact rate — these exist in every contact center's reporting. Your AI needs to move one of them, and you need to say which one before the demo.

2. Vertical depth protects margin. Commodity voice AI is a race to Rs 1 per minute. Vertical AI that integrates with specific systems and compliance workflows holds pricing power longer.

3. Cost per minute is a commodity frame. Outcome per interaction is not. The moment you can show that your system achieves a resolution or a payment or a renewal — not just completed a call — you're selling a different product at a different price.

For founders thinking about how to build a one-person startup with AI, voice AI infrastructure is increasingly available as a building block rather than a core engineering challenge. The work is in the vertical integration and the outcome measurement — which is exactly where small, focused teams can move faster than large horizontal players.

The price compression in voice AI is not slowing down. The companies that survive it will be the ones that got specific enough, fast enough, that cost per minute stopped being the main thing their customers were comparing.

---

Is your company eligible? Submit to the leaderboard → onepersonunicorn.co/submit