The Sachetization Trap

India is in the middle of a Voice AI gold rush. Quite a few Voice AI startups have emerged in recent months — Bolna, Ringg, Arrowhead, Vaani, Navana, to name a few.

The premise makes sense: replace the telecaller, reduce costs, and scale the conversation. Many have also shown early success with design partners, with BFSI leading the charge. On paper, the value proposition is obvious. You can handle different languages and dialects at scale without relying on a telecaller vendor that needs constant training and monitoring. Voice AI doesn't tire. It adapts to dialects. It is infinitely patient.

On the surface, this looks like a classic India success story. Just as we made shampoo, mobile data, and payments cheap and accessible, we are now sachetizing AI — low-cost, bite-sized, ubiquitous. Vendors are selling minutes for pennies. Voice AI has dramatically reduced the cost of conversation.

But there is a distinction that is easy to miss.

Sachetization as access is revolutionary.

Sachetization as extraction is noise.

The technology is the same. The architectural choice is not.

Doctrine Concept — Access vs. ExtractionWhen capability becomes inexpensive, organizations instinctively use it to increase volume rather than increase value. The trap is not cheap technology. It is using abundance for extraction rather than access.

Extraction here does not mean exploitation in the political sense. It simply means using technological abundance to produce more without increasing relevance, continuity, or value.

Sachetization as access is revolutionary — shampoo sachets, Jio's data, UPI on feature phones. Sachetization as extraction is noise — the same capability deployed to push more volume without relevance or continuity. The technology is the same but the architectural choice is not.

1. The Spam Factory

Most Voice AI deployments in India are built for Resource Substitution. The basic model is simple: replace the human telecaller with Voice AI — the same scripts, now delivered at machine scale.

• Old Model: Hire humans to read rigid scripts.
• New Model: Spin up AI agents to read the same rigid scripts at machine scale.

It helps to step back and look at the customer experience. Did anyone actually enjoy credit card and loan calls? Truecaller became increasingly useful as unsolicited calls became more common and intrusive.

When technological abundance is used for extraction, the channel itself begins to degrade.

We have seen this before with SMS. Fifteen years ago, the SMS inbox was still a place for conversation. Today, it is largely a graveyard of spam, with much of its everyday utility reduced to OTPs, alerts, and transactional messages. This happened because we optimized the channel for cost, not relevance. Once it became cheap enough to blast a million messages, businesses had less reason to care about whether any individual message mattered. The signal-to-noise ratio collapsed, and users moved to WhatsApp — a channel with more friction and stronger consent — for actual conversation.

The SMS story is not a historical anecdote. It is evidence of a recurring pattern: when technological abundance is used for extraction, the channel degrades under its own volume.

Voice AI is now at risk of following the same path. When a lender blasts a million leads from a generic database, it isn't engaging. It is fishing, with little context and little personalization. Increase the scale by another order of magnitude, and the degradation arrives sooner.

If we continue down this path, the phone call risks becoming what SMS became: a channel that remains functional but loses its value as a place for meaningful interaction.

2. The Uncanny Valley

The second failure is structural. Most deployments are retrofitting Generative AI into call-center logic built for an earlier era.

Consider the typical experience:

• Voice AI: Great conversation, understands intent, but hits a wall.
• Human Agent 1: Zero context passed. Customer repeats everything.
• Sales Agent: Zero context passed. Customer repeats everything again.
• Outcome: The AI worked, but the system failed.

The voice sounded human, but there was no continuity behind it. The customer doesn't care that “the AI is good.” They care about the experience of engaging with the brand — whether with AI or a human — and whether that interaction actually moves their intent forward.

A system without continuity cannot build a relationship. It can only process transactions.

3. The Real Opportunity

The promise of Voice AI in India is not cost reduction. It is structural leapfrogging.

We have seen glimpses of this with the Soundbox. Why did Soundboxes work in India? Because merchants didn't want to look at screens or log into dashboards. An audio confirmation of payment solved a basic problem of trust. Millions of Indian businesses — kiranas, traders, service providers — never fully adopted the PC era and have often struggled with the SaaS era. ERPs and dashboards are difficult to use for businesses that run on WhatsApp, relationships, and intuition.

Voice AI offers a different path.

Imagine a business with no website, no app, and no login — just a Google Maps listing front-ended by Agentic Voice AI. In this world, Voice is not a dialer or IVR. Voice is the Operating System. The business owner doesn't manage software; they speak to it.

In this world, Voice is not a dialer or an IVR. Voice becomes the interface to the business.

That is the opportunity: not replacing the telecaller, but removing layers of software and interaction that were necessary in the previous architecture.

We need to apply the Ditto Insurance philosophy here. Ditto led with understanding the user's hesitation, rather than pushing a policy at them. They created conversations, understood needs, educated the customer, and then arrived at a solution without making the interaction feel like a sales process.

That is access: using abundance to increase relevance, continuity, and value — not merely volume.

4. Build the Personal Secretary

To unlock this, we must shift the metaphor from Telecaller to Personal Secretary. You have a transactional engagement with a telecaller. With a secretary, it's contextual.

A good secretary knows:

• The Business: deeply understands the product, pricing, and edge cases.
• The User: knows where the prospect is in the lifecycle.
• The Context: maintains continuity across conversations — what was said, promised, and left unresolved.

This requires omni-channel, multi-session continuity. It is not a linear, transactional sell. It is a relationship: the AI front-ends the initiative, knows exactly when to hand off to a human, and the user engages with the brand — sometimes with AI, sometimes with humans — always with continuity. The context remains intact.

The Bottom Line

If you treat Voice AI as an outsourced service ("Get me leads at ₹X/min"), you will get the Spam Factory. If you treat Voice AI as an unlock to build relationships at scale, you get something different: an interface that can expand access to the business.

Voice AI startups can provide the technology and capabilities. But operationalizing them is the brand's job. That means redesigning workflows, roles, handoffs, and incentives around the new capability. Without that change, Voice AI may look good in a pilot and still end up as another cost center.

The pattern extends beyond India

This is not a Voice AI problem. It is not an India problem. It is a recurring organizational pattern.

India makes the pattern unusually visible. It skipped or compressed multiple technology eras, so AI arrives under different institutional conditions than in markets where those eras were adopted sequentially. That history makes both the opportunity and the risk easier to see.

But the structural dynamic is the same everywhere: when capability changes, institutions repeatedly reuse architectures designed for the previous era.

The question is no longer how we make AI cheaper. It becomes: why do institutions repeatedly fail to redesign themselves, even when the limits of the existing architecture are already visible?

That question belongs to the next essay.