Resources

Field notes on conversational AI that has to work in the real world

Written for operators who have to make a decision, not for search engines. Everything here is something we had to work out ourselves: Meta’s category pricing, per-market messaging registration, what a resolution rate is actually worth, and none of it is behind a form.

Cornerstone guide

Pricing & economics 14 min read · updated Aug 2026

Cost per resolution: the only conversational AI metric that survives a CFO

Published Indian voice AI rates run from about ₹2.00 to ₹20 a minute, and comparing them is almost entirely useless. A cheap agent that resolves four calls in ten hands six back to your team, and those six cost you a salaried human, a callback, and often the customer. This guide gives you the arithmetic, the four inputs you need to measure it on your own traffic, and a worked example from a 400-lead-a-month brokerage.

The arithmetic

Rate × duration ÷ resolution rate, and why the denominator decides everything

Four inputs to measure

Containment, handoff rate, repeat contact, agent minutes saved

Worked example

400 portal leads a month at three minutes a call

Read the guide

Start here by role

Founder or owner

Start with cost per resolution and the five-minute rule. Twenty minutes total.

Head of support or CX

WISMO, handoff completeness, and how containment is actually measured.

Head of sales

Lead response latency, dead-lead reactivation, and what qualification the AI can carry.

Technical evaluator

The state layer, the 21-pair continuity suite, latency methodology and the open API.

One email a month

What changed in Meta’s rate card, what TRAI did, what we shipped. No drip sequence.

9 of 9 guides · 85 min of reading

11 min

Meta’s four WhatsApp categories, and why your bill suddenly changed in 2026

Service, utility, authentication and marketing are priced very differently, and the free 24-hour service window is the most under-used cost advantage in Indian business messaging.

Read →

9 min

TRAI DLT registration, explained for people who would rather not learn it

Three separate registrations, a queue you cannot compress, and four constraints that silently drop your messages. Plan six weeks.

Read →

12 min

Why Hinglish breaks most voice AI, and how to test it in ten minutes

Code-switching mid-sentence is the norm in Indian customer conversation, and it is where benchmark scores and reality diverge most.

Read →

14 min

Cost per resolution: the only metric that survives a CFO

A worked example on 400 monthly leads, and the four inputs you need to measure it on your own traffic rather than trusting anyone’s deck.

Read →

8 min

Twenty-one channel pairs: what “omnichannel” actually costs to build

Four channels is six ordered pairs. Seven is twenty-one. Every pair is a place a conversation can lose its thread, which is why continuity is a test suite and not a claim.

Read →

7 min

The five-minute rule, and what happens to leads after 7 PM

Response latency is the single biggest determinant of conversion in Indian residential sales, and nights and weekends are structurally unstaffed.

Read →

10 min

WISMO is 60% of your support volume. It should be 0%.

Answering order-status questions from the live record instead of a template, and why an AI reply inside the 24-hour window costs almost nothing.

Read →

6 min

Latency, interruption and the two-bar connection

Why p95 latency on a real mobile network, Mumbai, Manila or Sharjah, is a completely different number from the one in a vendor datasheet.

Read →

8 min

When a specialist dialler is the better fit than a conversation platform

High-volume outbound calling with no messaging layer and no CRM depth is a different problem to solve. A frank guide to telling the two apart before you shortlist.

Read →

Article template: how a guide is laid out

Every long-form piece follows this structure: an answer in the first sixty words, a sticky table of contents, a data table or worked example, and one honest limitation. That order is what both a reader and an answer engine extract from.

On this page

The short answer

The three registrations

Who signs what

Four constraints that catch everyone

What we take on for you

One honest limitation

WhatsApp LinkedIn Copy link

Resources → Compliance → TRAI DLT registration

TRAI DLT registration, explained for people who would rather not learn it

By the iVaak platform team · 9 min read · Updated August 2026

The short answer

Business SMS in India requires DLT registration of your entity, your sender header and every content template, separately, per customer. From a standing start it commonly takes seven to twenty-one business days, and unregistered messages are silently dropped at the operator rather than returned as an error. Plan six weeks, not one.

The most expensive mistake in Indian messaging is treating regulatory onboarding as a technical task. Connecting an SMS provider takes an afternoon. Being allowed to send takes weeks, and the delay lands during onboarding, the worst possible moment, when a new customer is still deciding whether the purchase was a mistake.

Below: the three registrations, who signs what, the four constraints that catch everyone out, and what we take on for you as a managed service.

Registration

Typical time

What it needs from you

Entity registration

3–7 business days

GST or CIN, authorised signatory, letter of authorisation

Sender header approval

3–7 business days

Your six-character sender ID, and it must not be already taken

Content templates

1–5 days each

Exact message text with variables marked. Deviation fails scrubbing silently

One honest limitation

We cannot compress a regulator’s queue. What we can do is prepare submissions that do not bounce, chase them, and give you a customer-live date rather than a code-complete one.

Reading is not the same as trying it.

10,000 credits, fourteen days, no card. The fastest way to know whether any of this applies to your business is to point it at your real customers.

Book a demo