"Six weeks" is the answer most Indian development companies give. It is not a lie, but it describes the engineering only. It ignores what is genuinely out of everyone's hands: Meta business verification, DLT registration, template approval, and the days your own team needs to answer questions nobody has asked them before. Here is the honest version, with the waiting written in.
The short answer
| Type of agent | Engineering time | Approvals and waiting | Realistic go-live |
|---|---|---|---|
| Web chat agent on your own site | 2-3 weeks | None | 3 weeks |
| WhatsApp agent, new Meta account | 3-4 weeks | 7-15 days | 6-7 weeks |
| WhatsApp agent, verified account exists | 3-4 weeks | 2-4 days | 4-5 weeks |
| Voice agent, inbound only | 4-5 weeks | 5-10 days (numbering) | 6-7 weeks |
| Voice agent, outbound calling | 5-6 weeks | 3-6 weeks (DLT, headers, consent) | 9-12 weeks |
| Agent wired into an existing ERP or CRM | 6-9 weeks | Varies | 8-12 weeks |
|---|
The number that surprises people is outbound voice. The code is no harder than inbound; the regulatory path is.
Week by week: a WhatsApp agent, the common case
This is the most requested build in India right now. Assume a business with one WhatsApp number, an existing enquiry flow, and three people answering escalations.
Weeks 0-1: scoping and the conversation inventory
Not workshops. We ask for the last 200 real conversations from your WhatsApp or phone log. Screenshots are fine. This one artefact decides whether the project succeeds, because it shows what customers actually ask instead of what you think they ask.
Most businesses discover that 60-70% of enquiry volume is four or five questions repeated endlessly: price, availability, location, timing, "are you open". That is the automatable core. The rest goes to a human, and defining that boundary is most of the design work.
What slows this week down is nobody on your side owning the project. If the decision-maker is travelling for a fortnight, the project slips a fortnight. No engineering effort recovers it.
Weeks 1-2: Meta business verification, running in parallel
Start this on day one, not when the code is ready. This is the mistake that costs projects a fortnight.
A WhatsApp Business API number needs a verified Meta Business account, which needs documents matching your legal entity exactly: incorporation or GST certificate, a utility bill or bank statement with the same registered address, and a website with a working contact page on your own domain.
| Approval step | Typical wait | Common reason for rejection |
|---|---|---|
| Meta business verification | 2-10 working days | Address on document differs from Meta profile |
| WhatsApp number registration | Same day - 2 days | Number already on consumer WhatsApp, not deleted first |
| Message template approval | 1-24 hours each | Promotional wording in a utility template |
| Green tick (optional) | 2-6 weeks, often refused | Insufficient press coverage of the brand |
|---|
Two things worth saying plainly. Most SMEs will not get the green tick, it is not required, and any vendor promising it is guessing. And a business number on the normal WhatsApp app must be deleted from it before API registration, losing the chat history. Plan a cutover evening and warn your team.
Weeks 2-4: building the agent
The engineering itself is unglamorous and predictable.
- Knowledge base built from your real pricing, catalogue, policies and service areas. Accuracy comes from here, not from the model.
- Conversation flow and escalation rules: which questions must never be answered by a machine.
- Integrations with your CRM, Sheets, payment links or calendar, whatever actually exists.
- Retrieval and guardrails, so the agent declines rather than invents.
- Logging, so every conversation is reviewable. Non-negotiable.
Two to four working weeks for a competent build. A four-day quote means someone is reselling a configured SaaS template. Sometimes that is the right purchase.
Week 5: testing with real people, not with the team that built it
Reserve five working days and put ten actual customers, or staff not involved in the build, in front of it. Hindi-English code-mixing, voice notes, misspellings, a photo of a product with "ye kitne ka hai" — these break demo-quality agents immediately. Expect two rounds of corrections. Anyone scheduling zero correction days is planning a demo, not a deployment.
Weeks 6-8: live, but supervised
Go live with a human reading every conversation daily for the first fortnight. Accuracy on day one is usually 75-85%, reaching a usable 90-95% after two weeks of corrections, because only live traffic reveals the questions you never anticipated. This supervised fortnight is the difference between an agent that survives and one quietly switched off in month two, and it is why we prefer a retainer over a one-off build.
Why outbound voice takes three times longer
Nothing to do with the AI. Outbound commercial calling and SMS in India sit under TRAI regulation, which means DLT registration through an operator portal: entity registration, header registration, content template registration, and a documented consent trail.
| DLT step | Typical wait |
|---|---|
| Entity registration and KYC | 3-7 working days |
| Header (sender ID) approval | 2-5 working days |
| Content template approval | 1-7 working days per template |
| Consent scrubbing against DND setup | 3-10 working days |
|---|
Each rejection restarts a queue. Six weeks of regulatory calendar is normal for outbound, and a vendor who does not raise DLT early has not done this before. Inbound voice is far easier because the customer calls you. If you want voice quickly, start inbound.
Where a custom build is the wrong choice
Honesty is more useful than a pitch:
- Under roughly 300 enquiries a month. The maths does not work. Buy an off-the-shelf chatbot at ₹2,000-8,000 a month, accept that it is clumsy, revisit in a year.
- You need it live in ten days. No custom build survives that. Take the worse thing that exists and is running.
- Your process changes weekly. Automating a process you are still redesigning means paying twice. Stabilise first.
- Your data lives in one person's head. If the owner prices every deal by phone, there is nothing to automate yet. Writing the rules down is the real project.
- A form and an autoresponder would do. Sometimes they would. We have said so and lost the sale.
If you are still weighing automation against hiring, our comparison of a WhatsApp AI agent against hiring another salesperson puts the costs side by side.
The five things that actually cause delays
In order of how often we see them:
- Meta verification started late: 1-2 weeks.
- No single decision-maker on the client side: 1-3 weeks, repeatedly.
- Pricing or policy nobody has written down: 1-2 weeks.
- Scope added mid-build ("can it also raise an invoice"): 1-4 weeks.
- Legacy ERP with no API and no documentation: 2-6 weeks, occasionally fatal.
Four of those five sit with the buyer. That is not a complaint but a planning instruction: assign one owner, gather documents in week zero, write down your pricing rules before kickoff. Clients who do this finish two to three weeks earlier.
How we sequence it
We run a fixed-price AI Readiness Audit over two weeks before any build, precisely because of the failure modes above. It establishes whether there is enough written-down process and enquiry volume to justify a build, and sometimes ends with us recommending you build nothing. Implementation then runs as a monthly retainer, because accuracy is earned in the supervised weeks after go-live.
For a timeline on your own situation, tell us your monthly enquiry volume and whether your WhatsApp number is already on the API. Those two facts determine most of the calendar. Reach us at innovativebiz.in, info@innovativebiz.in, or +91 98260 26029.
FAQ
How long does it take to build a WhatsApp AI agent in India?
A production WhatsApp AI agent in India typically takes six to seven weeks from kickoff to supervised go-live. Roughly three to four weeks is engineering, one to two weeks is Meta business verification and template approval running in parallel, and one week is testing with real customers. If your number is already on the WhatsApp Business API and your business is Meta-verified, four to five weeks is achievable.
Why does an outbound AI calling agent take longer than an inbound one?
Outbound commercial calling in India falls under TRAI regulations and requires DLT registration: entity KYC, header approval, content template approval and DND consent scrubbing, each with its own queue and its own rejections. That regulatory path commonly adds three to six weeks. Inbound voice agents have no such requirement because the customer initiates the call, so inbound can go live in six to seven weeks.
What is the fastest an AI agent can realistically go live?
A web chat agent on your own website can be live in about three weeks, because it needs no external approvals. If you genuinely need something running in under two weeks, a configured off-the-shelf chatbot at ₹2,000 to ₹8,000 a month is the realistic option. It will handle fewer cases and sound more rigid, but it will be answering customers while a custom build is still in design.
What causes AI agent projects to run late in India?
Four of the five most common causes sit with the buyer, not the developer: starting Meta verification late, having no single decision-maker available, having pricing or policy rules that were never written down, and adding scope mid-build. The fifth is a legacy ERP or CRM with no API or documentation, which can add two to six weeks and occasionally makes integration impractical.
Do I need a WhatsApp green tick before launching an AI agent?
No. The green tick is an optional verified-business badge and is not required to run a WhatsApp Business API number or an AI agent on it. Meta grants it based largely on independent press coverage of the brand, so most small and mid-sized businesses are refused. Any vendor promising a green tick as a deliverable is guessing. Launch without it.
How much of the timeline is the client's responsibility?
More than most buyers expect. Document collection for Meta verification, supplying real past conversations, writing down pricing and escalation rules, and giving daily feedback during the supervised fortnight after launch are all client-side tasks. Projects where one named person owns these finish two to three weeks earlier than projects where responsibility is spread across a committee.