How long it takes to implement an AI WhatsApp chatbot (real timeline stage by stage)

An AI WhatsApp chatbot goes into production in 3 to 10 business days when the company already has a verified Business Manager, an available number and the response content ready. If any of those three things is missing, the real timeline stretches to 3-8 weeks. Development is not the bottleneck: the flow gets built in 2 to 5 days of work. What stretches the calendar is Meta's verification (1 to 5 business days with the paperwork in order, 2 to 4 weeks if there is a rejection), getting access to the CRM or the calendar, and getting someone on the company side to sit down and write the prices and the policies.

Timeline stage by stage: how long each one takes and who it depends on

These are the typical ranges we see in WhatsApp chatbot implementations for small businesses on Meta Cloud API. The column that matters is not the duration: it is whether the timeline is controlled by the consultancy, the client or Meta.

• Scope definition (half a day). Depends on the consultancy and the owner. What the bot answers, what it hands off and during which hours.
• Business verification with Meta (1 to 5 business days; 2 to 4 weeks if there is a rejection). Depends on Meta. It is the only timeline that cannot be sped up with budget or with people.
• Number registration and display name (2 hours to 2 days). Depends on Meta, though the display name is usually approved the same day.
• Building the knowledge base (2 to 15 days). Depends 100% on the client. It is the stage with the most variance in the entire project.
• Handover of access credentials (1 to 10 days). Depends on the client and on third parties: CRM, calendar, online store, management system.
• Building the flow (2 to 5 days). Depends on the consultancy. It is the part everyone believes is the whole project, and the only one that waits neither for Meta nor for the client.
• Internal testing (1 to 3 days). Depends on both.
• Pilot with real traffic and tuning (2 to 3 weeks). Depends on the volume of conversations coming in.

Adding up the stages that depend only on the consultancy, the technical work is 3 to 8 days. Everything else is paperwork and content.

Meta verification: 1 to 5 business days if the paperwork is right, 2 to 4 weeks if it is not

Business verification is done in Meta Business Suite, under Business settings, inside the Security Center. What you need ready:

• Tax registration certificate: ARCA (formerly AFIP) in Argentina, CIF in Spain, RFC in Mexico.
• A proof of address in the company's name, not the owner's. A utility bill in a personal name is grounds for rejection.
• Your own domain and a mailbox on that domain. A Gmail account does not work as a verifiable contact email.

The most common rejection has nothing to do with the company's line of business: it is that the legal name entered in the form does not match the one on the tax certificate letter for letter. "Clínica Dental del Sur" is not the same as "Odontología del Sur S.R.L.". Each correction and resubmission cycle costs between 5 and 15 days, and two rejections in a row turn a two-week project into a project of well over a month.

Two things worth separating from the launch timeline:

• The green check mark (official account) is a separate, optional process, with brand notability criteria. It is not needed to operate and should not condition the launch date.
• Messaging limits start low on a new number: the company can start a limited number of conversations every 24 hours, and that cap rises as the number builds history with a good quality rating. Conversations started by the customer do not consume that quota, so an inbound support bot is not affected. A mass outbound send is.

The WhatsApp number: the decision that delays more projects than it seems

There are three paths, and the one almost everyone picks first is the worst one to start with.

Migrating the number you already use. To connect it to the API you have to delete that account from the WhatsApp app: you lose the history that lives on the phone, the groups and the statuses, and going back requires removing the number from the API and registering it again in the app. Because it is irreversible in the short term, the owner postpones it "until after the end of the month" and the project sits still waiting for a decision that is not technical.

A new line. It costs whatever a prepaid line costs and it is sorted out the same day. The real downside is that customers already have the old number saved: you have to leave an auto-reply on it with the wa.me link to the new one and change it on the website, the email signature and the Instagram bio. Traffic takes 2 to 4 weeks to move over on its own.

A landline number. It is verified by voice call. It works, but someone has to be next to the phone at the exact moment of verification: in practice it adds a day of coordination.

The recommendation: start the pilot with a new line and decide on migrating the historical number once the bot is stable. Deleting the WhatsApp account of the number your revenue currently comes through is the last thing you touch, not the step that blocks registration in Cloud API.

The knowledge base and the access credentials: this is where most of the time is lost

A WhatsApp bot with RAG answers well about what exists in writing. What you need before building anything:

• The current price list (the current one, not last year's with a mental adjustment).
• Between 40 and 60 real frequently asked questions.
• Cancellation, shipping, warranty and payment method policies.
• Hours, addresses and the rules for when to hand off to a person.

The shortcut that saves a week: export the chat history from the phone (Settings, Chats, Export chat) for the last 3 to 6 months. That is where the questions customers actually ask come from, not the ones the owner thinks they ask. In most of the cases we see, between 5 and 8 topics account for the bulk of the queries, and the rest is a long tail the bot can hand off.

Access credentials have to be requested on day one, not on build day: a HubSpot private app token or a Pipedrive API key, a Google account with write permission on the calendar that will be used (the company's, not the personal one of whoever works the front desk), credentials for Tiendanube, Shopify or Mercado Libre if there is a catalog — Tiendanube and Mercado Libre are the dominant online store and marketplace platforms in LATAM — and the Chatwoot user or the inbox where the human handoff will land. Every credential missing on build day adds between 2 and 4 calendar days, because the back and forth to obtain it is never immediate.

What actually fits in a 72-hour launch (and what does not)

The 72 hours are implementation work, not paperwork. With the three keys ready, within that timeframe the following is operational:

• The WhatsApp number connected and answering, with RAG over the loaded knowledge base.
• Lead qualification with 3 to 5 questions and automatic entry into the CRM or a spreadsheet.
• Human handoff with explicit rules: keyword, intent unresolved twice, or outside business hours.
• Between 2 and 4 templates submitted for approval.
• A conversation dashboard so the team can see and take over any chat.

What does not fit, no matter the budget:

• The green check mark and high messaging limits. Both depend on history, not on configuration.
• Integrating a system without an API. If appointments live in desktop software with no accessible database, no integration is possible: it is either a system change or double data entry.
• RAG over documentation that does not yet exist in text.
• Payments inside the conversation when a Mercado Pago or Stripe account still has to be opened and verified — Mercado Pago being the dominant payment platform in LATAM. That onboarding has its own calendar.
• Fine-tuning the tone and the responses. That needs 2 to 3 weeks of real traffic. The day 3 bot is not the day 30 bot, and promising otherwise is selling smoke.

A day-by-day case: dental practice, from signature to pilot in 8 days

An illustrative scenario from the healthcare vertical, with figures typical of the sector. It is not the audited result of a specific company.

Starting point: three chairs, two people at the front desk, between 400 and 600 monthly queries through a single number, a little more than half of them outside business hours. They already had a Business Manager because they ran Instagram campaigns, but no business verification.

• Day 1, Monday. A 15-minute audit and scope definition: appointments, health insurance plans, prices for six procedures and emergencies. That same afternoon the business verification is submitted to Meta and the request for access credentials goes out.
• Day 2, Tuesday. They buy a new line. The number is registered in Cloud API, display name approved in about six hours.
• Day 3, Wednesday. They export four months of WhatsApp chats: around 1,900 conversations that distill into 58 distinct questions, of which six topics explain roughly 70% (the price of a cleaning, which health insurance plans they take, whether they handle emergencies on Saturdays, where to park, whether they do invisible orthodontics and how much the first consultation costs).
• Day 4, Thursday. Meta approves the business verification, three business days, documentation without objections. The flow is built: RAG over six documents, connection to the clinic's calendar and handoff to Chatwoot.
• Day 5, Friday. Internal testing with about 40 simulated conversations. It turns out the bot was quoting an orthodontics procedure the clinic no longer offers. The document gets fixed, not the prompt.
• Day 8, the following Monday. Launch in pilot mode. The old number is left with an auto-reply and the link to the new one.
• Days 8 to 30. Tuning with real traffic, three weeks: 14 responses nobody had anticipated get added and the handoff rule gets corrected, because any message containing the word "pain" escalated to a person and saturated the front desk at 3 in the morning with nobody on the other end.

Eight calendar days to the pilot, thirty to have it fine-tuned. If the verification had been rejected over the legal name, that would have added 5 to 15 days. If they had insisted on migrating the historical number, roughly as much again. If, instead of exporting the chats, the owner had written the responses "whenever he had time", that is where six-week projects come from.

What turns a 2-week project into a 2-month one

Ordered by how much each one costs, in calendar days:

1. Content without an assigned owner: 7 to 21 days. The most expensive and the most invisible. "Let the team review it" with no name and no date means the bot has nothing to answer with, and there it sits with the number already registered.
2. Business verification rejected: 5 to 15 days per cycle.
3. Access credentials held by third parties: 5 to 15 days. The CRM managed by a former provider, the store token left with the previous agency.
4. Scope that grows halfway through: 7 to 14 days. "While we're at it, have it issue the invoice too." The expensive part is not building it: it is that it reopens testing of the entire flow and, if new templates are needed, another round of Meta approval.
5. Approval of the response copy by committee: 5 to 10 days. Four people reviewing how the bot says hello.
6. Insisting on the number that is already in use: 3 to 10 days of postponed decision.
7. A system without an API. This is not measured in days. If the data the bot needs lives in closed desktop software, it is not a delay: it is a different project.

When it is better to hold off instead of rushing the launch

Going live in 72 hours is not always the right call. Cases where we recommend not accelerating, or simply not connecting the number yet:

• Fewer than about 50 queries per month. A WhatsApp chatbot does not move the needle at that volume. Saved quick replies and clearly communicated business hours are enough, and they cost nothing.
• Consultative high-ticket sales with few conversations. If out of eight conversations a month you close five, automating the conversation destroys value. Automate the reminder and the scheduling over WhatsApp, not the sales dialogue.
• Prices or policies that are not defined. The bot does not resolve that ambiguity: it exposes it, and at scale. If today the answer to "how much does it cost?" depends on who picks up, the version written down in the document becomes the official one for everyone who writes in.
• A management system migration in the next 60 days. The CRM and calendar integrations will have to be redone against the new system. Redoing them costs more than the weeks gained by launching earlier.
• An outbound campaign starting Monday. A new number starts with low limits on business-initiated conversations and no approved templates. Going out with mass sends on day three is the fastest way to get Meta to lower the number's quality rating.

In the 15-minute audit this is literally part of what gets reviewed: if the volume or the state of the content does not justify the project, it is better to say so before charging a setup fee, not after.

How long a custom chatbot takes (and what can be parallelized)

A custom WhatsApp chatbot, with process analysis, its own architecture and non-standard integrations, runs between 4 and 8 weeks end to end: 1 to 2 weeks of discovery, 2 to 4 of building and 1 to 2 of testing with real users before opening the number to the public. The difference with the fast launch is not in the engine, which is the same: it is that the custom build defines the process before automating it, and that takes people's time, not code time.

One single rule applies to both models: start the business verification with Meta the day you sign, even if you still do not know what the bot is going to answer. It costs nothing, it runs in parallel with everything else and it is the only timeline in the project that cannot be compressed. Projects that fall behind almost never fall behind because of the flow; they fall behind because the paperwork started late.

If you want to know the specific date it could be running in your case, the 15-minute audit is enough to tell you which of the three keys you are missing. Write to us at info@striqtech.com or book from striqtech.com/#servicios.