How to set up automated WhatsApp responses for your business - step by step

What this guide covers, and what you need before starting

This is a hands-on setup guide. By the end you'll have an automated WhatsApp response system running on a real number, handling real customer messages, with humans on standby for the stuff AI shouldn't touch. It takes about a working day end-to-end: 2 hours of active work plus a 2-4 hour Meta verification wait you don't have to babysit.

Before you start, gather three things. A WhatsApp Business account, or a phone number you're willing to register as one - it can be mobile or landline, but it must not be currently registered to a personal WhatsApp. Facebook Business Manager access - if you've ever run Facebook or Instagram ads for the business, you already have it; if not, creating one takes about 4 minutes. And your business verification documents: business registration certificate, a utility bill or bank statement at the registered address (within the last 90 days), and a contact person's full name as listed on the registration.

If you're missing any of those, stop here and collect them first. The biggest cause of multi-day delays in WhatsApp verification is starting the process with mismatched documents - Meta rejects, you re-upload, you wait again. Get the documents right once and the whole flow takes a few hours.

We'll assume throughout this guide that you're connecting WhatsApp to Seekadu, because that's the platform we know best. The same six steps apply to any BSP - the screens and labels differ, but the order and the prerequisites are identical.

WhatsApp Business App vs WhatsApp Business API - which one you actually need

Meta offers two products for businesses on WhatsApp. They look similar at a distance but they're different stacks, different costs, and different capabilities. Most local businesses end up on the API; some don't need it. Here's the decision rule.

The WhatsApp Business App

A free mobile and desktop app, exactly like personal WhatsApp, with a business profile, labels, a quick-reply library, and the old “away message” feature. One device, one user, no AI, no API access. It's fine if you're a solo operator answering fewer than 100 messages a week and you don't mind that the conversation lives on one phone.

The WhatsApp Business API (Cloud API or On-Premises)

A programmatic interface to WhatsApp - what every modern customer messaging platform connects to. Multiple agents can share one number, AI can reply automatically, conversations sync to a CRM or inbox, and you get analytics. Cloud API is hosted by Meta and recommended for almost every business under 100,000 monthly users; On-Premises API is a legacy self-hosted version Meta is slowly deprecating. You won't touch the API directly - you go through a BSP like Seekadu.

The 60-second decision rule

Under 100 messages a week, one person handling them, no need for AI replies or 24/7 cover - the WhatsApp Business App is fine. Stop reading this guide and keep using the app. Over 100 messages a week, more than one person involved, or you want any of: AI auto-replies, after-hours coverage, multi-agent inbox, booking capture, CRM sync, analytics - you need the API, and that means you need a BSP. Continue to Step 1.

Step 1: Set up WhatsApp Business and verify your business

This is the longest single step - not because it's hard, but because Meta sits in the middle. Two screens of yours, then a verification wait. Plan to do this in the morning so the wait happens during business hours.

1.1 Create or claim your Facebook Business Manager

Go to business.facebook.com. If you've ever boosted a post for the business, your Business Manager already exists - log in and confirm it lists your business name. If not, create one: business name, your business email, and your role inside the company. This takes 3-4 minutes.

1.2 Add the business phone number

Inside Business Manager, go to Business Settings → WhatsApp Accounts → Add. Enter the number you want as your business number. Two important rules: it cannot currently be on a personal WhatsApp account (you'll be prompted to deregister), and it must be able to receive an SMS or voice call for the one-time code. Landlines work - they verify by voice call.

1.3 Submit business verification

Meta asks for proof your business is real. Upload your business registration certificate (or equivalent - sole-trader registration is fine for most countries), a utility bill or bank statement at the registered address dated within the last 90 days, and a contact person's name. The names on the documents must match the Business Manager profile and each other - this is where most rejections happen.

1.4 Wait

Verification takes 2-4 hours typically; up to 2 business days if your documents need a manual review. You'll get an email when it's done. While you wait, move to Step 2 - you can configure most of the Seekadu side without verification finished.

Step 3: Build your first auto-reply flow - welcome + the three most common questions

With the connection live, you're ready for the first conversation flow. Don't try to handle everything on day one. Handle the welcome and the three questions your business gets most often, beautifully - then expand.

3.1 Write the welcome message

The welcome message fires on the first message of every new conversation. Keep it short - two sentences max - and name the business so it doesn't sound like a generic bot. For a Sofia restaurant: “Hi! You've reached Manastirska Magernitsa - I can help with reservations, the menu, or anything you need before your visit.” For a salon: “Hi! You're chatting with Beauty Studio Vitosha - I can book you in, share prices, or answer anything about treatments.”

3.2 The three most common questions

Every local business gets roughly the same three repeat questions, by industry. For restaurants it's “are you open tonight?”, “do you have a table for X at Y?”, and “do you have gluten-free options?”. For salons and beauty it's “how much is a haircut/colour?”, “do you have availability tomorrow?”, and “how long does X take?”. For healthcare and clinics it's “what insurance do you take?”, “when's your first appointment?”, and “do I need to bring anything?”. For festival software it's “is glamping still available?”, “what time do gates open?”, and “can I re-enter on a day pass?”.

3.3 Train the AI on those three answers

For each of the three questions, write the answer the way you'd say it out loud, in two or three sentences. Don't write a script - write an answer. The AI uses your wording as a tone reference and will improvise around it for follow-ups. Restaurants: “Yes, we're open tonight until 23:00. We have a few tables free around 21:00 - would that work?” Salons: “A women's haircut is 45 lev, men's is 30. I have tomorrow at 14:30 and 17:00 - does either suit?”

3.4 The fallback

If the customer asks something the AI isn't confident on, the fallback should never be “I don't understand.” The fallback is: “Let me get a colleague to confirm - they'll be back to you shortly.” And the conversation flags into your inbox for human pickup. This is also the right default for anything you haven't explicitly trained yet - better a 10-minute human reply than a wrong AI one.

Step 4: Set up booking capture inside WhatsApp

Most of the revenue from automated WhatsApp comes from one specific job: capturing bookings inside the conversation, without kicking the customer to a separate form or website. The four fields below are non-negotiable for any local-business booking.

4.1 The four required fields

Name.Single field, full name. Don't overthink it - “Maria K.” is fine for most businesses, “ Maria Kostadinova” is better. Contact. The phone number is the WhatsApp number by default - but ask explicitly if you also need an email for a confirmation. Date and time.Specific, not “sometime tomorrow.” The AI offers slots; the customer picks one. Service.What are they booking? “Table for 4”, “women's haircut”, “30-minute consultation.” This is the slot type, not a description.

4.2 The booking flow in conversation

The flow is: customer expresses intent → AI offers two or three specific slots → customer picks → AI captures the missing fields in one or two follow-up turns → AI confirms with the full details. Six turns max for a booking. If you find the flow running longer than that, the prompts are too long - tighten them.

4.3 Connect to your booking system

The captured booking has to land somewhere. For restaurants - OpenTable, Resy, SevenRooms, Tock, or a Google Calendar. For salons - Fresha, Booksy, Schedulista. For clinics - SimplePractice, Jane, your EMR. Seekadu writes to any of these via the AI booking system; if your platform is more custom, a Google Calendar fallback covers 80% of cases on day one and you can wire up the real system later.

4.4 The 24-hour confirmation

Once the booking is in, schedule a confirmation message for 24 hours before the appointment using a template message. This is where you reduce no-shows. Industry baseline no-show is ~20%; with a conversational 24-hour confirmation, expect 12-14%. The customer replies “yes”, “reschedule”, or silence - the AI handles all three.

Step 6: Set escalation rules - when the AI must call a human

The escalation rules matter as much as the conversation flow. Get them wrong and either the AI bothers you for nothing - or, worse, doesn't bother you for something that needed a human. Three tiers; you can copy these and adjust the keywords.

Tier 1 - 60-second escalation

Triggers a push notification to the on-call manager's phone within 60 seconds, with the full message thread. Use it for: complaints (keywords like “refund”, “terrible”, “raw”, “rude”, “sick”), severe allergy mentions, press or media enquiries, anything mentioning a previous illness or injury, and high-value bookings (parties over 12 for restaurants, full wedding-party bookings for salons, anything urgent for clinics).

Tier 2 - 30-minute escalation

Lands in your team inbox to be picked up within 30 minutes. Use it for: corporate enquiries, group bookings over 8, requests outside your normal capacity, deposit disputes, same-day reschedule requests, anything where the AI's confidence score is below the threshold.

Tier 3 - handle without escalation

The AI handles cleanly, no human notification. Standard bookings under 8 covers, routine menu or service questions, hours and location queries, cancellations inside the policy window, gift voucher enquiries, “are you open today?”.

The principle

Anything that affects revenue at scale, anything with a public reputation risk, and anything with a medical or safety implication goes to a human. Everything else, the AI handles. If you find yourself debating where a rule goes - put it in Tier 2. Better a slightly cautious AI than one that mishandles a complaint.

Common mistakes in the first week

Four mistakes account for roughly 80% of first-week problems. None of them are technical - they're all small editorial or policy errors that are easy to fix once you know what to look for.

1. Letting the AI sound robotic

“Thank you for your enquiry. We will respond to you shortly.” is the death-knell of an AI auto-reply. Customers instantly disengage. The fix is the tone-reference paragraph in Step 5 - re-read it after Day 2 and tighten it. The AI mirrors what you write, so if your tone reference sounds corporate, so will the AI.

2. Missing the last-appointment / last-seating rule

The AI will happily accept a 21:45 booking on a Tuesday when your kitchen closes at 21:00 - unless you tell it the last seating rule. The same applies to salons (“last appointment 30 minutes before close”) and clinics (“no new patient appointments after 16:00”). Take five minutes to list every day's last-slot time.

3. Forgetting weekend hours

Saturday hours often differ from weekday hours, and Sunday hours differ from both. The most common Day 3 ticket is “the AI told a customer we were closed on Saturday and we're not.” Walk through the hours field again before going live; the AI is only as right as the data.

4. No handoff path for complaints

If a Tier 1 escalation has no destination - no manager phone, no on-call inbox - the AI will quietly try to handle the complaint, badly. Verify your Tier 1 destination is correct on Day 1. Send yourself a test complaint and confirm the notification arrives.

What success looks like in 30 days

Don't judge the AI on Day 3. Judge it on Day 30. Three measurable outcomes tell you whether automated WhatsApp is working for your business.

1. A volume baseline you can read

By Day 30 you should see a clear weekly volume: total conversations, AI-handled vs human-handled split, peak hour patterns. For a typical local business this lands at 80-300 conversations per week, with 60-80% AI-handled end-to-end and the rest escalated or taken over manually.

2. Captured-after-hours rate

Of all bookings the AI took in 30 days, what percentage came in during hours your team wasn't working? For most local businesses this is 30-50% - bookings that would otherwise have died in an unread DM at 22:47. If this number is below 20%, customers either don't message you outside hours, or your AI is over- escalating after-hours messages instead of handling them.

3. Handoff accuracy

Of the escalations the AI sent to humans, what percentage actually needed a human? Healthy range: 80-95%. Lower than 80% - the AI is escalating too cautiously, train it on more edge-cases. Higher than 95% - fine, but watch for false negatives (cases the AI handled that should have escalated). Spot-check 10 random AI-handled conversations each week to catch these.

The kill-switch is always there

One more thing: at any point in any conversation you can tap “take over” and reply as yourself. The AI pauses. This is not a failure mode - it's how the system is designed. Use it whenever the conversation matters. The AI is your team's cover for the easy stuff so they can show up properly for the hard stuff.