Skip to content
PM Master Handbook
⬇ PDF
Module 21

AI Chatbots for Business: WhatsApp, Slack and Web

Design, build, test and price AI chatbots that answer customers on WhatsApp, help employees in Slack, and sit on a website, with safe handoff to people.

≈ 8 h studyDraftedSources: Meta WhatsApp Business Platform docs and pricing page; BSP documentation; n8n WhatsApp, Telegram and Slack node docs; Module 10 (RAG) and Module 12 (agents); handbook dataset support_eval_set.csv; checked 23 Oct 2026

On this page
  1. What a business chatbot does
  2. WhatsApp: App or Platform?
  3. How an AI WhatsApp bot works
  4. No-code options
  5. Slack, Telegram and website bots
  6. Design the bot before you build it
  7. Guardrails, privacy and quality
  8. Costs at a glance (illustrative)
  9. Labs
  10. Exercises with model answers
  11. Key takeaways
  12. Check your understanding
  13. Recommended resources

Messaging rules and prices change

Meta updates WhatsApp pricing and policies several times a year. Always check the official pricing page and your provider's rates before launch. Examples use hypothetical businesses; screens are illustrative wireframes.

21.1 What a business chatbot does

A business chatbot answers questions and completes simple tasks in a chat channel: order status, booking an appointment, store hours, return policy, lead capture. An AI chatbot uses a language model (such as Claude) plus your business's knowledge and tools, so it can understand free-text questions instead of only menu buttons.

WhatsApp
  • Where Indian customers already are
  • Orders, bookings, reminders, support
  • Needs the Business Platform for automation
Website chat widget
  • Pre-sales questions, lead capture
  • Fully under your control
Instagram / Messenger
  • D2C brands, creators
  • Same Meta platform family
Slack or Microsoft Teams
  • Internal bots: HR policy, IT help, sales enablement
Telegram
  • Free and quick to set up
  • Perfect for practice and communities
Voice (phone)
  • Calls instead of chat
  • See Class 22
Channels and when to use them. The AI part is the same across channels. What changes is the channel's rules, costs and setup.

21.2 WhatsApp: App or Platform?

WhatsApp Business AppWhatsApp Business Platform (API)
Who it is forSmall shops replying by handBusinesses automating or scaling
CostFree appPer delivered template message, plus any provider fee
AutomationGreeting, away message, quick replies onlyFull bots, AI, integrations
UsersPhone plus a few linked devicesUnlimited agents through a provider's inbox
SetupDownload and verify the numberMeta business portfolio, number, display name, templates
AI chatbotNot supported directlyYes, via webhooks to n8n, a provider or your code
WhatsApp Business App vs WhatsApp Business Platform. For an AI chatbot you need the Business Platform, used directly through Meta's Cloud API or through a Business Solution Provider.

21.2.1 Direct Cloud API or a Business Solution Provider (BSP)?

  • Meta Cloud API directly: no middleman fee; you connect it to n8n or your own code. Best when someone technical owns it.
  • A BSP (for example Twilio, Gupshup, Interakt, WATI, AiSensy, 360dialog): onboarding help, a shared team inbox, broadcast tools and no-code flow builders; they add a monthly fee or a markup per message.

21.2.2 What Meta requires

The 24-hour customer service window. When a customer messages you, a 24-hour window opens in which you can reply freely. Outside that window, a business can only start the conversation with an approved template.

21.2.3 Pricing basics (checked 23 Oct 2026)

  • Since July 2025 Meta charges per delivered template message. The rate depends on the template category (marketing, utility, authentication) and the recipient's country. Marketing costs the most; authentication (OTP) the least.
  • Rates differ by country. Several 2026 guides quote India authentication messages at a fraction of a rupee and marketing messages at a few times that. Use Meta's current rate card, not a blog.
  • Independent guides report that from 1 October 2026 Meta began charging for some messages inside the 24-hour window that used to be free (service and utility replies). Confirm this on Meta's pricing page before you forecast costs.
  • Add the BSP's fee and the AI model's cost (a few paise to a few rupees per conversation, depending on length and model).

21.3 How an AI WhatsApp bot works

21.4 No-code options

Tool What it is Good for
BSP flow builders (Interakt, WATI, AiSensy, Gupshup) Menus, forms and AI replies inside the provider Fast WhatsApp launch for SMBs
Botpress Visual bot builder with AI and many channels Multi-channel bots with knowledge bases
Voiceflow Conversation design and AI agents, chat and voice Teams designing complex flows
Chatbase Train a bot on your website or documents Website FAQ bot in minutes
Intercom Fin, Zendesk AI AI agents inside support suites Companies already on those helpdesks
n8n + Claude Build exactly what you need Custom tools, privacy, low running cost

21.5 Slack, Telegram and website bots

  • Slack internal bot: create a Slack app with the Events API and bot scopes (app_mentions:read, chat:write), connect it to n8n's Slack Trigger, and answer from your HR or IT policy documents. Restrict it to the channels it's invited to.
  • Telegram practice bot: message @BotFather, send /newbot, copy the token, add it to n8n's Telegram Trigger. It's free and takes ten minutes, ideal for learning before WhatsApp.
  • Website widget: the "Ask us" button from Class 16 (Lovable server function + Claude), or a Botpress or Chatbase embed code pasted into your site.

21.6 Design the bot before you build it

The one-page bot spec

  1. Job: the top 10 questions and 3 tasks the bot handles (from real support tickets).
  2. Not allowed: medical, legal or financial advice; refunds above a limit; anything about other customers.
  3. Knowledge: the exact documents and their owner; how often they're updated.
  4. Tools: each action with inputs, and whether it needs confirmation.
  5. Handoff: when, to whom, within what time, and what the customer is told.
  6. Tone and language: for example English and Hindi, friendly and short.
  7. Privacy: what data is stored, for how long, and the opt-out keyword ("STOP").
  8. Success metrics: containment rate, resolution rate, handoff rate, CSAT, cost per conversation.

21.7 Guardrails, privacy and quality

Rules that protect customers and the business

  • Answer only from the knowledge base; say "I'm not sure, let me connect you" otherwise.
  • Confirm before acting ("Shall I move it to Friday 11:00?").
  • Never collect card numbers, passwords or OTPs in chat.
  • Get opt-in before marketing messages and honour opt-outs immediately.
  • Follow data-protection law (in India, the DPDP Act): collect the minimum, explain the purpose, delete on request.
  • Log conversations for quality review, with access limited to staff who need it.

21.7.1 Evaluate before launch

Use datasets/support_eval_set.csv from this handbook (or 50 real anonymised questions): run every question through the bot, then score correct, safe, handed off when it should, tone. Fix the knowledge base or instructions, and re-run until it passes your bar (see template 09 for the evaluation plan).

21.8 Costs at a glance (illustrative)

Item Typical cost driver
WhatsApp template messages Per delivered message by category and country
BSP Monthly plan or per-message markup
AI model Per token; short FAQ chats are cheap, long ones cost more
n8n Free self-hosted (server cost) or a cloud plan
People Staff to handle handoffs and keep knowledge current

21.9 Labs

Lab 1 · Telegram FAQ bot (free, 40 minutes)

  1. Create a bot with @BotFather and copy the token.
  2. n8n: Telegram Trigger → AI Agent (Claude, or Gemini free tier) with instructions and a 15-line FAQ pasted into the system prompt → Telegram "send message".
  3. Test ten questions, including two it should refuse.

Lab 2 · WhatsApp test number (free test, 45 minutes)

  1. Create an app at developers.facebook.com, add the WhatsApp product, and use the provided test number. Add your own phone as a recipient.
  2. n8n: add a WhatsApp credential (access token and phone number ID) and a WhatsApp Trigger; register its webhook URL in the Meta dashboard.
  3. Reuse the Lab 1 AI Agent and reply with the WhatsApp node.
  4. Message the test number from your phone and check the reply.

Lab 3 · Knowledge base and a booking tool (45 minutes)

  1. Put FAQs in a Google Sheet (question, answer, category). Give the agent a Sheets tool to look them up.
  2. Add a Bookings sheet and a tool find_booking(phone); then reschedule_booking(id, new_time) with confirmation required.
  3. Test the reschedule conversation from the phone picture above.

Lab 4 · Human handoff (20 minutes)

Add an IF node: if the agent sets handoff: true (complaint, health issue, low confidence), post the conversation to Slack #practice-support and send the customer the handoff message.

Lab 5 · Slack HR policy bot (30 minutes)

Create a Slack app with Events and the scopes above, connect n8n's Slack Trigger, and answer from a 2-page policy document. Mention the bot in a channel and ask three questions.

Lab 6 · Evaluate (30 minutes)

Run 30 questions from support_eval_set.csv through your bot with a loop in n8n, write answers to a sheet, and score them. Fix the top two failure patterns and re-run.

21.10 Exercises with model answers

Exercise · Template or free reply?

(a) A customer asked about delivery 2 hours ago; you reply with tracking. (b) You want to tell all past customers about a Diwali sale. (c) You send an OTP to log in.

Model answer

(a) Inside the 24-hour window, so a normal reply (check whether your country's current rules charge for it). (b) A marketing template, approved by Meta, only to opted-in customers. (c) An authentication template.

21.11 Key takeaways

  • AI chatbots need knowledge, tools, guardrails and a handoff to people.
  • WhatsApp automation needs the Business Platform (direct Cloud API or a BSP), approved templates, the 24-hour window and opt-in.
  • Pricing is per delivered template message by category and country, plus provider and AI costs.
  • Practise on Telegram for free, then use Meta's test number before going live.
  • Evaluate with a question set and track containment, handoffs and CSAT.

21.12 Check your understanding

Check your understanding

Which WhatsApp option supports an AI chatbot connected to n8n?

  1. WhatsApp Business App
  2. WhatsApp Business Platform (Cloud API or BSP)
  3. Personal WhatsApp
  4. WhatsApp Web
Why: Automation and webhooks require the Business Platform.

A business wants to message a customer who last wrote 3 days ago. What's needed?

  1. Nothing
  2. An approved template message
  3. A voice call
  4. A new phone number
Why: Outside the 24-hour window, businesses can only start with an approved template.

A customer describes a worsening injury to a clinic bot. The bot should…

  1. Give treatment advice
  2. Hand off to a person and give emergency guidance
  3. Ignore it
  4. Offer a discount
Why: Health concerns are out of scope for the bot and must reach a qualified person.
← PreviousModule 20: Automation with Zapier, Make and n8nNext →Module 22: Voice Agents for Businesses