The three jobs a WhatsApp bot actually does well
A WhatsApp Business chatbot earns its keep on three narrow jobs. It books and confirms appointments. It replies with order status and tracking numbers. It screens incoming messages and routes humans to the right person. Everything else — open-ended sales, complex negotiation, anything that requires judgment calls — is a money pit.
The 80/20 rule holds hard here: roughly 80% of your inbound WhatsApp messages are five to six repeat questions. What time are you open? Where's my order? Can you hold this for me? A bot trained on these patterns handles them in under ten seconds. But the moment someone pitches you a "24/7 AI salesperson," walk away. Customers know the difference between a kiosk and a person. The bot that tries to close deals gets blocked.
Scope the bot to those three jobs — bookings, status lookups, triage — and it pays for itself in the first month. Expand it beyond that and you're paying platform fees to frustrate your customers.
WhatsApp's rules: the 24-hour window, templates, and why they matter
WhatsApp is not an open SMS channel. Meta's Business Platform enforces a 24-hour conversation window: once a customer sends you a message, you have 24 hours to reply. After that, you cannot send anything without using a pre-approved message template.
Templates are critical. They're static, branded reply patterns — Your order #[order_id] shipped on [date]. Track it at [link]. — that Meta approves in advance. This is not a workaround; it's the only legal way to message customers who have gone silent. If you want to send anything outside the 24-hour window, it must fit a template and fall into one of four conversation categories: utility (order updates, booking confirmations), marketing (promotions, newsletters), authentication (verification codes), or service (support escalations).
Utility and service conversations are free in most countries. Marketing conversations carry a per-conversation fee — typically €0.01–€0.05 depending on region. Authentication is free but limited to verification flows.
Opt-in is non-negotiable. A customer must explicitly agree to receive messages from you on WhatsApp. This kills cold outreach entirely. You're building a channel for existing customers, not a new acquisition funnel.
There are two ways to connect: the WhatsApp Business App (free, limited to one person, no API) or the WhatsApp Business Platform Cloud API (requires a Business Solution Provider like Twilio or 360dialog, starts around €20–€50/month plus per-conversation fees). You need the Cloud API for any bot with integrations or scale. The Business App is a dead end the moment you want automation.
What a realistic build costs — and what drives the number
We've quoted enough WhatsApp bots to give you honest ranges. A scoped FAQ bot with a handoff to a human runs €2,000–€4,000 to build. Add integrations — Shopify for orders, Google Calendar for bookings, HubSpot for lead routing — and you're at €6,000–€15,000. These are one-off build costs, not maintenance.
Monthly running costs stack like this: your Business Solution Provider charges a base fee (Twilio: €25–€50/month; 360dialog: €20–€100/month depending on volume). Meta's conversation fees add up: zero for utility messages, €0.015–€0.05 per marketing message. LLM tokens (if you're using Claude or GPT for intent routing) run €0.001–€0.01 per message depending on model and volume. A bot handling 1,000 messages/month costs €40–€80 in platform and LLM fees combined.
Our guide to AI automation costs in 2026 breaks down the real numbers — build, monthly run, and when a fixed-price contract beats hourly. Hidden costs matter: template approval can take 24–72 hours (not instant); Business Manager and phone number verification add one to two days of setup; if your integrations are flaky, you're debugging on someone's hourly rate.
What separates a good build from a bad one
A good build hands off to a human in the same WhatsApp thread. A bad one sends the customer to a support email or a web form. That's friction. The customer should never leave WhatsApp.
Integration matters. Your bot should read from the system of record: Shopify or WooCommerce for orders, Calendly or Google Calendar for bookings, HubSpot or Pipedrive for lead data. If it's guessing or making up order numbers, you're selling a toy.
Idempotency and retries prevent double-bookings and duplicate charges. If a webhook fails halfway through — your calendar API times out, your bot sends a confirmation but the booking doesn't actually save — the system should retry cleanly. Most bad builds ship without this, and the cost emerges in month two when a customer books the same slot twice.
Logged conversation history is non-negotiable. Your ops team needs to audit every message — what the bot said, why it said it, where it escalated. This is compliance, not nice-to-have.
Guardrails matter. A good bot refuses to quote prices it can't verify. It won't promise "we'll ship tomorrow" unless it's actually checked inventory. If the LLM gets unsure, it falls back to a human within 30 seconds, not five minutes.
For customer-facing AI assistants and agents for business, these patterns—clean handoff, verified integrations, guardrails—are what turn a demo into something ops can rely on.
Who owns it on Monday morning
A WhatsApp bot is a product, not a one-time project. Someone on your team owns it: usually the ops lead, not IT.
Their job: review flagged conversations daily (the bot logs anything it couldn't confidently answer). Monthly, walk the top 5–10 intents and check if template language is landing or if customers are getting confused. Decide whether to tweak the template wording or retrain the intent classifier. Update templates faster than you update the code.
The KPIs that matter: deflection rate (what % of common questions the bot fully handled), time-to-first-reply (usually under 10 seconds), handoff rate (what % went to a human), and CSAT after bot interactions (do customers feel helped or stalled?). For a well-scoped bot in the first 90 days, expect 50–70% deflection on repeat questions, nearly instant first replies, and 70–80% of handoffs resolved in that same WhatsApp thread.
When it breaks, it's usually not the bot—it's the integration. An order system that goes offline, a booking API that starts returning wrong data, a template that violates WhatsApp's policy and gets suspended. This is why process automation for SMBs requires a system of record you trust. The bot is only as good as the data it reads.
When a WhatsApp bot is the wrong answer
Low message volume (fewer than 200/month) means a shared inbox is cheaper and faster than a bot. Hire one person to reply, not an LLM.
Highly regulated advice — legal questions, medical diagnosis, financial recommendations — should not run through a bot, even with guardrails. The liability is yours, and WhatsApp's terms don't protect you.
Complex B2B sales cycles where every deal is bespoke require negotiation, not automation. A bot can't close enterprise contracts.
If your order or booking system has no API — it's a paper spreadsheet or an old desktop system — fix that first. A bot built on top of that is like building a house on sand.
If your real bottleneck is staffing, not response time, a bot won't save you. You need more people, not a chatbot.
The next step: scope it in a week
If you can name your top five repeat questions and point to a Shopify store, a Google Calendar, or an order system with an API, a WhatsApp bot makes sense. We've scoped dozens of these in a week: audit your current messages, map the intents, design the templates, estimate the cost.
Reach out at /contact and we'll scope the build for you — timeline, cost, what deflection rate you can actually expect. No hype, no promises about AI salespeople. Just three jobs done well and an ops lead who owns it.