Most WhatsApp chatbots in e-commerce are not artificial intelligence. They are a short list of trigger words, one fallback message, and a person who picks up whatever the rules miss. That is less impressive than a demo and more useful than one: for a Shopify store, three well-chosen rules answer more messages than a clever bot nobody has trained.
- Three keywords are already in your Keaz account — Widerruf, Stop and Datenschutz. You can change what they answer, not whether they exist.
- One fallback, capped. The Default Reply goes out at most once every 24 hours per contact, and only when no keyword matched.
- 24 hours is the design constraint. WhatsApp's customer service window resets with every customer message. Outside it, only a template goes out.
So what is a WhatsApp chatbot?
The word covers two different things, and the difference decides what you can promise a customer.
A rule-based bot matches what the customer wrote against a list you defined. The message contains "Retoure", so your return policy goes out. It has no memory of the previous message and no opinion about anything you did not anticipate.
A conversational bot generates an answer instead. It copes with phrasing you never thought of — and it also invents delivery times when it has nothing better to say.
Keaz sits firmly in the first camp. Inbound messages are handled by rules: keyword replies, one fallback, and your Inbox. We would rather say that plainly than sell you something that improvises about your shipping. Here is what Keaz does on the inbound side.
Which questions should the rules answer?
Start from your Inbox, not from a template gallery. The questions worth automating are the ones you have already answered more than ten times this month — usually delivery time, the return process, where an order is, and sizing.
A keyword in Keaz has an internal name, one or more trigger words, and one stored answer. When an incoming message contains a trigger word, that answer goes out and nothing else does. The help center has the detail: how trigger words are set up.
Three keywords ship with every account and cannot be removed: Widerruf, Stop and Datenschutz. Rewriting what they say is worth an afternoon. They are the messages your least happy customers read, and a curt answer there costs you the contact and the goodwill in one go. More on that here: setting up automatic replies with keywords.
What happens to everything the rules miss?
The order is fixed. Knowing it stops you building two automations that fight each other:
- The customer sends a message.
- Keaz checks it against your trigger words. A match wins, and that keyword's answer goes out.
- No match, and the Default Reply goes out — if you switched it on, and at most once every 24 hours per contact.
- The Default Reply is off, so nothing is sent. The message waits for you in the Inbox.
A keyword always beats the fallback, so the two never answer the same message. The help center covers the order: how the three automations fit together.
Why the 24-hour window decides the design
Every reply you write freely lives inside WhatsApp's customer service window. Meta's own documentation is blunt about it: non-template messages "can only be sent in an open 24-hour customer service window, which opens and resets with each user message. Thus non-template messages can only be used to respond - not reach out - to users" (Meta, WhatsApp Business Platform documentation, retrieved 2026-09-02).
Two consequences for anyone planning a bot. Your rules can only ever answer — they cannot start a conversation. And once the window has closed, the only thing that goes out is a template, because templates are the one message type sendable outside the window (Meta, retrieved 2026-08-24).
There is also a date worth putting in the calendar: from 1 October 2026, Meta charges for non-template messages that were free inside the window until then. What changes and what stays is a subject of its own: what changes for service messages on 1 October.
When should a person take over?
Sooner than a demo suggests. A rule-based setup is good at the first message and weak at the second. The moment someone replies "yes, but my order is 4711 and it says delivered", no keyword helps.
That is why the fall-through in step 4 matters more than the automations above it. An unanswered message that lands visibly in a shared inbox is a working system. An unanswered message that a bot has smothered with a cheerful non-answer is not — the customer thinks they have been dealt with, and you never find out they were not.
If more than one person answers your number, that only works when everyone can see what is already handled. More on that: how small teams share one WhatsApp number.
Common mistakes
- Calling it AI. If your bot matches words, say so. Customers forgive a rule. They do not forgive being told a machine understood them when it did not.
- Overlapping trigger words. Two keywords both containing "retoure" means the first match wins and the second silently never fires. Keep triggers distinct, and keep them few.
- No fallback at all. With the Default Reply switched off, someone writing at 22:00 gets silence. Silence is a legitimate choice — make it on purpose, not by omission.
- Confusing a delay in a flow with the reply window. A 24-hour Delay block inside an automation is your timing. The 24-hour window is Meta's. Two unrelated numbers that happen to match.
- Expecting the bot to reach out. Rules answer. Starting a conversation means a template, which is a separate piece of work with its own rules.
A WhatsApp chatbot that works in a shop is unglamorous: five trigger words, one fallback, a person on the rest, and a design that respects the 24-hour window. That is also why it can be live this week instead of next quarter.
What would that be worth in your shop? The Forecast runs the numbers on your own data. Calculate your Forecast.
Frequently asked questions
Does Keaz have an AI chatbot?
Can a chatbot message a customer first?
How many keywords should I start with?
What if nobody replies within 24 hours?
Nils Spölgen
Founder at Keaz. Serial founder in chat marketing — built a local agency into Keaz, the WhatsApp marketing platform for Shopify. Bootstrapped the MVP, raised funding, and scaled to 200+ merchants in DACH. Writes about flows, opt-in & GDPR, Klaviyo, and realistic revenue benchmarks — prefers conservative math over impressive claims. View profile



