WhatsApp automation for ecommerce is usually sold as a chatbot that will handle your customer service. That framing produces disappointing projects, because it starts from the technology rather than from the inbox.
Start from the inbox instead. Read the last two hundred messages your store received and sort them by what they are actually asking. Nearly every store that does this finds the same shape: one question dominates, a handful of others recur, and a long tail needs a person.
The one question worth automating first
It is almost always "where is my order". Depending on the store it is between a third and well over half of everything received, it has a definite answer, and that answer lives in a system you already have.
Automating it well is not a chatbot conversation. It is: recognise the intent, identify the customer, look up the order, reply with the status and the tracking link. No personality, no menu tree, no attempt to be charming. The customer wants one fact.
Do that and you have removed a large share of the volume with something that cannot really go wrong. Every other automation on the list is worth less and carries more risk, which is why this is the one to build first and measure before doing anything else.
What else is worth doing, roughly in order
Abandoned cart recovery, where the mechanics are permission rather than technology — you need the number and consent to message it, which has to be collected properly at checkout rather than assumed.
Delivery notifications, which are proactive rather than reactive and reduce the first question before it is asked. A store that tells people when the parcel moves receives far fewer messages asking.
Returns initiation, if your returns policy is genuinely simple. If it has conditions and judgement calls, automating the start of it produces a customer who believes they have started a return they have not.
Product questions, which sound like the obvious use and are the weakest. Answers vary by product, change with stock, and are exactly where a confident wrong answer costs you a sale and the trust behind it.
The rules that catch operators out
WhatsApp's business platform is not an open messaging channel and it is worth knowing the shape of the constraints before you plan around them.
Outside an open conversation window, you may only send pre-approved message templates, and approval is a real process with real rejections — write them early rather than discovering the rule at launch. Once a customer messages you, a window opens in which free-form replies are allowed; most automation lives there.
Messaging people who have not opted in is both against the rules and the fastest way to have your number rated poorly by recipients, which degrades delivery for everything else you send. Consent is not a formality here; it is the mechanism the channel is policed by.
Collecting consent without ruining checkout
Everything above depends on having a number you are allowed to message, and that is collected at checkout or not at all. The common mistakes are equal and opposite: burying it so nobody opts in, or pre-ticking it so everybody does and half of them later report the messages as unwanted.
What works is an unticked box with a plain description of what you will actually send — order updates, not marketing, unless you mean marketing and say so. Opt-in rates on a clear, honest checkbox are usually better than people expect, because customers genuinely do want to know where the parcel is. And a smaller list of people who agreed beats a larger one that damages how your number is rated.
The line not to cross
Automation should never make it hard to reach a person. The pattern customers hate — and the reason the whole category has a poor reputation — is a bot that loops, misunderstands, and offers no exit.
Build the exit first. Any message the automation does not confidently recognise goes to a human, and the customer is told that is what happened. Any message containing frustration goes to a human immediately. Anything touching money goes to a human until it has earned trust, and some of it stays that way permanently.
An automation that handles half the volume and hands the rest over cleanly is worth far more than one that attempts everything and traps people, and it is also considerably cheaper to build.
How to tell whether it paid
Measure the number of conversations a person had to handle, before and after. Not "messages automated", which counts the wrong thing and always looks good — the number you care about is human hours reclaimed.
Then watch whether resolution time went down or up. A badly-built automation increases it: the customer spends four minutes with the bot and then reaches a person anyway, having become annoyed on the way. That shows up here and nowhere else, which is why this is the metric to agree before anyone builds anything.
Full disclosure, since it is relevant: we build these and we also own ReploChat, a product in this space, which was built because clients kept asking for the same thing. Treat our enthusiasm accordingly. The bespoke version of this work is described on the WhatsApp page, and the broader argument about what is worth automating at all is on the automation page.
