The same three questions, a thousand times: what support really costs

Open the support inbox of an ecommerce store on any given day and you find the same three questions, written twenty different ways. Where is my order. How long does it take. Will this size fit me. Whoever answers already knows, and answers anyway, because a customer is waiting.

It looks like unavoidable work. It is not, but to stop doing it you first have to understand why it repeats.

If the question keeps coming back, the customer is not the problem

A question that arrives a hundred times a month is not a distracted customer: it is information that is not where the doubt appears. Delivery times live in a help page, the returns policy in the footer, the size guide behind a link that opens in a window. All present, none where they are needed.

The support message is the symptom. And the person who writes is the lucky part of the problem: for every one who writes there are many who simply close the page, and you will not see them in any inbox.

The cost that is on no invoice

The bill has three lines, and only the first is obvious.

Time. Answering the same question by hand is work that leaves nothing behind. It does not improve the store, it does not reduce tomorrow's questions, it has to be done identically again next week.

Waiting. Hours pass between question and answer; at night, the night passes. A doubt about delivery or sizing has a short shelf life: if the answer arrives the next day, it often arrives at an already abandoned cart.

The sale that never starts. Many of those questions are not support, they are the last step before purchase. Treating them as a ticket to close rather than a sales conversation is the difference between answering and selling.

Why a generic chatbot usually makes it worse

The natural reaction is to add a chatbot. And the reason almost everybody tried one and switched it off is always the same: it knows nothing about the store.

An assistant that cannot see orders cannot say where your parcel is. One that cannot see stock does not know whether the size is available. One that does not know the real shipping terms makes things up, and an invented answer about a return costs more than silence. At that point the customer writes to support anyway, but annoyed, and you have added a step instead of removing one.

The problem was never chat technology. It was that the chat sat outside the data.

What an automated answer needs in order to be useful

Three connections, and they are always the same ones.

Orders, so "where is mine" is answered with the real status rather than a polite formula.

Catalogue and availability, to say whether a size is in stock and to offer the alternative when it is not, which is the moment an automated answer stops being support and becomes selling.

The store's real terms, written once: lead times, shipping days, islands, thresholds, returns, warranty. Not a generic tone of voice, that store's rules.

And a fourth thing people forget: whoever answers has to know when they do not know. An assistant that hands the conversation to a person at the edges is trustworthy; one that always tries to manage is not.

What changed for two stores

At an artisan cheese producer nearly every question was about shipping a fresh product: lead times, dispatch days, islands, how to keep it. Training the assistant on the real policies and connecting it to orders and catalogue cut repeat questions to support by 60%, around 70% of questions are now handled without a person, and chat conversion rose 18%.

At a womenswear brand the questions were about sizing and how to combine pieces. There the assistant did not just answer: it builds the look and recommends the size. Chat conversion +22%, items per order +15%, returns -18%, and repeat questions about size and fit down 55%.

Two very different sectors, the same shape of problem. Both cases, with all the numbers.

Where to start

Not with the chatbot. Start by reading the last two hundred conversations and counting how many distinct questions there actually are. Usually fewer than fifteen, and the first five cover most of the volume.

Those five are two things at once: the work worth automating, and the list of what is missing from your product pages. Fixing the pages reduces the questions at source; automating the answers handles the ones that remain. Doing only the second half works only halfway.

Frequently asked questions

How many questions can an assistant really handle without a person?

It depends on how repetitive that store's questions are and how complete the information available to it is. In the two cases above it is around 70% at the cheese producer, where questions are few and recurring. On a catalogue with many variants and many exceptions the share is lower, and that is normal.

What if it gets an answer wrong?

That is the right question to ask. The protection is not hoping it will not: it is that it only answers on what it can verify in the store's data, and hands the conversation to a person when it goes beyond that. An assistant that admits it does not know does less damage than one that improvises.

Do I need it on WhatsApp, or is the site widget enough?

It depends on where your customers already write, not where you would prefer. In Italy a substantial share of commercial conversations happens on WhatsApp, and answering there with the same information you use on the site avoids the worst outcome: two channels giving different answers.

Do the conversations stay mine?

They should. They are customer data and must be exportable, because they hold the most useful thing you own: the exact words people use to ask for what you sell. Useful to support, and just as useful to whoever writes the product pages.

Does it really cut returns, or is that a side effect told nicely?

In the fashion case the effect is direct and explainable: most returns start with the wrong size, and advice given before the purchase removes the problem at source instead of managing it afterwards. In categories where returns come from something else, do not expect the same result.