I know it sounds harsh, and I don't mean your customers are wrong for asking. I mean something in the shop wasn’t clear enough, so they asked.

The size isn't on the product page. The shipping time is buried. The confirmation email has no tracking link. The return policy is technically on the website, just not anywhere a person would look while they're holding the thing they want to send back.

So they write to you. And you answer. And next week someone else asks the same thing.

Most webshops treat this as support volume, so you staff for it, or put an AI agent in front of it, and measure it as a resolution rate.

It’s the wrong place to start.

A repeated question is a bug report about your business. The inbox is just where it shows up.

If twenty people ask "do you ship to Norway, and how long does it take?", you don't have a support problem. You have a shipping page problem. Or a product page that never mentions delivery. Or a checkout that springs the cost on them at the last step.

If they keep asking whether a jacket runs small, the size guide is missing, or it's “right there!”, but nobody can find it.

And if they write two days after ordering to ask where the package is, maybe the confirmation email never included tracking. Or they can't find the order on your site without creating an account they didn't want.

None of this is mysterious. You already have the data. Sort your conversations by what people ask. The top ten are your list.

Then go fix the source.

Two things happen when you do that.

First, fewer people write in. That's the obvious one, and it's why support teams like this exercise.

The second one is bigger, and you won't see it in the helpdesk. It's the people who never wrote in at all. They had the same question, didn't want to wait for an answer, and left. Or they added the jacket to the cart, got to checkout, couldn't tell if it would fit, and dropped off.

Well, you didn't just lose a ticket. You lost an order.

That's why this isn't only a customer service project. It's a conversion project that happens to show up in your inbox first.

Still, you can't design every question away. And you shouldn't try to.

"Where is my order?" is not a fault in the shop, as long as you already told them how shipping works. It's a personal question about a specific parcel. Same goes for "Can I change the address on this one?" or "I think I ordered the wrong size." Those are real, legitimate reasons to get in touch.

That's where live chat, self-help and AI agents earn their keep.

Self-help has a bad reputation, and honestly, shops (and useless AI chatbots over the past couple of years) have earned it. A FAQ hidden three clicks down. A widget that interrogates you so you won't bother a person. "Have you read our help center?" as a way to stall.

That's not self-help. It’s just freakishly annoying!

Good self-help is the opposite. The answer is right there, at the moment they would have asked. And a person is still one tap away if the article isn't enough.

Looking up your own order in the chat, without waiting in a queue. Reading how returns work before you write a message. Getting a clear answer from an AI agent that actually looked up your order after checking who you are.

Those are exactly the features we’ve built into the Herodesk live chat that you can put on your website. We’ve turned it from a simple live chat into a mini self-help center.

The customer who finds that answer themselves didn't get worse service. They got faster service! They didn't have to explain the situation twice. They didn't have to wait until Monday morning.

Remember that, because a lot of the industry talks about self-help and AI as a way to spend less on support. That's a side effect. The point is to make it easy for the customer (good customer service is all about making it easy for the customer, remember?)

Easy to find the answer. Easy to get a person when they need one. Easy not to ask in the first place, because the shop was clear.

And don’t just take my word for it. Emil Nissen, CEO of blandselvfrø.dk, talked about it on stage with me at E-handelskonferencen a month ago. This exact recipe is how he managed to scale his business to 10 markets and +300k (that's 300,000!) orders per year with a single person running all of customer support (using Herodesk, of course 😎). They have the lowest ticket-to-order ratio I’ve ever seen. They’ve got their processes straight, focus on self-service that actually helps customers, and have a super-fast way to handle what comes through.

But still, some questions come through, and they’re useful in two ways.

First, they tell you what to go fix in the shop. If the same thing keeps showing up after you've published an article/FAQ about it, the article isn't in the right place, or the page is still unclear. The agent answering it automatically doesn't make that less true.

Second, once your team has answered a question well, that answer can help the next customer.

Our AI agents can draw on your help center and on past conversations (anonymized first, and only if you turn that source on, of course). So the fifth person with the same question doesn't have to wait for someone to type the same reply again.

At this point, you might think… “Why the hell is he promoting self-service when that’s counter to how he makes money?”

I’m doing it because it’s a brilliant way to give great customer service to your customers. And that’s what we’ve ultimately built Herodesk around. Good (no, fuck it – great!) customer service!

And we’ve built tons of features to support it. Our live chat with order lookup and easy access to FAQs. Help center where you can manage and publish FAQs and help articles. Reports showing you what your customers are actually asking about. AI agents that can automatically resolve sometimes >80% of all customer questions (I’ll rant about why you cannot just claim “we can always solve XX % with our ai agents” in another post, ‘cause that’s just a stupid claim) and, of course, our Helpdesk making it super easy for you and your team to collaborate about managing the questions that still (legitimately) land in your inbox.

That's the loop I care about. Your business gets clearer. The questions that remain get answered immediately. And a person is there for the ones that shouldn't be guessed at.

Speed matters, but the cleanest support conversation is the one that never had to happen. And the next-best is the one where the customer got the answer immediately, without waiting for someone to type it again.

A few of these make the questions that do come in easier to answer, without leaving the conversation. Here's what shipped over the past two weeks:

Smaller improvements: Herodesk is now available in Swedish, Norwegian, Dutch and Finnish. AI agents are better at telling Danish from Norwegian, handover rules only hand over when their condition is met (including an optional silent handover that leaves an internal note without messaging the customer on email, Messenger, Instagram and WhatsApp), and a question about payment status is no longer treated as a sensitive money topic. You can import old Outlook emails from a .pst file as closed conversations. In live chat, you can choose which team members' pictures appear in the header and set how far the widget sits from the edge of the page. The signature manager lets you choose who gets the test email, attachments now show a file-type icon, you can upload images in the help center, and in Smartpack you can pause an order that's ready for packing.

We also fixed a few things worth mentioning: large attachments on Microsoft mailboxes now come through, images and other media sent in Instagram or Messenger DMs are stored and shown (for new messages; older expired images can't be recovered), and rules that should run when an AI agent hands a conversation over now fire. Plus Facebook and Instagram reconnect, a translation check that was blocking correct messages, DanDomain Webshop order lookup, and errors when testing rules.

Below are some of my favourite newsletters. So if you like this one, you should consider subscribing to these, too.