Ecommerce / 18 Jun 2026

Where to build your first factory

I wrote last time that the job has changed, that we build factories now instead of features. The fair reply to that is simple. Fine, where do I start, in my actual business, this week? For most ecommerce stores between two and ten million in revenue, the answer is the same, and it is sitting in plain sight. Build your first factory on the support queue.

The repetitive-question tax

Here is the problem nobody puts on a slide. A growing store pays a tax, and the tax is answering the same handful of questions forever.

Run the rough numbers on a four million dollar store. Depending on your category and price point, that is somewhere between twenty and fifty thousand orders a year. A fifth to a third of those orders generate a support contact. Call it a thousand tickets a month. Now look at what those tickets actually are, and the pattern is brutal:

• Where is my order. The industry has an acronym for it, WISMO, which tells you how common it is.
• Returns, exchanges and refunds.
• Sizing, fit and compatibility.
• Can I change or cancel my order.
• Is this back in stock.

In most stores, well over half of all tickets are some version of those five. That matches what I see in the business I run. You are paying full salaries to type the same answer to "where is my order" several thousand times a month. Worse, the genuinely hard tickets, the ones that actually need a human, are stuck in the queue behind the easy ones.

Why the two obvious fixes both fail

Faced with this, most founders reach for one of two things. Both disappoint.

The first is to hire another support person. It works for a quarter. But you have not removed the tax, you have just paid someone to keep paying it. Hire the fifth rep and you are buying the same five answers, over and over, at full price.

The second is to bolt a chatbot onto the website. This fails for a reason I have written about before. The bot answers like a clever stranger, because the business was never made legible to it. It does not know your real returns window, your fit guidance, your live stock. So it guesses, gets things wrong in public, and the team quietly switches it off. The point is not that chatbots are bad. The point is that a chatbot sitting on top of a company that cannot be queried is just a faster way to give wrong answers.

Build the system, not the bot

The thing worth building is not a bot. It is a support factory, and it runs in a strict order. This is where the rest of the series comes together.

1. A brain to read from. Before anything answers a customer, it reads your written-down knowledge: returns policy, fit guides, shipping times, the product truths. If your company is not queryable yet, that is the real first job. No brain, no factory.
2. A gate for anything risky. The system resolves the safe, known questions on its own. Refunds, complaints, anything legal or about safety, it hands to a human. The model stays on the cheap side of the gate, where a mistake costs a click, not trust.
3. A clean handover for the rest. What it cannot answer well, it routes to a person with the context already gathered, so the human starts a hard ticket warm instead of cold.

Most of the volume, the where-is-my-order tier, never needs a person again. The hard third still does, and now your team has the time and the head space to do it properly.

It pays you back twice

The first payback is obvious. The repetitive tier stops eating salaries, and your people move up to the work that actually needs judgement.

The second payback is quieter and bigger. A support factory built this way writes things down as it works. Every time a human answers a question the system could not, that answer can flow back into the brain as a help centre article. The store gets more legible as a side effect of operating, which means next month the system resolves a little more on its own. The tax shrinks over time instead of growing with revenue.

Start by reading your own tickets

You do not need to buy anything to begin. You need to know your own tax first.

Open your helpdesk and read the last few hundred tickets. Sort them into buckets. You will almost certainly find that a small number of question types make up most of the volume, and that most of those are answerable from things you already know but have never written down. That list is your build order. Start with the biggest, most boring bucket, because that is where the money is.

The rule

Do not set out to automate your support. Automate the five questions that are not really support, the ones you answer on autopilot anyway.

Find your most boring, most repeated question. The one you could answer in your sleep. That is not a chore to hand to a junior. That is the first factory to build.