Before we sold a single automation to anyone, we ran a Shopify store.
Not as a hobby. Not as a portfolio piece. As an actual direct-to-consumer brand selling physical products, taking real orders, handling real returns, responding to real customer queries, posting on social, writing blog content, chasing carts, managing inventory. We did all of it ourselves. Then we automated all of it ourselves.
The store is called Lumovex. It still runs. The point of this post is what we learnt from it that an AI consultant will never learn, and why we think dogfooding is the only honest entry point to selling automation services.
You cannot automate what you have not done
Six months in, we had a client ask us to "automate customer support" for a Shopify brand much like ours.
The instinct, if you are a consultant, is to start with a workshop. Discovery. Map the customer journey. Define the support categories. Workshop the response taxonomy. Build a chatbot. Ship it. Move on.
The instinct, if you have actually been the person at 11pm answering "where is my order" emails, is different. You know that the request is not "automate support." The request is "stop me from ever again typing the same shipping-confirmation message twice." Those are different problems.
Most "support automations" fail because they are designed by people who have never been on the receiving end of the problem. They build a chatbot that parrots the FAQ page back to customers. The customers were already not reading the FAQ page. The chatbot does not help.
The thing that helps is recognising that 70% of inbound customer messages are about three things: order status, returns logistics, and product fit. Build a system that resolves those three perfectly, and escalates everything else to a human cleanly. That is not a chatbot. That is a triage layer with hard guardrails.
We only know that because we sat through the support inbox first.
What our own store taught us about scope
The first automation we built for ourselves was for the wrong thing.
We built a competitor monitoring system. It scraped competitor pricing weekly and sent us a report. It worked. We were proud of it. We used it twice and then stopped reading the reports.
Why? Because competitor pricing was not actually our biggest pain. The pain was that we were posting on social maybe twice a week when we needed to be posting daily, and the brand was invisible during the gaps.
The second automation was the social content pipeline. Different impact entirely. Daily posts across three platforms. Content that actually performed. Three months later we noticed we had been writing zero blog posts and getting zero organic search traffic. Built the SEO content engine. Three months after that, twenty-eight posts published, ranking for thirty-plus keywords, traffic compounding.
The lesson is not "we are clever". The lesson is "you cannot prioritise correctly until you have skin in the game". A consultant doing discovery would have asked us "what hurts most" and we would have answered "competitor pricing intelligence" because it was the most technically interesting problem. We were wrong. We only learnt we were wrong because we shipped the wrong thing first and watched ourselves not use it.
This is now one of the most valuable parts of any new client engagement: we ask "what would you build first?" and then we ask "what have you already tried, and what did you stop using after a week?" The second answer tells us more than the first.
A consultant doing discovery would have asked "what hurts most." You will answer with the most technically interesting problem. You will be wrong.
What it taught us about pricing
When we started building automations for clients, we had no idea what to charge.
The honest first instinct was to bill hours. We did some quick maths. A typical service takes us 40-60 hours of dedicated work to ship. At £100 an hour that is £4,000-£6,000 a service. So we started with that.
Three problems showed up immediately.
Hourly billing rewards slow. If we got faster (and we got significantly faster the more we built), we would earn less. That is the wrong incentive structure for a business that wants to stay sharp.
Hourly billing makes scope arguments inevitable. "Why did this take 60 hours and not 40?" "Because the integration was harder than we thought." "Can you do it in 40 next time?" "Maybe, maybe not." Nobody enjoys this conversation. Including us.
Hourly pricing penalises customers who chose us for being fast. A vendor that ships in two weeks for £6,000 looks "more expensive" than one that ships in eight weeks for £4,000 even though the first one is wildly better value.
The fix was fixed-price-per-outcome. Sales follow-ups: from £900. Customer support 24/7: from £1,200. Same scope, same outcome, same price, regardless of how long it takes us. If we get faster, we earn more per hour. If we get slower, we eat the difference. This puts us on the same side of the table as the client every time.
We could not have figured this out from a spreadsheet. We had to feel the friction of hourly billing on three or four real engagements before we knew what to fix.
Why this matters for you
If you are about to engage anyone for an AI build, ask them one question that almost no agency will pass.
"What did you ship for yourself before you started shipping for clients?"
If the answer is "we built a SaaS product that pivoted six times" or "we ran a consulting practice for seven years" or "we have a strong AI methodology", you are talking to a strategy firm dressed up in a hoody.
If the answer is "we ran a real business in your industry, hit the same problem you have, and built the same system we are about to sell you", you are talking to operators.
We are not the only operators. Pick one. Just make sure you do.