Empire Essence: my store is my lab
Outcome: Everything I offer, I test first in a business that actually sells.
evidence: The store is live and selling empireessence.co ↗
For the owner
Business outcome
- Grow sales without paying for ads, with a team of two.
- In production. Best month so far: about COP 700,000 in sales from SEO alone, no ads. The Christmas pages went live on October 17, 2026. The WhatsApp agent is still in testing and doesn't serve customers yet.
- I run what I build. I know what it takes to keep a system alive: prices change, orders go missing, the source of truth drifts.
The short version
Everything I offer, I test first in a business that actually sells. Empire Essence is the online perfume store my wife and I have run since 2023. It’s in production, and it’s my lab.
Context
Empire Essence sells perfumes inspired by well-known brands, through the website and WhatsApp, shipping across Colombia. We’re a team of two. My wife and I make pricing decisions together. I build everything else, with Claude Code and Codex.
Why it matters to you: when I show you a system, it isn’t an idea. It’s something I keep running, with real customers, changing prices and orders that can’t get lost.
The problem
- Grow sales without paying for ads. A small business budget can’t sustain constant ad spend.
- Help gift shoppers find the right bottle among hundreds of references.
- Track essence inventory in grams, not in a notebook.
- Never lose an order. A paid order nobody sees is a lost sale and an annoyed customer.
- Answer fast on WhatsApp, at any hour, without someone glued to the phone.
The approach
Same method I use with clients: one problem at a time, measured, with a person who approves before anything goes live. Every change to the store runs through automated tests and needs my yes. Each release has its rollback written down before it ships.
What I built
- The storefront. Next.js on Vercel, designed mobile-first.
- Search-intent SEO. Gift pages for people searching by recipient, style or budget. Seasonal articles for Secret Santa, Christmas and graduations. With structured data and FAQs.
- Custom inventory and customer system. Python with FastAPI and SQLite, so we don’t pay for an external CRM. In October 2026 we loaded the essence count in grams and merged duplicate references against the supplier’s list.
- Order and payment alerts. Order logging, instant Telegram alerts, payment confirmation from the gateway and an automatic daily check. It came out of a real incident where orders slipped through.
- The QA robot that walks through checkout before every change. It has its own case study.
- A WhatsApp sales agent, in testing. It recommends by occasion, gender and budget, and hands off to a person when it can’t close. It never claims to be human.
Result and status
- In production at empireessence.co.
- Sales: our best month so far was about COP 700,000, from SEO alone, no ads.
- Christmas: gift pages and holiday design went live on October 17, 2026. Seasonal articles followed on October 19.
- WhatsApp agent: still in testing, not serving customers yet. I’d rather wait until it’s safe than risk the business number.
- The goal I set: more than COP 3,000,000 in December 2026. That’s a goal, not a result. When it happens, I’ll post it here.
What I learned
- The source of truth drifts. Once an agent built a gift guide with outdated prices because I gave it an old branch as reference. Since then the rule is fixed: prices come only from the published version or the live site. That lesson is now part of how I install systems: every client declares their sources of truth.
- Urgency shows up in orders. The alert system wasn’t a nice-to-have. It came from a scare.
- An agent that sounds human must not lie. Our WhatsApp agent talks like we do, but if someone seriously asks, it says it’s the store’s assistant. Because of WhatsApp Business rules, and because I’ll sell it to other businesses.
What broke and how I found it
- Orders slipped through. A paid order nobody saw. That scare is where the order log, the instant Telegram alert, the payment confirmation from the gateway and the automatic daily check came from. The alert system was not a nice-to-have.
- A gift guide went out with old prices. I pointed an agent at an old code branch and it wrote a guide with outdated prices. Now prices and catalog come only from the published version or the live site, never from a branch.
- Any change could break checkout. I had no way to know until a customer could not pay. That is why the QA robot exists: 20 checkout cases on desktop and mobile before every change.
Want something like this for your business?
If you run an online store or sell on WhatsApp, I can show you what works in mine and what could carry over to yours.
For the team that hires
Technical notes
- Architecture
- Next.js storefront on Vercel; a custom inventory and customer system (Python, SQLite) on our own server; n8n, Telegram and webhooks for alerts.
- Data
- Orders, payments (Nequi, Bancolombia and the gateway) and inventory in our own system; daily backup and health check; prices and catalog only from the published version or the live site.
- Tests
- QA robot: 20 checkout cases on desktop and mobile plus 60 unit tests before every change; a weekly reviewer checks the live store for design, content, SEO and the purchase flow up to payment.
- Where AI runs and its guards
- A WhatsApp sales agent, in testing and not serving customers yet; SEO and gift pages drafted with agents and reviewed by me before publishing.
- Monthly running cost
- Vercel hosting plus our own server; zero ad spend; the Christmas pages went live on October 17, 2026.
- Repo and status
- Live at empireessence.co since 2023. In production. The agent stays in testing until it passes its cases.
Stack