Automation
Automation that can speak for a brand
A direct-to-consumer supplement brand selling online, and the automation behind it that promotes the products, keeps three social accounts fed and routes customer questions to a person — with the whole thing driven and approved from Telegram, including an agent the founder can instruct to post on demand.
- Our role
- Built the store and the automation
- Built with
- Shopify · n8n · Telegram · Webhooks · Email
- Where it got to
- Live, selling
- Open it
- zorevexa.com
The idea
Put a supplement brand online and stop the founder spending every morning copying the same thing between the shop, the mailing list and three social accounts.
What it actually needed
The shop was the easy half. The work that actually consumed the week was everything around it: telling customers about an offer, keeping three social accounts fed, and answering the same questions in whichever channel they arrived through. None of that is a storefront problem. It is the systems not talking to each other.
What got built
- The storefront
- A direct-to-consumer supplement shop: catalogue, a bundle offer, reviews and the content around them.
- A promotional path that uses the brand's own domain
- Offers reach customers from the brand's address rather than a generic sender, so the mail lands and looks like the company it came from.
- Social posting across three networks
- Short-form video and posts go out to the brand's own accounts, with the captions drafted rather than hand-written each time.
- Support routed to a person
- Customer replies arrive in the brand's own inbox instead of scattering across whichever channel the customer used.
- Control from Telegram
- What is about to go out arrives as a message on the founder's phone and is approved with a tap, rather than waiting in a dashboard somebody has to remember to open.
- An agent to ask, on demand
- A conversation in the same place: describe a post and have it made, instead of waiting for the next scheduled one. It runs inside the same fence — the interface got easier, the limits did not move.
The decisions that shaped it
Limits before features
This automation can email customers and post in public under the brand's name. Software that can speak for a company can embarrass it, and an apology is more expensive than the feature was. So the sending path was fenced to the brand's own domain and a guard added before the promotional flow was allowed to run at all.
The shop is not the product
A storefront is a few days of work and any agency will sell you one. What a small team actually lacks is the hour every morning spent copying between the shop, the mailing list and three social accounts. The automation is where the time comes back, and it is the half worth paying for.
The gate goes where the person already is
A fence is only as good as the gate in it, and a gate in a dashboard becomes a task you have to remember. Either the queue backs up or somebody widens the rule to stop it backing up, and both end with the automation posting unreviewed. Putting the decision in Telegram, already open on the founder's phone, makes approving cheaper than ignoring.
Drafted, not decided
Captions are generated, and a person still sends. The system removes the blank page rather than the judgment, because the cost of an automated post that reads badly falls on the brand and not on the tool.
In production
Live and selling, with the automation running behind it. The social path is the most recently exercised and is still being tuned.
What this is worth
An hour a day comes back to the founder, and the automation cannot email a customer or post in public outside the fence it was given. What that avoids is the version of this that saves time until the morning it says something the brand has to apologise for.
The shop was the easy half, and it is the half people ask for. A catalogue, a bundle offer, reviews, the content around them. Any agency sells you one and it takes a few days.
What was actually consuming the week sat around it. Telling customers about an offer. Keeping three social accounts fed. Answering the same handful of questions in whichever channel they arrived through. A person was employed, in effect, as the connection between four systems that do not talk to each other.
Software that can speak for you can embarrass you
This automation sends mail to customers and posts in public under the brand's name. That is a different category from software that only moves data between two databases. A mistake is not a failed job in a log. It is something a customer received, or something sitting on a public timeline with the company's logo on it.
So the sending path was fenced to the brand's own domain, and a guard put in front of the promotional flow before it was allowed to run at all. Neither was requested. Both were cheaper to decide while the system could not yet reach anybody, which is the only time that decision is cheap.
The approval had to live where the founder already was
A fence is only as good as the gate in it. The first version put the decision in a dashboard, which sounds right and fails in a predictable way: approving something becomes a thing you have to remember to go and do, so either the queue backs up or somebody widens the rule to stop it backing up. Both end with the automation posting unreviewed.
So the control surface moved to Telegram, which is already open on the founder's phone. What is about to go out arrives as a message. Approving it is a tap, in the place they already are, rather than a task competing with the rest of the day.
And then it could be asked, not only answered
The same channel now carries an agent the founder can talk to directly — describe a post and have it made, rather than wait for the next scheduled one. That is the part that changes how the system is used day to day, because the automation stops being a thing that runs at you and becomes a thing you can ask for something.
It did not widen what the automation is allowed to say. An instruction arriving in a chat is still an instruction into the same fenced path, subject to the same rules about what may be claimed and where. The interface got easier; the limits did not move. Those are separable, and keeping them separate is the whole discipline.
Drafted, not decided
Captions are generated and a person still sends. The system removes the blank page rather than the judgment, because the cost of a post that reads badly falls on the brand and never on the tool that wrote it.