Show Up Online: An Agentic Website Business

Client

Show Up Online sells websites to small local service businesses that have a weak web presence or none at all. It researches the business, builds the site, sells it, and then keeps it running. The site exists before the owner has ever heard of the service.

The offer is a one time setup fee and a small monthly subscription, priced so a contractor with a van and a phone does not have to think about it. At that price the traditional agency model collapses. A salesperson, a designer, and a project manager cannot be paid out of a contract that size, which is why this end of the market has been left alone for years.

A-CX designed the engine behind the service, built it, and operates it. The venture that owns the brand stays out of the day to day. A-CX runs the system on its behalf and decides which market each campaign goes after.

Challenge

Every step of the traditional path is a person’s time. Finding the business, qualifying it, building something worth showing, chasing the reply, closing, launching, supporting. Eight touches, and not one of them gets cheaper as volume grows.

The trap is what actually works as outreach. A contractor does not respond to a pitch. They respond to seeing their own business, their own reviews, and their own phone number on a page that looks better than what they have now. So the most expensive thing in the process has to be produced before anyone has agreed to buy anything.

Three constraints shaped everything that followed. Nothing on a generated page could be invented: no years in business we could not verify, no credentials, no text or photographs taken from the prospect’s existing site. Cold outreach carries a legal floor that applies from the first message rather than the first complaint. And a system that reads text written by strangers on the public internet and then generates from it has a prompt injection problem that is operational, not theoretical.

One more constraint decided the architecture. A single person, working part time, had to be able to oversee the entire book of customers.

Solution

Eight steps, one automated pipeline, one person watching it. What used to be eight separate jobs is now a single run, and the work moves from stage to stage without anyone carrying it.

How it enters a new market. Campaign selection. A-CX picks the trade and the geography, the system runs the campaign, and the numbers that come back sharpen the next one: which prospects convert, which messages land, and where the operating time actually goes.

Results

Eight Jobs, One Run, One Part Time Operator.

All eight steps run as a single automated pipeline. From the moment a prospect pays, a live site is delivered inside 48 hours, including domain setup, structured data, analytics, and the performance work.

One person oversees the whole book, part time. The work per additional customer is close to flat, which is the only reason the price point works at all.

Entering a new market is a campaign decision rather than a build, and every campaign returns numbers that improve the next one: sharper targeting, better messages, and a clearer picture of where the operating time is spent.

Every production site has to clear a performance threshold before it can go live, because the whole pitch rests on the prospect’s current site being slow.

The design assumes the human role keeps shrinking rather than disappearing. The gate that exists today is the one that still earns its place.

Technical details

Every step runs on A-CX AI Nexus, which carries the orchestration, the event flow between stages, agent execution, and the escalation paths that reach the operator. Anything that touches data or model output goes through it. Nothing is wired service to service.

Sites are assembled from a block registry rather than generated as free form markup, with theming driven by CSS variables set from the per business brand brief. The same blocks serve the private site and the production site, so what the prospect approved is what goes live. Output is static and served from the edge, one project per customer, with contact form handling in an edge function that carries an encrypted recipient token, so no recipient address appears in page source and no submission is stored.

The generated sites are plain HTML, CSS, and JavaScript. Python and Java sit behind the pipeline, running discovery, scoring, generation, classification, and the deployment path. Payments run on hosted payment links, so the recurring relationship starts without asking the customer a second time.

The guardrails are part of the build rather than a policy document. Input taken from the public internet is sanitized before it reaches a prompt, generated markup is sanitized before deploy, and the generators are built to omit an unknown field rather than fill it. The whole system is source controlled with architecture decisions recorded as they are made, so the reasoning behind each choice outlives the people who made it.

Does your economics break the moment the deal gets small?

When the work in a funnel is priced by the hour, and the deal is not, the answer is usually to move the work rather than the price. We can walk through which stages of your process can run on agentic AI, and where a person still belongs. Talk to us.