← All case studies
Web agency · Contractor sitesAnonymized · under NDAIndustry · Epoxy & concrete coatings

A founder-capped website line became one Discord command.

A US agency builds multi-page websites for epoxy & concrete-coatings contractors. Their ceiling wasn’t demand — it was how fast two founders could personally push each 25–70 page site through. We mapped their entire process and rebuilt it as a semi-autonomous cascade they trigger from Discord and run on their own server.

Weeks → days

signed to first drafts

25–70 pages

per client site, not one landing page

/new-client

one Discord command starts it all

Owned VPS

they keep the system, data & control

The challenge

Demand was never the problem. Delivery was.

Every contractor site is a real build — 25 to 70 pages of services, locations, education and trust content, not a single landing page. And the quality bar was personal: only the founders could tell whether a page truly read like their work.

Generic AI made it worse, not better. It happily produces pages that score well and feel nothing like their brand. So growth stayed capped by founder hours — the more clients they won, the deeper the production backlog got.

What we mapped

We mirrored their operation — then handed it back as a system.

Their process, not a template

We mirrored the entire internal flow — intake, client research, page planning, drafting, SEO, QA, review — into one documented operating model.

Their quality bar, captured

We pulled their definition of “good” from real past work and counter-examples, so the system pattern-matches their standard instead of generic AI defaults.

One trigger, twelve stages

A single `/new-client` Discord modal kicks off a 12-stage cascade — from intake to a designer-ready brief to a delivered, SEO-scored draft set.

The system · 12-stage cascade

From a Discord command to a delivered, on-brand draft set.

Twelve stages, fifteen artifacts. They press /new-client in the channel they already live in — and every stage posts back to their Discord thread as a live audit trail, pausing for human review where it matters.

01

Trigger

02

Source ingest

03

Info dump

04

Content spine

05

Keyword research

06

Sitemap

07

Website brief

08

Designer hand-off

09

Design extension

10

Draft

11

SEO + QA

12

Deliver

How the build works

Their design, extended by the system — never drifting off-brand.

Full-AI site builders all hit the same wall: design. So we drew the line where the human matters — their designer hand-builds the core pages, and the system extends the rest ~80% hands-off, inheriting the exact look instead of guessing at it.

01

Onboarding cascade

Intake form → web-presence scrape → keyword research → a 17-field info dump → content spine → an AI-derived sitemap (25–70 pages) → a designer-ready website brief.

02

Hybrid design model

Their designer hand-builds 8 core pages. The system extends every remaining page from those cores — same tokens, same section patterns, page-unique copy — so it never drifts off-brand.

03

Production pipeline

Per-page brief → section-decomposed drafting (each section written to its own budget, not padded) → SEO / GEO scoring → humanizer + QA → delivered to Google Docs & Sheets.

04

Owned & logged

Runs on a VPS they own and keep. Work stays offline in Google Workspace + Drive; every dispatch posts to Discord. Human review gates the whole way through.

The math

Most of the build team’s work, automated. Designers kept.

Delivery capacity was the ceiling — not leads. The system absorbed the bulk of the build work, so the same team ships several times the sites it used to. Monthly build capacity moved from the high five figures into the mid six figures, the design team stayed, and roughly 60–70% of the dev team’s workload was automated away.

~4×

more sites, same team

High 5 → mid 6 figures

monthly build capacity

7-figure

added annual build capacity

60–70%

of dev-team workload automated

Figures shown as client-confirmed ranges; exact numbers withheld pending client approval.

The proof

Validated on a real brand before it shipped.

We stress-tested the model on a real epoxy brand’s site: rebuild the core pages, then generate brand-new pages from the same system and measure parity. Because generated pages inherit the same section library, they don’t re-attempt brand fit — they’re born with it.

8 / 10

core-page reproduction

9 / 10

generated-page brand parity

~14 sections

each drafted to budget, not padded

Days

signed → first on-brand drafts

Client anonymized under NDA. Figures describe the production system and a pre-launch validation build; per-client commercial terms are confidential. See our fully-named flagship case study →

Capacity-capped, not demand-capped?

If your delivery line is the ceiling — and quality is too personal to hand to generic AI — we’ll map your process and rebuild it as a system you own and trigger yourself.