Our own site: an ROI calculator priced by task, real HTML
A bilingual marketing site that prices a task two ways, a person by the minute and the model by the token, with every route served as real prerendered HTML.
1Pick the task: support reply, IT request, lead research and more
2Saving in year one, rolled up from the per-task gap
ROI calculator: person vs. model, priced per task and rolled up to a year-one saving.
1Pick the task: support reply, IT request, lead research and more
2Saving in year one, rolled up from the per-task gap
What runs automatically
01 / PRERENDER
Prerendered, then hydrated
The production build renders every route in both locales to static HTML, rewriting title, description, canonical and hreflang per page, then the browser hydrates that HTML instead of rendering from scratch.
02 / ROI
Per-task ROI calculator
One task gets priced two ways, a person by the minute at a fully loaded hourly cost and the model by the token at published prices plus a human-review pass, then rolled up to a year-one saving.
03 / LEADS
Lead capture, four hops
A submitted form is checked by hCaptcha, written to Supabase, sent as a mandatory email over Gmail SMTP and mirrored to Telegram as a best-effort extra; the Ami chat widget on the same site answers questions live.
Built with Claude Code in 6 months
Before
An agency site made the case for AI with a pitch page and a contact form.
After
Visitors price their own task against the model and read twelve articles, in production at anmflow.com.
[→]/More screens
Home: the pitch and a live status panel above the fold.
Learn: one of twelve articles, prerendered like every other route.
Under the hood
01Context
An agency site that has to prove the pitch
An agency that sells AI systems has to read like a company that runs its own advice, not a template with a contact form on top. That means a real cost model instead of a vague promise, content that ranks instead of one About page and a build that behaves the same for a search crawler as for a visitor. The home page opens with the pitch and a live status panel, then routes into the service catalog, the Learn library and the calculator itself.
02How it works
Every route prerendered, then hydrated
The site is a Vite and React 18 single-page app, but visitors and crawlers never see a blank shell. The production build runs the client build, an SSR build of the same component tree and a prerender script that renders every route in both locales to static HTML, rewriting the head tags per page. The browser then hydrates that HTML instead of rendering from scratch, so the real headline is already on screen before any JavaScript runs; the last build wrote 97 per-route HTML files.
03Outcome
Live at anmflow.com, in two languages
In production at anmflow.com in English and Russian, with a URL-based locale for every page. Twelve Learn articles cover AI ROI, implementation cost and adoption topics, each one registered with its own EN and RU content and prerendered like every other route. Analytics loads through GA4 under Consent Mode v2, so tracking stays denied until a visitor accepts the cookie banner; the calculator and the Learn library are the two pieces doing the actual selling.
ANM SOLUTIONS / CONTACT US
Want something like this?
First 30-minute call is free. We scope it, you decide.
