aiCRO

Point it at a website. Get a full consulting engagement.

aiCRO is an open-source AI growth-consulting engine. Give it a URL and it produces the deliverable a consulting firm hands over — website analysis, positioning, ICP & sales strategy, and a per-segment go-to-market kit — delivered as self-contained HTML that opens from file:// with zero server, zero CDN, zero dependencies.

See a real engagement → · The landing page · How it works · Run an engagement

---

What it produces

Five self-contained artifacts per engagement, under companies/<slug>/:

#	Artifact	What it is
0	The Cover — open-me.html	One page: the one-number diagnosis, a 20-minute action path, the deliverable map. The only file you have to tell the client about.
1	The Book — book/index.html	A single-file consulting book — cover, numbered chapters, captioned exhibits, appendix, glossary. Print-quality; it survives a conference table.
2	The Plan — plan/index.html	The execution plan: access through demand-gen, every task with an owner, a minute count, a day, and a click-path. Persistent checkboxes.
3	The Website — website/preview.html	The rebuilt site rendered inside a simulated browser — URL bar, tabs, device toggle — with a scroll-synced guidance sidebar explaining what each section does and why, citing the book. The raw site also ships standalone, ready to publish.
4	GTM Kits — gtm/<group>.html	One presentation per customer group: messaging matrix, paste-ready web copy, marketing angles, LinkedIn posts, blog concepts, email sequences, a channel plan, a 90-day calendar.

See the whole thing for a real (fictional) company: open demo/open-me.html — a complete engagement for Marigold Home Cleaning, produced with this pipeline.

Why it's different

This is the second generation of an earlier pipeline, rebuilt around three lessons from a 10-firm consulting audit:

• Adoption beats analysis. Every deliverable is built backwards from one question: will a busy owner with 20 spare minutes act on this page? Insight that sits in a drawer is a failure, not an output.
• Show, don't prescribe. Not "rewrite your hero" — here is your new hero, inside your rebuilt site, with the reasoning beside it. The guided simulated-browser preview is the flagship.
• Evidence or it doesn't ship. Every number is tagged [measured], [benchmark: source], or [estimate], or it's cut. Every named company needs a citable URL. A standing arithmetic auditor re-multiplies every deal-math strip. The page that sells aiCRO obeys the same rules — it's zero-dependency and grades its own claims.

How it works

A four-phase pipeline that runs inside Claude Code (the Workflow tool), with one human gate:

intake ──▶ strategy ──▶ [confirm customer groups] ──▶ build ──▶ QA ──▶ ship
  │            │                                         │         │
fact base   artifacts (markdown)                    analysis/*  verdict

Agents read from disk, write to disk, and pass small manifests — so the engagement is reproducible, the cost stays bounded relative to the output volume, and deterministic assemblers (build-book.mjs, build-preview.mjs, check-links.mjs) guarantee a consistent, numbered, cross-linked, print-quality whole. Full detail: process/README.md and process/PIPELINE.md.

Quickstart

companies/<slug>/   ← where engagements are written (gitignored — your client data never leaves your machine)
process/            ← the framework: specs, design system, templates, and the Workflow scripts that ARE the pipeline
demo/               ← a complete sample engagement (fictional) you can open and inspect
index.html          ← this project's own landing page

Run an engagement by following process/workflows/RUN.md: point phase 00-intake at a URL, answer the operator questions, confirm the customer-group set at the one gate, then build and QA. The output lands in companies/<slug>/; zip it (excluding _work/) and email it — it opens offline.

Your client data never enters this repo

The pipeline writes real engagements to companies/<slug>/, and all of companies/ is gitignored. Nothing a client gives you — their URL, their customers, their numbers — is ever committed. The only sample in the repo is the fictional demo/. See .gitignore and companies/README.md.

What aiCRO refuses to do

These refusals are load-bearing, carried over from the consulting-hardening audit:

• No SWOT, PESTLE, 7S, balanced scorecards, or Blue Ocean theater. The framework rubric (process/PIPELINE.md §6) lists what's in, what's conditional, and what's banned.
• No upside multiples ("this program is worth $200K"). Breakeven framing only.
• No unverifiable named attributions. No numbers without source tags. No named entity without a citable URL.
• No operator jargon in client-facing files (grep-enforced in QA).
• Done-with-you, not done-for-you. aiCRO ends at paste-ready copy and publishable HTML. It is not an agency and refuses to become one.

The landing page & demo are part of the repo

index.html (this project's own storefront) and demo/ are hand-tuned static HTML — no build step, no dependencies, system fonts only. They are served as GitHub Pages from the repo root and double as a worked example of the quality bar aiCRO holds. Regenerate the landing after editing its sources with node process/templates/build-landing.mjs ..

License

MIT © 2026 Chris Kluis. Built with aiCRO.