Demo case · AI Pipeline · February 2026
SmileClinic — a full proposal package in 6 minutes
People often ask me: «What exactly can AI do for my business?» Instead of long explanations, I put together an answer you can hold in your hands: one run of the pipeline — and a ready commercial proposal package.
// The scenario — and why it is honestly fictional
SmileClinic is a typical dental chain: 3 clinics, 15 doctors, 3000+ active patients, competition in a crowded city market. I have seen this profile in real clients more than once — so the scenario is realistic, but the client itself is fictional, and that is stated plainly.
The demo answers a concrete question: if you take a typical business and run it through an AI pipeline — what exactly comes out and how long does it take?
// How I thought about the task
A proposal is not just «text». It is a package: a client dossier, a document, slides for the meeting, and a page you can send. By hand, such a package takes several people several days to assemble. So the task for the pipeline is not «to write text», but to go through the whole chain in a single run.
That is why under the hood there is Claude Code with a custom skill: the agent does the OSINT research itself (website, news, the decision-maker on LinkedIn, job postings), assembles the dossier, generates personalized copy, renders the PDF and slides, and builds a landing page with analytics.
All in one run — with no switching between platforms.
// What one run produces — 5 artifacts
Website, news, decision-maker profiles, job postings — collected and structured.
Client pain, process, metrics, pricing, FAQ — personalized to the dossier.
Ready to send, fully laid out — without a designer.
Generated via Gemini from the proposal copy.
A dedicated proposal page + Umami, to see whether it was opened.
// Stack
| Layer | Tools |
|---|---|
| Orchestration | Claude Code (Sonnet 4.6) + a custom skill |
| OSINT | Web research: website, news, decision-maker on LinkedIn, job postings |
| Proposal copy | Claude — 205 lines: pain, process, metrics, pricing, FAQ |
| Document gen | Puppeteer (PDF), Gemini 2.5 (slides) |
| Landing | Static landing page + Umami analytics |
| Hosting | Cloudflare Workers + R2 for storing the PDF |
// What the proposal itself offers — numbers from the scenario
The generated proposal offers a typical clinic a project for 142 000 грн with an estimated payback of ~3 місяці. The logic inside the proposal:
- 2 admins on routine work — that is ≈ 432 000 грн/рік;
- the AI agent handles ~60% of routine inquiries;
- bookings outside working hours — another ≈ +50 000 грн/міс in potential revenue.
Important: this is a projection for the demo scenario, not the reporting of a real client. The point of the demo is to show the depth of work the pipeline delivers in 6 minutes.
// What went wrong — lessons learned
The first versions of the pipeline generated text that was too generic. It took a few days of iterating on the prompts and the skill structure to get a result on par with a live marketer. «6 minutes» is after the refinement, not on the first try.
The structure ate the most time, not the generation. Until the skill got a clear proposal format (pain → process → metrics → pricing → FAQ), the text came out polished but empty.
// The reaction
The post about this demo gathered 12 000+ views and dozens of comments on Facebook — mostly with one question: «can you do this for us?». You can — see the FAQ below.
// FAQ
Is this a real client? +
No, and it is stated plainly: SmileClinic is a demo scenario of a typical dental chain (3 clinics, 15 doctors, 3000+ patients), put together to show the pipeline on realistic data. What is real here is the pipeline itself and everything it generated.
What exactly is real in this case? +
The pipeline and the artifacts. One run of Claude Code with a custom skill produces in ~6 minutes: an OSINT dossier on the company, a 205-line proposal (pain, process, metrics, pricing, FAQ), a PDF document, a 6-slide deck, and a landing page with analytics. Building the pipeline itself took a few days of iteration.
Where does the 142 000 грн figure come from? +
It is the budget the generated proposal itself offers for a typical clinic from the scenario. The calculation inside the proposal: 2 admins ≈ 432 000 грн/рік, AI handles ~60% of routine work, bookings outside working hours ≈ +50 000 грн/міс. Hence an estimated payback of ~3 місяці. This is a projection for the scenario, not the client’s reported figures.
Why Claude Code and not ChatGPT or Make.com? +
ChatGPT is good for one-off prompts, but without agentic logic. Make.com is fine for simple chains (form → email), but it «chokes» on complex branching. Claude Code lets you write custom skills: one run runs OSINT, generates the text, renders the PDF/slides — all deterministically.
Can something like this be built for my business? +
Yes — that is the whole point of the demo. The pipeline is built around your process and your documents. We build the strategy and architecture together with me, implementation is led by the Auspex or Grow2.ai team.
Case author
Andrew Maryasov
AI consultant, founder of Auspex (CRM automation) and Grow2.ai (AI agents + community). 25+ years in business automation: accounting → CRM → AI. I build AI strategy for owners and teams; implementation goes through my own brands.
// Other cases
// Takeaway
Building the pipeline took a few days of refinement. But now every next proposal is assembled in the same 6 minutes: a one-time engineering investment — and from then on the process runs without person-days.
Pipelines like these are built in real projects by Grow2.ai (AI agents) and Auspex (CRM automation). I build the strategy personally — implementation is led by the team of the relevant brand.
Curious where AI gives you the leverage?
Let's start with a conversation about your situation. We build the strategy together, implementation — through Auspex or Grow2.
Personal advisory →