How we automated our own content and sales pipeline with AI

Six building blocks, one system: from LinkedIn posts through website articles and proposal PDFs to the funnel from content to booked call. Built for AlpiType itself, since adapted in client projects — with the numbers it produced.

For a long time marketing, sales and content production ran as separate processes at AlpiType. Every step needed a manual handover — from the idea to the appointment in the calendar. Content created visibility but no measurable leads; scaling depended on people. This report describes what we built from that, block by block. Blocks 01 to 04 can be used on their own; 05 and 06 connect them into a system.

01 LinkedIn content from existing material

Starting point

Existing technical content sat scattered across PDFs, notes, slides and old posts. New LinkedIn posts were written ad hoc — if at all. Writing effort per post: 30 to 90 minutes. Tone varied between authors, content from old materials was not reused, posts appeared irregularly.

Solution

A pipeline that reads existing materials (PDF, Markdown, audio) and generates LinkedIn posts from them. Format, tone and structure are defined as reusable templates; each source yields 3 to 5 post variants. Every output passes a review step — one click to approve or adjust — before publication.

Result

Writing effort per post: under 5 minutes
Posts per week: from 1 to 5
Tone stays consistent across all posts
Existing content is reused systematically

02 Website and SEO articles from talks and transcripts

Starting point

Raw material existed — talks, interviews, technical notes, transcripts — but rarely became website articles. Transcripts ended up in a folder. SEO structure (H1, H2, meta description, internal links) was missing; crawlers did not find the content.

Solution

Whisper transcribes audio and video locally. Claude structures the text into H2 sections and generates the meta description and internal links. Publication happens as a draft in the WordPress back-end via the REST API — a person decides before anything goes live.

Result

Effort per article: from 4 hours to 30 minutes of review
SEO structure is correct from the start; content gets indexed and ranks
One talk becomes several articles instead of one note

03 Sales enablement: structure requests, prepare proposals

Starting point

Sales requests required technical validation, research in old projects and manually assembled proposal PDFs. Engineering was pulled into every sales cycle; response times depended on the availability of individual experts, and knowledge from past projects was not reused systematically.

Solution

A sales-operations layer accepts requests in any format, structures them, compares them with past projects, checks feasibility and generates a first proposal draft — PDF proposal, technical summary, follow-up questions. Engineering only steps in for complex cases.

Result

First response within hours instead of days
Engineering involvement per request drops significantly
Consistent proposals regardless of who handles them; knowledge from past projects is retrievable

04 Multi-channel distribution from one source

Starting point

Content was created per channel — an article for the website, a separate post for LinkedIn, another format for the newsletter. The same content was prepared several times by hand, website and LinkedIn said different things, and not every channel was served.

Solution

A central content hub. A single input — a talk, an article, a technical document — generates the website article, the LinkedIn post, the video snippet, the newsletter entry and the media page in parallel. Each channel gets its own format but the same source; publication is coordinated via REST APIs (WordPress, LinkedIn, mail).

Result

One input becomes 4 to 6 outputs in different formats
Publishing effort drops by a factor of 5
Messages stay consistent across all channels; no channel is forgotten anymore

05 The full system: content → website → download → profile → call

Input: talks, transcripts, technical documents, customer requests
Processing: Whisper, Claude API, format generation per channel
Output: articles, LinkedIn posts, media pages, PDF downloads, sales summaries
Funnel: content → website → download → profile → call

Every stage is measurable and can be optimised on its own. Lead capture, follow-up and profiling — previously spread across different tools — now hang off the same content.

Result

Content publication follows a reproducible process instead of ad-hoc actions
Leads come from the content itself, not from separate campaigns
Engineering involvement in the sales funnel is reduced to qualified conversations
The system scales with the content, not with team size

06 Distribution as a sales channel

Starting point

Content existed — articles, videos, insights from running projects — but creation and distribution were entirely manual. High effort per piece, inconsistent quality, no structured funnel from content to enquiry.

Solution

Claude structures existing content by audience and turns it into LinkedIn posts, website articles, SEO texts and PDF summaries, with the tone adapted per channel. The pipeline behind it: content is created and distributed → readers download a PDF → the lead is captured in the CRM → follow-up content is delivered → the sales conversation is prepared. Every step is automated and traceable.

Result

Production time per piece: from hours to minutes
Consistent tone across all channels, scalable without additional staff
Content becomes a measurable sales channel — full transparency from first click to project start

What transfers to engineering teams

The mechanics are the same as in our engineering use cases: read existing material, structure it, bring it into a target format with templates, put a review step in front. For requirements analysis, code review, test generation and knowledge management we have documented this on AI for Engineering Teams — including the measured numbers. The Engineering AI Workshop (in German) is the entry point for teams.