- type
- summary
- created
- Tue Apr 07 2026 02:00:00 GMT+0200 (Central European Summer Time)
- updated
- Tue Apr 07 2026 02:00:00 GMT+0200 (Central European Summer Time)
- sources
- raw/articles/SERVICE-AS-SOFTWARE-PLAN
- tags
- strategy saas pivot revenue ai-broker
SaaS Plan Summary
Overview
The SERVICE-AS-SOFTWARE-PLAN document lays out a fundamental business model shift. Instead of building a marketplace where mills list surplus and buyers browse, B2BPaper becomes an AI paper broker that replaces human brokers (who charge 3-8%) with automated agents charging 2-4%.
Revenue Model
- Commission: 2-4% per deal
- Target volume: 50-100 deals/month
- Average deal size: EUR 100k
- Revenue target: EUR 150k-400k/month at scale
- Ramp: EUR 50k/month by month 3, EUR 150k by month 6
How It Works
Buyer Flow
- Buyer submits a natural-language request (e.g., "40t uncoated woodfree, 80gsm, reels, Central Europe, 3 weeks")
- AI agent searches supply database and live scrapes marketplaces
- Agent presents 2-3 matching offers with pricing
- Buyer picks one; agent handles negotiation and paperwork
- Deal closes; B2BPaper takes commission
Mill Flow
- Mill submits surplus info (e.g., "60t off-spec coated going obsolete in 30 days")
- AI agent scans demand profiles for matching buyers
- Agent drafts personalized offers to qualified buyers
- First buyer to commit gets the deal
- Commission taken from mill side
Thierry's Role
Thierry remains the human relationship anchor for high-value deals (EUR 50k+), opens mill networks via personal contacts, validates industry knowledge (pricing, grades), and closes the ~20% of deals that require human trust. The AI handles the other 80%.
Technical Architecture
The plan builds on top of the existing marketplace codebase (Django REST + PostgreSQL). The architecture has five layers:
- Supply Side -- scrapers, mill API, manual entry
- Demand Side -- intake forms, outbound, CRM
- Matching Engine -- grade, GSM, format, volume, geography, price optimization
- Deal Pipeline -- Lead, Qualified, Offered, Negotiating, Closed
- Communication Layer -- email, WhatsApp, agent outreach
Existing assets at time of writing: Product models, mill data, Extractor pipeline (8,919 docs, 4,246 products, 440 mills).
Phase Breakdown (42 Tickets, ~44 Days)
| Phase | Focus | Duration |
|---|---|---|
| Phase 0 | Foundation (supply, demand, deal, match models + admin shell) | 5 days |
| Phase 1 | Supply scraping (Crawlee framework, 4 scraper targets, dedup, scheduler) | 8 days |
| Phase 2 | Demand collection (intake form, buyer profiles, outbound email, qualification) | 5 days |
| Phase 3 | Matching engine (algorithm, scoring, auto-triggers, review dashboard) | 5 days |
| Phase 4 | Communication layer (email templates, WhatsApp API, conversation tracking) | 5 days |
| Phase 5 | Deal management (kanban pipeline, contract PDFs, commission/invoicing, analytics) | 5 days |
| Phase 6 | Service landing pages (buyer-facing, mill-facing, SEO) | 3 days |
| Phase 7 | AI agent layer (qualification, matching, outreach, negotiation, daily digest) | 5 days |
| Phase 8 | Testing and hardening (E2E flow, load testing, error handling) | 3 days |
Go-to-Market Strategy
- Thierry calls 10 known mills for surplus lists (manual supply entry)
- Outbound to 500 converters/printers to build demand profiles
- First 10 deals manually assisted by Thierry + AI
- After 20 deals, let agents run 80% autonomously
- Scale scraping to cover all major paper marketplaces
Key Observations
- This plan was written as a future vision, not as the MVP that was actually built. The MVP (phases 0-13) focused on the marketplace foundation.
- The scraper component was partially built in a separate repo (
/home/claude/customers/b2bpaper-scraper/) reaching 24/56 stories before being paused at SaaS-060. - The AI agent layer (Phase 7) represents the most ambitious part -- five specialized agents for different stages of the deal lifecycle.
Sources
- raw/articles/SERVICE-AS-SOFTWARE-PLAN -- full strategic plan with ticket-level detail
Related
- wiki/concepts/service-as-software-pivot -- concept page on the pivot model
- wiki/concepts/spec-based-matching -- matching engine that underpins the AI broker
- wiki/concepts/make-to-order-marketplace -- evolution beyond surplus into planned production
- wiki/entities/b2bpaper -- the platform entity
- wiki/summaries/progress-summary -- what was actually built vs. this plan