Xolib — Strategy & Competitive Intelligence¶

For Claude Code: This document is the strategic compass for every development decision. Before building any feature, check: Does this deepen the data moat? Does it create switching costs? If a feature is pure CRUD with no AI layer and no data value — question it.

1. Core Thesis¶

Software is the interface. Data is the product. Every property management competitor (Aareon, Haufe, Immoware24, Win-CASA) sells software. Xolib sells a system that gets smarter with every action — and takes institutional knowledge with it when a customer leaves. The moat is not the feature set. The moat is: 1. The events we log (every action, every agent call, every outcome) 2. The memory we build per entity (tenant, unit, property) 3. The feedback we collect (thumbs up/down on every AI recommendation) 4. The outcomes we track (did the recommendation actually work?) A competitor can copy our UI in 3 months. They cannot copy 500,000 real Hausverwaltung decisions with documented outcomes. That takes 3-5 years and cannot be accelerated with capital.

2. Strategic Priorities — Every Feature Must Answer These¶

When building any new feature, Claude Code must ask: 1. Where can AI help here? — No pure CRUD. Every module needs an AI layer. 2. What data does this generate? — New fields = new training data. Collect more, not less. 3. Does this create switching costs? — Memory, history, learned preferences = things customers lose when they leave. 4. Can this feed the Xolib Score? — Building scores score data -> bank API -> B2B revenue. 5. Is this i18n-ready? — Always. No hardcoded text. 8 languages minimum.

3. The Data Moat — Architecture Decisions¶

3.1 Event Sourcing (Sprint Priority: Phase 2)¶

Every user action, agent action, system event -> xolib_events table. Append-only, never deleted. Key fields: tenant_id, event_type, entity_type, entity_id, actor_type, payload JSONB, embedding vector(1536), created_at Event taxonomy: verwaltung.* / ai.* / system.* Target volume: ~140 events/day/tenant -> 255M events/year at scale.

3.2 Episodic Memory (Sprint Priority: Phase 2)¶

entity_memory table: compressed, semantically enriched knowledge per entity (tenant, unit, property). Updated nightly by background job. Three types: episodic (what happened) / semantic (what it is) / procedural (what works). This is the #1 switching cost mechanism. A customer who leaves Xolib loses years of accumulated institutional knowledge about their tenants and properties.

3.3 Feedback Loops (Sprint Priority: Phase 2 — thumbs up/down UI)¶

Every AI recommendation gets explicit feedback. Combined with implicit signals (acceptance rate, time-to-decision, modifications before accepting). Target: 50,000 feedback entries -> first LoRA fine-tune on open-source model. Target: 200,000+ entries -> full RLHF, own domain model in production, AI-COGS drop 75-90%.

3.4 Outcome Tracking (Sprint Priority: Phase 3)¶

agent_outcomes table: was the agent's prediction correct? What was the EUR/hours delta? This enables: provable ROI dashboard per tenant. "Xolib saved you EUR 4,200 this month."

4. Xolib Score — The Key Differentiator¶

Concept: Building evaluation score 0-100, 5 categories: - Substance (25%) — building condition, age, renovation history - Technology (25%) — smart systems, energy class, metering - Revenue (25%) — occupancy rate, rent performance, payment reliability - Compliance (15%) — certificate status, legal obligations met - Maintenance (15%) — ticket resolution rate, preventive maintenance Why this is strategically critical: - All data already exists in Xolib — zero extra effort for the customer - Real-time updates (not point-in-time like an appraiser) - Objective, audit-trailed, tamper-proof - Bank API: standardized building report -> B2B revenue stream - Investor pitch narrative: "We make housing economically measurable" Building decision: Every new module should feed the Score. When building Tax Exemption Certificates -> Compliance score. When building Heating Costs -> Technology + Revenue score. Always ask: which Score category does this affect?

5. Competitive Landscape — Current State (March 2026)¶

Immoware24 (Biggest threat)¶

4,000+ customers, 1.8M managed units, German market leader in cloud
AI features they have: Invoice recognition, call answering (40+ languages -> tickets), chatbot, text generation (since Jan 2025), KI-based text reformulation
Acquired we-do.ai GmbH (Nov 2023) — dedicated AI team, serious investment
What they do NOT have: Agent stack, episodic memory, predictive layer, outcome tracking, data moat architecture, Xolib Score equivalent
Strategic framing: They bolt AI onto legacy architecture. We build AI-native from scratch.
Known weaknesses: UI "like Excel from the last millennium", steep learning curve, non-transparent pricing, no key management, no smart meter integration, German-only
Pricing: ~EUR 0.29-0.39/unit/month + EUR 15 base fee
Faruk training intel: Questionnaire sent, results expected after 13.03.2026

Aareon¶

Acquired by TPG for EUR 3.9B (2024) — validates that this market has exit value
Enterprise focus (large housing companies), not SME Hausverwaltung
Legacy on-premise architecture being cloudified — slow

Haufe (axera/wowinex)¶

Legacy desktop software being modernized
No AI-native positioning
Strong brand but weak product innovation

Win-CASA / Domus¶

Desktop/legacy, no cloud-native, no AI
Customer base: traditional small HV firms resistant to change

Strategic conclusion¶

The market has exits (Aareon EUR 3.9B), has established players with AI features (Immoware24), and has clear pain points (UI complexity, no predictive intelligence, no data monetization). The window for an AI-native challenger is open — but Immoware24 is moving faster than expected.¶

6. Revenue Model & Monetization Roadmap¶

Primary: SaaS subscription¶

Basic: EUR 4.90/unit/month
Pro: EUR 3.90/unit/month (volume)
Enterprise: EUR 2.90/unit/month (volume)
Annual plan: 2 months free
Break-even: ~43 units on Basic covers all current operating costs (~$225/month)

Secondary: Data monetization (Year 3+)¶

All based on pseudonymized aggregates — never individual tenant data. GDPR-compliant. | Phase | Product | Target | Revenue potential | |-------|---------|--------|------------------| | Phase 1 (M12-18) | Market benchmarks | HV associations, law firms | EUR 50K-200K ARR | | Phase 2 (M18-24) | Risk scores for lenders | Regional banks, insurers | EUR 200K-500K ARR | | Phase 3 (M24-36) | Real estate market intelligence | Investors, PE funds | EUR 500K-1M ARR | | Phase 4 (M30-48) | Contractor network ratings | Handwerk, FM companies | EUR 100K-500K ARR | | Phase 5 (M36-60) | Regulatory early warning system | Compliance consultants | EUR 200K-1M ARR |

Tertiary: Xolib Score bank API¶

Standardized building reports for financing decisions. Separate B2B product. Timeline: after Score reaches statistical significance (~500+ scored buildings).¶

7. Exit Strategy — Who Buys Xolib and Why¶

Target exit: Year 5-7, EUR 30-80M range (base case), EUR 100M+ (optimistic with own AI model) Most likely buyers: 1. Aareon / TPG — just paid EUR 3.9B, needs AI-native tech and SME market access 2. Haufe Group — wants cloud-native modernization + AI layer 3. PATRIZIA / Vonovia — vertical integration of property management software 4. European PropTech PE (Accel, Index, Balderton) — platform consolidation play 5. OpenAI / Microsoft — if own domain model proves vertical AI thesis Valuation drivers (in order of importance): 1. ARR x revenue multiple (PropTech AI-native: 16-20x ARR) 2. Own trained domain model (biggest multiple jump: +30-50%) 3. Data moat size (events logged, feedback entries, outcomes tracked) 4. NRR > 120% (expansion > churn) 5. Xolib Score adoption + bank API revenue What increases exit multiple most: Fine-tuned domain model + provable data asset. Build toward this from day one.

8. Funding Strategy¶

ProFIT Berlin (IBB) — Top Priority¶

Up to EUR 500K, no equity dilution, grant/loan
Requires: Berlin company seat (virtual office/co-working suffices)
Tarek has personal IBB contact — activate in Month 1-3
Prerequisite: Holding-UG structure must be in place first

ZIM (Federal innovation grant)¶

Up to EUR 310K grant
R&D focus: AI agents, domain model training
Can run parallel to ProFIT

Corporate structure (TIME-CRITICAL)¶

Setup: personal Holding-UG per founder -> both Holdings co-own Xolib GmbH 50/50
Holding-UGs MUST be founded BEFORE the GmbH (7-year lock-in otherwise)
Tax advantage at exit: ~1.5% effective tax via Par. 8b KStG vs. ~26% personal income tax
This week: Notar + Steuerberater appointment

9. Operating Reality (Stand: 22.03.2026)¶

Current status: Sprint 0-4 CODE DONE. Go-Live blocked by external tasks (DSGVO, AGB, Pentest).

Aktuelle Metriken: siehe CLAUDE.md (validiert durch scripts/validate-docs.sh).

Metrik	Wert
Operating costs	~$225/month (89% Claude Max subscription)
AI Agents	8

Infrastructure: Contabo VPS ($18/mo) + Cloudflare + PostgreSQL 16 + PostGIS 3.4 + H3 + Resend + Sentry (Production) Tech stack: Next.js 14 + TypeScript strict + Prisma + Multi-Model AI (siehe CLAUDE.md) + Tailwind + JWT + i18n Vollstaendiger Projektstand: docs/XOLIB-MASTERPLAN.md

10. Cross-Tenant Intelligence — Der naechste Moat-Level¶

Kernidee: Xolib lernt aus JEDER Interaktion JEDES Mandanten — anonymisiert, DSGVO-konform — und gibt systemweites Wissen an alle zurueck. Vollstaendiges Konzept: CROSS-TENANT-INTELLIGENCE.md

5 Saeulen: 1. Benchmark Engine — Anonymisierte Vergleichskennzahlen (BK/qm, Zahlungsquote, Score-Verteilung) ueber Kohorten (Stadt + Gebaeudeart + Groesse). k-Anonymitaet (k>=5), spaeter Differential Privacy. 2. Pattern Learning — Bank-Matching-Gewichte, Score-Normalisierungs-Breakpoints und Agent-Confidence werden aus cross-tenant Outcome-Daten kalibriert statt statisch kodiert. 3. Template Intelligence — Welche Mahnungen, Einladungen, Wirtschaftsplaene performen am besten? Best Practices werden systemweit geteilt. 4. Anomalie-Erkennung — Abweichungen eines Mandanten vom Kohortentrend loesen Fruehwarnungen aus. 5. Systemweites Wissen — Regulatorisches RAG (WEG-Recht, BetrKV), Handwerker-Intelligence, Energie-Benchmarks.

Datenmodell: SystemBenchmark, SystemModelParam, SystemTemplate, SystemAggregationJob (Prisma Models)

Implementierung: 5 Phasen ueber ~6 Monate. Phase 1 (Benchmarks, SQL-Aggregation, 2-3 Wochen) liefert sofort Kundenwert.

DSGVO: k-Anonymitaet → Differential Privacy → DSFA-Dokumentation. KEINE Mietpreis-Empfehlungen (RealPage-Kartellrisiko).

Wettbewerb: Immoware24 und Aareon haben KEIN cross-tenant Learning. Yardi Matrix (19M Einheiten) ist das Vorbild. Xolib hat alle Bausteine (Event Store, Agents, Feedback, Score) bereits — es fehlt nur die Aggregationsschicht.

Network Effect: Je mehr Mandanten → bessere Benchmarks → bessere KI → mehr Mandanten. Das ist der staerkste Data Moat einer Vertical SaaS.

11. Current Sprint Priorities (Phase 4 — Go-Live)¶

P0 — Critical (this sprint): - Tax Exemption Certificates (Par. 48 EStG) — legal obligation, direct liability - Extend service provider module: certificate number, valid-until, PDF upload - Auto-warning 30 days before expiry - Invoice check: valid cert? If not -> 5% withholding + booking suggestion - AI: Finanzwaechter checks every invoice, Rechtswaechter alerts on expired certs P1 — High (this sprint): - Operating Costs: Maintenance vs. Repair separation (Par. 1 BetrKV compliance) - Heating Cost Module + Metering Interfaces (Techem/Ista/Brunata-Metrona) - Stripe account setup (Tarek) P1 — High (next sprint): - Tenant Portal MVP (payments, tickets, documents, multilingual, QR access) - Xolib Score concept implementation Blocked: - Immoware24 competitive analysis — Faruk training 13.03.2026

12. Non-Negotiable Rules (for every Claude Code session)¶

No hardcoded text in JSX — always t('key')
No emojis — Lucide React icons only
TypeScript strict — no any, no ignored errors
Every new feature: ask "where can AI help here?"
Every new data field: ask "does this feed the Score? the moat?"
Every commit = push = auto-deploy
Update Notion after every session
i18n mandatory — en.json + de.json minimum, 8 languages target
No quick fixes that don't scale to 10,000+ Mandanten
All cross-module data must be linked (Mieter <-> Zahlungen <-> Tickets <-> Vertraege <-> Objekte)

13. Data Intelligence — 7 Cluster, 35 Predictions, 8 Industries¶

Full concept: Notion → Feature-Konzepte → "Xolib Data Intelligence — Gesamtkonzept"

7 Data Clusters (107 Prisma Models mapped)¶

Mieter-Verhalten (12 Models) — Churn 3-6 Mon, Zahlungsausfall 2-4 Mon, Zufriedenheits-Index
Gebaeude-Gesundheit (15 Models) — Komponentenausfall 6 Mon, Schadensmuster, Sanierungspriorisierung
Finanz-Intelligence (18 Models) — Cashflow 90T, BK-Optimierung, Rechnungsplausibilitaet
Markt-Signale (11 Models) — Mietentwicklung 12 Mon, Leerstand, Stranded Asset Risk
Dienstleister-Netzwerk (12 Models) — Handwerker-Ranking, Preisprognose, Kapazitaetsengpass
KI-Meta-Learning (9 Models) — Konfidenz-Kalibrierung, Kosten-Routing, Autonomie-Reife
Compliance-Radar (18 Models) — Risiko-Score, Fristen, Anfechtungsrisiko

Non-Obvious Predictions ("Pregnancy Prediction" Equivalents)¶

Gentrifizierung aus Ticket-Typen (75-85% Konfidenz, 18 Mon voraus)
Gefaehrdete Senioren aus Heiz+Verbrauchsmustern (50-65%)
Gebaeude-Strukturversagen aus Riss-Kaskaden (40-60%)
Schimmel-Epidemie im Haus aus erstem Fall + Nachbar-Heizverhalten (70-85%)

Datensatz-Meilensteine¶

Einheiten	Mandanten	Unlock
2.000	10-15	Erste Cross-Tenant Benchmarks, k-Anonymitaet
10.000	30-50	Mietpreis-Benchmarks, Zahlungsausfall-Modell 80%+
50.000	100-200	Building Health API (Banken), Risk Score (Versicherungen)
200.000	500-1.000	Portfolio Intelligence, Gentrification Index
1.000.000+	3.000+	Marktstandard Deutschland, Daten-Umsatz > SaaS-Umsatz

Revenue Projection Data Products¶

Jahr 1: 100-300k EUR (15-25% of total)
Jahr 3: 2-5M EUR (25-40%)
Jahr 5: 10-20M EUR (35-50%)

Exit-Multiplikator (CoStar-Effekt)¶

SaaS-only: 8-15x ARR. Data Platform: 15-30x ARR. Bei 20M ARR: 160-300M (SaaS) vs. 300-600M (Daten). Differenz = Data Moat Wert.

Go-to-Market (5 Kanaele)¶

Kostenlose Migration + 6 Monate gratis (OpenImmo Import steht)
Verbandspartnerschaften (VDIV 3.500 + IVD 6.000 Mitglieder)
Benchmark-Report als Lead-Magnet ("Wo stehen Ihre BK im Vergleich?")
Handwerker-Netzwerkeffekt (Handwerker empfehlen Xolib an HVs)
WEG-Beiraete als Hebel (Eigentuemer fordern Transparenz)

Kipppunkt: ~50 Mandanten (20.000 Einheiten) — Produkt verkauft sich durch Daten, nicht Features.

Last updated: March 2026 — synthesized from Strategy Chat + Notion Project Hub + Data Intelligence Research