Strategic Risks and Open Questions¶

A honest inventory of what could derail PropPie, and what we don't yet know.

Risk register¶

Risks are ranked by expected impact (likelihood × severity). High-priority items get a defence plan.

#	Risk	Likelihood	Severity	Mitigation
R1	SEBI guidance pulls AI conversational tools into IA territory	Low-Medium	High	Build info-only by design (already SOUL). Worst case: register as IA (planned escape hatch).
R2	SM-REIT compliance gap in existing Fractional schemes	Unknown — must audit	High	Urgent legal audit; restructure if needed; transparent investor communication
R3	Major hallucination incident with reputational fallout	Medium	High	Source-citation discipline, eval framework, AI red-team process, "I don't know" defaults
R4	Defamation suit from a developer over derived scores	Low-Medium	Medium-High	Percentile framing, citation, right of reply, never "fraudulent" language
R5	Pan-India incumbent (PropEquity, NoBroker) launches an info-only AI broker faster than us	Medium	High	We can't out-spend; we can out-discipline. Lock in Maharashtra depth before they re-architect.

#	Risk	Likelihood	Severity	Mitigation
R6	MahaRERA / IGR portal structure changes break scraping	Medium-High (over 2 yr)	Medium	Modular scrapers with monitoring; fallback to licensed feeds (PropEquity, CRE Matrix)
R7	Data licensing partner pulls / hikes prices	Medium	Medium	Diversify sources; self-scrape what's reachable; budget for upside
R8	LLM provider price/API change	Medium	Medium	Provider-agnostic abstraction in code; benchmark open-source models periodically
R9	Fractional issuer event (asset defaults, vacancy spike, manager fraud)	Low-Medium per asset	High per-incident	Asset diligence, conservative underwriting, clear investor disclosure, insurance
R10	DPDP rule notification requires significant rework	Medium	Medium	Build privacy-first from foundation; vault + masking; periodic audit
R11	AI cost runs higher than projected	Medium	Medium	Caching, batch inference, smaller models for cheap tasks, RAG over fine-tuning
R12	Talent retention — key engineer leaves	Medium	Medium	Mission alignment, documentation (this repo), ESOPs, dual-coverage

#	Risk	Likelihood	Severity	Mitigation
R13	Foreign player enters India proptech	Low-Medium (over 3 yr)	Medium	Localisation moat; relationships; data depth
R14	Macro real-estate downturn	Medium (cyclical)	Medium	Our value rises in downturns (more scrutiny per deal); B2B subscribers may cut budgets
R15	Cyberattack / data breach	Low-Medium	High if happens	Security-first architecture, encryption, access controls, breach process
R16	Founder burnout / co-founder conflict	Low-Medium	High	Clear role separation (CEO/COO documented), governance, advisors
R17	AI safety / public sentiment shift against AI advice	Low-Medium	Medium	Our info-only positioning insulates us; double-down on honesty
R18	State-level regulatory shift in Maharashtra	Low	Medium	Document compliance posture; engage state authorities professionally

These are not yet decided. They will be resolved over the next few sessions.

Q1: What's the existing PropPie Fractional regulatory structure (SM-REIT / AIF / SPV / private placement)?
Q2: What does the existing PropPie Fractional tech stack look like? (Determines how AI overlay integrates.)
Q3: Which channel for B2C Broker — web-first, WhatsApp-first, or mobile-app-first?
Q4: Bilingual (Hindi + Marathi) priority — Day 1 launch or later?
Q5: Voice interface — phase 1 or phase 2?
Q6: B2B Analytix — start with developers, brokers, or institutional investors as design partners?
Q7: Pricing model for Broker — free with affiliate revenue, freemium, subscription, per-report?
Q8: Pricing model for Analytix — seat-based, AUM-based, tiered?

Q9: Data licensing budget — do we license PropEquity / CRE Matrix data, or self-scrape only?
Q10: Buy CIBIL Commercial for developer credit scoring, or use proxies?
Q11: Buy social listening (Brandwatch / Talkwalker), or self-build?
Q12: News feed — vendor (Aggregate Intelligence) or self-scrape?
Q13: How do we handle Marathi documents — neural MT, fine-tuned model, or human-in-loop?
Q14: Sub-day IGR ingestion — which priority micromarkets first?

Q15: LLM strategy — OpenAI, Anthropic, AWS Bedrock, Azure OpenAI, or self-hosted open-source?
Q16: Vector DB — managed (Pinecone, Weaviate Cloud) or self-hosted (Qdrant, pgvector)?
Q17: Fine-tune vs. RAG-only — which derived attributes need fine-tuning?
Q18: Latency targets for B2C conversational — <2s for simple, <8s for complex?
Q19: Eval framework — automated, human-in-loop, or hybrid?

Q20: SEBI legal opinion confirming info-only stance — who do we engage?
Q21: DPDP — register as Significant Data Fiduciary (likely yes if scale)?
Q22: Right-of-reply mechanism for developer scores — what's the public process?
Q23: When (if ever) do we register as SEBI IA or RA?
Q24: How do we handle a developer suing over a score — playbook?

Q25: Among 5 developers, who owns which workstream?
Q26: When do we hire the first dedicated AI engineer (separate from full-stack)?
Q27: Compliance officer designation — internal or external?
Q28: Advisor board — who, when?

Q29: Funding — bootstrap from SORANA, or raise pre-seed/seed? Timing?
Q30: Geographic expansion — Karnataka or Gujarat first after Maharashtra?
Q31: Partnerships — IPC firms (Knight Frank etc.) as channels? On what terms?
Q32: PR strategy — when do we go public with the Honest Broker positioning?

If, in 18-24 months, a well-funded incumbent successfully copies our positioning:

Deepen Maharashtra to a level they can't match — micromarket-by-micromarket investor confidence
Lean into community — a buyer-side network of users who expect source-citation
Open-source the data dictionary (not the data) — set the industry standard
Become the trust certification layer — even competitors integrate with us for citation