Risk register

Twenty risks are registered in full. These are the ones that can change or end the business.

1 · The community browses but never contributes

Why it matters: without contributors there is no knowledge graph, no intent data and no unique page content — Glow would be a thinner Google Maps.

What we're testing: contribution measured separately for votes, comments and answers across ~2,000 users in 90 days. Below 1.5% is a clear signal.

Response: AI community agents ask one contextual follow-up per human post; voting is the low-friction entry point. If contribution genuinely fails, the fallback is a curated editorial directory — smaller, but real.

2 · Businesses like the free listing but won't pay

Why it matters: this is the binding constraint. Break-even needs roughly ₹2,499 ARPU at 25% trial-to-paid — above Justdial's blended ₹1,708.

What we're testing: 200 field visits, price randomised across three cells, each trial followed to a paid decision within 45 days.

Response: sell measured outcomes, not impressions. If subscription advertising fails, lead-generation and commission models get tested before the category is abandoned.

3 · Consumer acquisition never reaches escape velocity

Why it matters: paid channels alone plateau the model. Scale requires organic search, referral and business cross-promotion compounding together.

What we're testing: four channel cells — Meta, creators, salon QR, referral — each tracked to cost per activated user and week-4 retention.

Response: concentrate on one area. The model encodes a density gate: no new area opens until existing areas clear the level at which advertisers will pay.

4 · Outbound acquisition runs into telecom regulation

Why it matters: India's TCCCPR regime has no clean B2B carve-out, and most listed business numbers are personal mobiles that may be DND-registered.

Response: the model was already rebuilt onto compliant channels — field visits, inbound listing claims, post-consent follow-up. A written legal opinion precedes any outreach programme. AI-assisted calling is excluded from the model.

5 · Programmatic SEO triggers a quality problem

Publishing directory pages without unique content meets Google's own definitions of scaled content abuse and doorway abuse. Mitigation: a page publishes only when it carries genuine community content. Page count capped by content, not database size. No bulk publish in the first 90 days.

6 · Google Maps / Instagram substitution

Most consumers will keep using both. Glow has to be better at something specific — the individual provider, service-level comparison, community opinion — not better in general.

7 · Advertiser churn exceeds the model

Base assumes 5%/month. Justdial's paid campaigns grew 3.5% against 13% listing growth, implying meaningful incumbent churn. Above 10%/month the acquisition engine is a leaky bucket. Cannot be resolved inside 90 days — early conversion results stay provisional until retention is observed.

8 · Named service providers

The core loop names private individuals with no account and no right of reply. Positive attribution is benign; a negative allegation about a named stylist is a defamation and harassment risk. Mitigation: asymmetric moderation — positive attribution publishes freely, negative claims about named individuals are held for human review or confined to business-level criticism, with notice-and-takedown and a right of reply.

9 · Privacy and DPDP

Rules notified November 2025 with an 18-month runway. Consent must contemplate aggregated analysis at signup or the intelligence layer is foreclosed later. Designed now, while the user base is zero.

10 · Key-person and platform dependency

Founder dependency is real and unmitigated at this stage; the internal model carries a replacement-cost sensitivity. Google Places, Vertex and Supabase are external dependencies — mitigated by Glow entities being proprietary and portable, with only place IDs stored indefinitely.