Diligence
AI-native operating model
Not "AI-powered" as positioning. The directory that would traditionally need an operations team was built by a pipeline, and the run logs show it.
Automated — shipped and verified
- Business discovery across zones and categories
- Classification and enrichment
- Deduplication and brand/branch resolution
- Area banding and scoring
Automated — designed, not yet live
- Conversation seeding and community follow-up questions
- Content tagging
- Spam triage
- Advertiser reporting
- Trend detection
AI with human review
- Moderation escalation
- Claims naming individual service providers
- Listing disputes
- Quality audit of published pages
Human owned
- Product and strategy
- Closing advertisers
- Brand partnerships
- Strategic accounts
The claim is operating leverage, not a company without people. Where today's models aren't reliable enough — anything involving reputation, disputes or money — humans own the work outright.
The modelled headcount comparison
Held off the main narrative as supporting evidence rather than a headline. Human sales headcount is identical in both columns.
| MAU | Traditional FTE | AI-native FTE | Traditional OpEx | AI-native OpEx |
|---|---|---|---|---|
| 1M | 102 | 24 | ₹9.42cr | ₹2.42cr |
| 5M | 511 | 121 | ₹47.09cr | ₹12.12cr |
| 10M | 1,022 | 242 | ₹94.17cr | ₹24.24cr |
Sanity check: Yelp runs roughly 4,400 employees against a comparable local-review audience — about 4.4 FTE per 100k MAU, sitting between this model's traditional benchmark (10) and its AI-native assumption (2.2). That is the right place for it to sit.
Geographic expansion economics
| Footprint | Areas | Languages | Build | Maintain / yr | Localisation | First year |
|---|---|---|---|---|---|---|
| Bangalore | 12 | 2 | ₹0.8L | ₹0.5L | ₹36L | ₹37L |
| Top-8 metros | 96 | 5 | ₹6.3L | ₹3.7L | ₹90L | ₹100L |
| 50 cities | 280 | 9 | ₹18.3L | ₹10.9L | ₹162L | ₹191L |
| 100 cities | 420 | 11 | ₹27.4L | ₹16.4L | ₹198L | ₹242L ($252k) |
Two findings. First, supply construction across 100 cities costs about $28,500 — less than two operations hires, and the figure the narrative headlines. Second, localisation is 82% of the first-year geographic cost, not data acquisition. The ₹2.42cr first-year total is kept here rather than on the main page: it is correct, but it bundles a large modelled localisation component and reads as a budget rather than an insight.
The precise claim, and its limit
Supported: national supply coverage scales on software. Mapping and maintaining beauty businesses across 100 cities needs no proportional field organisation.
Not supported: that total operations headcount stays flat. Moderation scales with content volume, native-language QA with languages, and advertiser servicing with advertisers.
The defensible sentence is therefore narrower than the ambitious one: Glow can achieve national geographic coverage without a national field organisation, because supply construction and maintenance run on software. Community and advertiser operations still scale with usage and revenue — sub-linearly against a traditional operator, but not flat.