The Specificity Playbook
How to seed supply in one city, build a taxonomy where segmentation actually works, and ship the directory and community in ninety days.
Seeding supply in one city
Start with three verticals, not five. Colorists, nail techs, and brow/lash techs have the highest churn, the strongest dependence on word of mouth, and the weakest platform loyalty. Derms are low churn and already served by Zocdoc and RealSelf. Trainers and tailors come after the loop is proven.
Seed supply from demand, not from scraping. Recruit the founding member cohort first, then ask each one who she already goes to and recruit those providers. The directory is vetted by construction. Fill gaps by compiling from Instagram, Booksy, Fresha, StyleSeat, and Yelp.
Do not build booking. Link out to whatever the provider already uses. Booking is a commodity where incumbents own the lock-in. Your product is matching, so the provider profile is structured around who they are good for, not around their calendar.
The provider pitch is fit, not volume
A verified profile with "experienced with" tags (lived-in color on level 2 Asian hair, hand-tied extensions on fine hair, brow mapping for hooded eyes) plus a portfolio sorted by client attributes. The value to them: fewer mismatched consults, clients who arrive pre-qualified.
Define "vetted" precisely
- Three attributed reviews from members with declared facets.
- A portfolio with at least six outcome photos.
- An automated license check. New York State cosmetology and nail licenses are publicly searchable through the Department of State, so this is a nightly job.
Make churn a feature
Members follow providers. A provider move is a first-class event that triggers notifications and a community thread. Run a monthly provider check-in and let members flag moves. This is the recurring hook that Reddit cannot offer.
| Day-90 supply target | Number |
|---|---|
| Providers across three verticals | 150 |
| Providers with a verified license | 80% |
| Providers with three or more attributed reviews | 60 |
| Neighborhoods with five or more providers per vertical | 8 |
How the directory and community reinforce each other
Taxonomy that makes segmentation work
Separate three layers. Member attribute facets are the segmentation variables. The service and provider ontology is what gets filtered. Content types are the third layer. Segmentation lives only in facets and is applied as a lens over the other two.
Facets, each with a small controlled vocabulary and a reference chart
Three admission rules for any facet
- A provider outcome must plausibly depend on it.
- A member must be able to self-assess it from a chart.
- The resulting segments must have at least thirty NYC members, or the facet gets collapsed.
Do not create a forum per segment. Empty rooms kill communities. Use one feed where every post carries a frozen snapshot of the author's facets at posting time (hair and skin change). Members filter by "people like me" with a weighted similarity score, weights set per vertical: hair facets dominate for colorists, eye shape for lash, skin for derms. Show "matches you on 5 of 7" rather than a hard filter.
Reviews count toward a provider's "for people like you" score only when the reviewer shares the vertical-relevant facets. Store facets as versioned key-value pairs, because the taxonomy will change and you will migrate.
Ancestry needs one hard rule. Providers tag "experienced with," never "serves only," and providers can never filter clients. Members self-segment. Under the NYC Human Rights Law the opposite design is a liability.
| Vertical | Facets that carry weight | Facets that don't |
|---|---|---|
| Colorist | hair type, texture, chemical history, gray coverage | eye shape, skin concerns |
| Nail tech | neighborhood, budget, skin tone + undertone (color recs) | hair, eye shape |
| Brow / lash | eye shape, brow density, skin tone | hair type, chemical history |
| Derm (later) | Fitzpatrick, undertone, concerns, ancestry | hair, budget |
Cold-start content
Structure, not conversation. Three formats seed the archive before anyone is chatting.
- Outcome cards. Photo, provider, service, price, facet snapshot, and a rebook yes or no. Two hundred cards from fifty founding members makes the directory feel alive on day one.
- Post-transaction prompts. The spend tracker already knows she spent money at a merchant on Tuesday. Ask for a ten-second rating. Zero-effort, purchase-verified content is the advantage Reddit and TikTok cannot copy.
- Ask-a-segment threads. "4C, color-treated, UWS: who does your gloss?" Founding members answer within the day.
Recruit the founding cohort for coverage, not convenience
A NYC consultant network drifts toward one demographic. Build a coverage table across the facets and recruit against the gaps. Keep the cohort in a WhatsApp or Discord group for the first month while you build, then migrate them in.
Founder-authored content twice a week
Ranked lists like "colorists who do lived-in balayage on level 2 Asian hair, rated by six members" are both SEO pages and group-chat-shareable objects.
| Month-one content target | Number |
|---|---|
| Founding members | 50–100 |
| Outcome cards | 200 |
| Members who post at least once | 30% |
| Founder-authored ranked lists | 8 |
What moderation costs
| Stage | Setup | Approximate cost |
|---|---|---|
| Under 2k members | Founder plus two volunteer mods with founding-member status | ~5 hrs / week |
| Every post and review | LLM pre-screen (Haiku 4.5): spam, provider self-promotion, medical claims, doxxing | < $10 / day at 1,000 items |
| Over 5k monthly actives | Part-time community manager, 20 hrs / week | $2.5k–3.5k / month |
Four rules to write before launch
Providers will contact you within two weeks. Have these ready.
- Astroturfing. Reviews need verified spend or an account with real age and declared facets.
- Negative reviews. New York defamation exposure means a documented takedown and dispute process, and keeping receipts.
- Medical content. Derm and biomarker talk gets a disclaimer and a block on dosage discussion.
- Body content. No numeric weight-loss goals in cohorts. Protects the community's tone and your liability.
The first 90 days
Stack. Next.js, Postgres on Supabase, faceted search via Typesense or pgvector, an LLM for tagging and moderation. Your RAG experience maps directly to "ask the community" retrieval over facet-stamped posts, which is the feature that makes the archive compound.
- Data model: members, facets (versioned), providers, services, reviews, posts, facet snapshots.
- Provider ingestion pipeline, half scraped and half manual.
- Facet onboarding with reference charts.
- Directory MVP: vertical, neighborhood, "people like me."
- Founding cohort running in a group chat in parallel.
- Outcome cards and reviews.
- Post-transaction prompt wired to the spend tracker.
- Similarity-scored feed.
- Provider claim flow with license verification.
- Migrate the group chat into the product.
- Follow-provider and move alerts.
- Ask-a-segment threads.
- LLM moderation queue with a human review lane.
- Provider self-serve profiles.
- Weekly metrics dashboard.
| Day-90 success criteria | Target |
|---|---|
| Members | 100 |
| Attributed reviews | 300 |
| Members with five or more facets declared | 40% |
| Directory searches returning three or more "people like me" results | 60% |
| Members who posted at least once in month one | 30% |
Real risks
Segment × vertical × neighborhood is combinatorial and most cells stay empty for a long time. Similarity scores instead of hard filters, and the narrow launch footprint, are the mitigations. No second city until the search-coverage metric holds.
Ancestry, skin, body, and eventually biomarkers joined to transaction history is a profile that demands real care. Treat it under the New York SHIELD Act now, encrypt at rest, opt in per facet, and make privacy a visible product feature. The biomarker layer needs HIPAA-adjacent design from day one.
Being rated by segment will irritate some providers. The dispute process and the "experienced with" framing keep this defensible.
Paid placement kills the trust that is the whole product. Plan for provider verified-pro subscriptions and a member subscription tied to the ops-for-body layer. Not ads, and not a booking take-rate you don't own.
No booking, no chat, no native app in the first 90 days. Each is a month you don't have.