Cleanroom positioning — privacy-preserving collaboration as default¶
Goal: turn Husky's existing Snowflake Marketplace presence into a cleanroom-native offering. Cleanroom support is no longer optional in 2026 — cookie deprecation + GDPR/DPDP/CCPA tightening make it table stakes for audience suppliers. Buying a third-party cleanroom is overkill at Husky's stage; building our own is strategic mis-investment. Snowflake Data Clean Rooms is free, edition-included, and already installed on our account — first-mover ROI is immediate.
Build vs buy decision¶
| Option | Time-to-market | Cost | Strategic fit | Verdict |
|---|---|---|---|---|
| Build own cleanroom | 6-12 months + ongoing engineering | $500K+ + 2-3 FTE | ❌ Wrong layer for an audience supplier | Skip |
| Snowflake Data Clean Rooms (native) | Already installed | $0 license, only warehouse credits | ✅ Native to existing distribution channel | Adopt — Phase 1 |
| AWS Clean Rooms | 1-2 weeks integration | ~$10K integration | ✅ Covers non-Snowflake AWS buyers | Phase 2 (Q3 2026), after AWS DX seller appeal clears |
| LiveRamp Safe Haven | 4-8 weeks integration | License + revenue share | ✅ Identity-resolved use cases | Phase 3 (2027 H1), alongside LiveRamp distribution |
| InfoSum / Decentriq | 8-12 weeks | $50K-500K/year + revenue share | ❌ Designed for enterprise/banking; oversized for our volume | Skip |
Husky is a data supplier. Cleanroom is infrastructure. Suppliers buy/integrate; only platform companies build.
Setup status (as of 2026-04-27)¶
Snowflake Data Clean Rooms Native App is fully operational on Husky's account WIB77365:
| Step | Status | Date |
|---|---|---|
| Snowflake account on Enterprise Edition (DCR-supported) | ✅ Confirmed via $3/credit pricing | Pre-existing |
SNOWFLAKE_DATA_CLEAN_ROOMS Native App installed |
✅ Auto-installed since 2024-05-08 | Pre-existing |
| App auto-upgraded to v1.0 | ✅ Patch 120, last upgraded 2026-04-23 | Pre-existing |
APP_USER application role granted to ACCOUNTADMIN |
✅ | 2026-04-27 |
CREATE DATABASE and EXECUTE TASK privileges granted to app |
✅ | 2026-04-27 |
ONE_TIME_SETUP() procedure invoked |
✅ Returns "Success" | 2026-04-27 |
SAMOOHA_SAMPLE_DATABASE created with templates + sample data |
✅ | 2026-04-26 |
Marginal cost: zero license fee. Only Snowflake warehouse credits when a query runs (estimated $10-60 per buyer overlap session on DEV warehouse). Provider-side setup is single-digit credits per cleanroom.
Buyer flow — three use cases¶
Use case 1 — Pre-purchase overlap analysis ("test before you buy")¶
Buyer pain: "How much of my CRM does Husky's APAC Travel Luxury segment cover before I commit to a paid bundle?"
Today (without cleanroom): buyer sends Husky a hashed-email list under NDA, Husky runs an offline overlap query, returns a number. Multi-day cycle, requires legal review for the data exchange.
With cleanroom: 1. Husky publishes a "Husky × Buyer-CRM Overlap Estimator" cleanroom template 2. Buyer accepts the cleanroom invitation, attaches their CRM (encrypted, never visible to Husky) 3. Both sides compute overlap inside the cleanroom; output is a single aggregate count 4. 30 minutes end-to-end, zero PII exchange, zero legal risk
Output: overlap_pct = 23.4% ; joint_size = 234,000 users (no row-level data shared)
Use case 2 — Lookalike modelling¶
Buyer pain: "I have a 50K seed audience of high-value customers. Expand to similar APAC users at scale."
With cleanroom: buyer's seed + Husky's full APAC population get joined inside a cleanroom-bound LOOKALIKE_BUILD template. Husky's segment behaviour signals are used as features; output is a scored expansion audience with similarity ≥ threshold.
Output: ranked expansion audience identifiers, accessible via Snowflake share to buyer; Husky's raw segment matrix never leaves Husky's side.
Use case 3 — Closed-loop attribution¶
Buyer pain: "Did my campaign on Husky-targeted users actually drive lift? I don't want to send my conversion log to a vendor."
With cleanroom: buyer attaches their conversion log; cleanroom joins to Husky's exposed-audience flag at user-level; computes test-vs-control lift; returns aggregate measurement metrics. Buyer's conversion data never visible to Husky; Husky's user-level audience flag never visible to buyer.
Output: lift = 18.2% ; confidence_interval = [12.1, 24.3] ; per-segment breakdown.
Provider flow (what Husky operates)¶
Built on Snowflake's SAMOOHA framework + native SNOWFLAKE_DATA_CLEAN_ROOMS app:
1. Provider studio: define a cleanroom template (one of the three use cases above)
2. Bind it to a Husky data source — typically HUSKY_DATA.SHARE.V_MAID_APAC_PLUS_US
or a curated segment subset
3. Set policies: aggregation thresholds (k-anonymity ≥ N), allowed query patterns,
audit logging
4. Issue a cleanroom invitation to a specific buyer Snowflake account
5. Buyer accepts, attaches their input data (CRM / conversions / seed)
6. Buyer triggers the template; cleanroom executes inside Snowflake's TEE-protected
compute; outputs aggregate result
7. Husky monitors session metadata via Provider Studio; row-level data on either
side stays in its origin
Key design constraint: Husky's existing SHARE views (V_MAID_APAC_PLUS_US, V_HEM_US, V_COVERAGE_APAC_US, SEGMENT_CATALOG) are reusable as cleanroom inputs without modification. The schema is already cleanroom-friendly: (user_id, seg_list, country, platform) with IAB_Taxonomy_1.1 segment IDs.
Marketing positioning¶
Hook for buyers¶
"Cleanroom-ready by default — match your CRM against APAC + US audiences without exchanging a single row of PII."
Three buyer-facing claims (use across web, sales decks, Snowflake listing copy)¶
- Test before you buy — pre-purchase overlap analysis in 30 minutes via Snowflake DCR. No NDA, no data movement.
- Compliance built in — GDPR / DPDP / PDPA / CCPA all permit cleanroom-based collaboration. Husky as Singapore-licensed provider + cleanroom = cross-border compliance fully resolved.
- APAC behaviour × US addressability, privacy-preserved — Husky's flagship cross-border product (US-addressable MAIDs carrying APAC behavioural attributes) is uniquely cleanroom-friendly because the differentiator is attributes, not raw identifiers. Cleanroom ships the value while keeping the identifier graph private.
What we are NOT claiming¶
- Not "Snowflake-certified Cleanroom Partner" (no such tier exists yet)
- Not "we built our own cleanroom" (we use the Snowflake-native one)
- Not "100% PII-free" (k-anonymity threshold is the contract; small queries can still leak signal)
Three-phase rollout¶
Phase 1 (now → 2026-Q2): Snowflake DCR live, marketed
─────────────────────────────────────────────────────
✅ App installed + setup complete
□ Build first sample cleanroom template ("APAC Mobile Reach Estimator")
□ Update Snowflake Marketplace listing copy to mention cleanroom support
□ Update huskydata.io: partners page (done 2026-04-27 via this commit),
homepage hero, /products pages
□ One-pager PDF: "How to overlap-test Husky audiences in 30 minutes via DCR"
□ Demo run with 1-2 friendly buyers; capture testimonial
Phase 2 (Q3 2026): AWS Clean Rooms second channel
──────────────────────────────────────────────────
□ Wait for AWS DX seller appeal #00874776 to clear
□ Build AWS Clean Rooms equivalent of the Snowflake template
□ Mirror marketing across both channels
□ Buyer choice: Snowflake-native or AWS-native, same offering
Phase 3 (2027 H1): LiveRamp Safe Haven for identity-resolved use cases
─────────────────────────────────────────────────────────────────────
□ Confirmed alongside LiveRamp distribution channel (already planned Q3 2026)
□ Adds RampID-keyed cleanroom collaboration for CTV/publisher buyers
□ Scope limited to USA + Cross-border products (per partners.js)
Risk register¶
| Risk | Likelihood | Mitigation |
|---|---|---|
| Buyer doesn't have Snowflake account → can't use DCR | Medium for non-warehouse buyers | Phase 2 covers AWS; plain Marketplace SHARE remains for non-cleanroom buyers |
| Cleanroom queries reveal too much (small k) → privacy regression | Low if thresholds set | Set k-anonymity ≥ 100 by default; reject queries below threshold; audit log |
| Snowflake DCR feature deprecated / restructured | Low (active product) | Architecture is layered — only template SQL changes if API moves; data shares survive |
| Compute cost surprise from buyer queries | Low (buyer pays own credits) | Provider-side warehouses isolated; usage monitoring in Provider Studio |
| Marketing claim "cleanroom-ready" before any buyer has actually used it | Medium (we say "ready" not "proven") | Get a Phase 1 demo / testimonial within 60 days to back the claim |
Provenance¶
Drafted 2026-04-27 after hands-on verification that DCR is operational on Husky's account (ONE_TIME_SETUP returned Success same day). Builds on the data-plane review captured 2026-04-26 in docs/strategy/2026-04-26-pipeline-automation-strategy.md. Website marketing copy committed alongside this doc — src/content/partners.js Snowflake card now lists Cleanroom-ready capability + 4 capability bullets.