How to build idempotent deduplicated per-location-aware CRM record creation

The 4-skill bundle on the crm-record-creation agent

One agent. Four coordinated skills. The Fingerprint + Dedupe + Stitch + Audit bundle runs above the CRM + marketing-CRM + CDP + commerce-identity substrate and writes one canonical per-customer per-event idempotent- record.

Compliance overlay

Five anchors run per-customer per-event before any record commits downstream. The first anchor is operationally distinctive: per-vendor Idempotency-Key semantics + dedup- window strategy + per-CDP identity-graph + per-location- aware identity-resolution + omnichannel customer-stitching converge on every CRM-record-creation event.

FAQ

What is idempotent deduplicated per-location-aware CRM record creation — and what is the per-vendor-Idempotency-Key-times-per-CDP-identity-graph-times-omnichannel-customer-stitching problem distinctive to this skill?

An omnichannel DTC ecommerce operator with 5-50M customer records ships 10-500M CRM-record-creation events per month across 8-15 CRM + marketing-CRM + CDP destinations (Salesforce + HubSpot + Klaviyo + Iterable + Attentive + Postscript + Customer.io + Braze + Twilio Segment + RudderStack + Tealium + Lytics + mParticle + Stripe + Shopify Customer). Each vendor ships its own Idempotency-Key header semantics + upsert primitive + 4xx/5xx retry semantics + rate-limit. Each CDP ships its own identity-graph + identity-resolution algorithm. Each customer touches multiple channels (email + phone + device + storefront + fulfillment-center) requiring omnichannel stitching. The four-skill bundle on the crm-record-creation agent — Fingerprint, Dedupe, Stitch, Audit — sits above the CRM + marketing-CRM + CDP + commerce-identity substrate and writes a per-customer per-event canonical idempotent-record. The operationally distinctive anchor: idempotency-key generation strategy (canonical-event-hash + content-hash + composite-business-key + per-vendor Idempotency-Key header semantics) + dedup-window strategy (W-minute + W-hour + W-day + W-week sliding windows + per-vendor upsert window) + per-CRM-vendor idempotency semantics (Stripe Idempotency-Key 24-hour TTL + Klaviyo + Iterable + Salesforce Bulk API 2.0 External Id upsert + HubSpot Batch API + Attentive + Postscript) + per-vendor 4xx/5xx idempotent-retry semantics (exponential backoff + jitter + Conflict 409 dedup + 429 retry-after + circuit breaker + dead-letter queue) + per-CDP identity-graph (Segment Personas deterministic + probabilistic match + RudderStack + Tealium + Lytics + mParticle) + per-location-aware identity-resolution (per-fulfillment-center + per-retail-store + per-warehouse identity-stitching) + omnichannel customer-stitching (email + phone + device-id + IDFA/AAID + cookie + Klaviyo profile-id + Stripe customer-id + Shopify customer-id) + per-event idempotency-fingerprint (canonical-event-hash + SHA-256 + Blake3 + xxHash + deterministic-serialization) + Outbox Pattern + Transactional Outbox + Change Data Capture (CDC) via Debezium + Kafka Connect + Maxwell + Striim + CloudEvents 1.0 + Lamport timestamp + Hybrid Logical Clock (HLC) causal ordering + idempotent consumer (exactly-once semantics via offset commits + Kafka transactions).

Why do Salesforce + HubSpot + Klaviyo + Iterable + Segment + RudderStack break at omnichannel-DTC-millions-of-customers-per-month scale?

Each CRM + CDP vendor ships per-account flat record-creation primitive at single-vendor single-channel level. None coordinates per-customer per-event idempotency-fingerprint generation + dedup-window matching against per-vendor Idempotency-Key semantics + per-CDP identity-graph + per-location-aware identity-resolution + omnichannel customer-stitching. None handles per-vendor 4xx/5xx idempotent-retry semantics + Conflict 409 dedup + 429 rate-limit retry-after + circuit breaker + dead-letter queue at the cross-vendor level. None gates against TCPA 47 USC 227 SMS consent + CAN-SPAM email consent + GLBA financial-customer + HIPAA health-customer + per-state biometric. None enforces SOX 302/404/906 when public-company CRM material. None writes a per-customer per-event WORM idempotent-record audit trail with regulatory-defense retention. The four-skill bundle Fingerprint + Dedupe + Stitch + Audit sits above the CRM + CDP + commerce-identity substrate — it does not replace it.

How does Fingerprint + Dedupe work?

Fingerprint runs per-event idempotency-fingerprint generation: canonical-event-hash + SHA-256 + Blake3 + xxHash + deterministic-serialization. Per-event composite-business-key (customer-id + event-type + event-timestamp + per-vendor-event-id + per-source-fulfillment-center). Per-vendor Idempotency-Key header generation conforming to per-vendor semantics (Stripe Idempotency-Key 24-hour TTL + Klaviyo + Iterable + Salesforce Bulk API 2.0 External Id + HubSpot Batch API + Attentive + Postscript). CloudEvents 1.0 envelope wrap with Lamport timestamp + Hybrid Logical Clock (HLC) for causal ordering. Dedupe runs per-event dedup-window matching: W-minute sliding window (W=1, 5, 15) + W-hour (W=1, 6, 24) + W-day (W=1, 7, 30) + W-week. Per-vendor upsert-window matching against per-vendor Idempotency-Key TTL. Per-vendor 4xx/5xx idempotent-retry: exponential backoff (base * 2^attempt) + jitter (full-jitter + decorrelated-jitter) + Conflict 409 dedup-confirm + 429 retry-after + 503 circuit-breaker (half-open after N seconds) + dead-letter queue (Kafka DLQ + AWS SQS DLQ).

What does Stitch + Audit do?

Stitch runs per-customer omnichannel identity-stitching via per-CDP identity-graph (Segment Personas deterministic + probabilistic match + RudderStack Identity Graph + Tealium AudienceStream + Lytics + mParticle Identity Sync) + per-location-aware identity-resolution (per-fulfillment-center + per-retail-store + per-warehouse). Stitching keys: email + phone (E.164) + device-id (IDFA/AAID) + cookie + Klaviyo profile-id + Stripe customer-id + Shopify customer-id + Salesforce Lead/Contact ID + HubSpot Contact ID + Iterable user-id. Identity-graph merge-rule resolution: most-recent-wins + most-confident-wins + per-channel priority + per-vendor source-of-truth. Per-customer canonical identity-record. Gate runs 5 anchors per-customer per-event before any record commits downstream. (1) Per-vendor Idempotency-Key semantics + dedup-window + per-CDP identity-graph + per-location-aware identity-resolution + omnichannel customer-stitching + Outbox Pattern + CDC + CloudEvents 1.0 + Lamport + HLC + idempotent consumer + exactly-once semantics. (2) FTC Section 5 + Pfizer 1972 + CFPB UDAAP + Lanham + USPTO + Robinson-Patman + per-state UDTPA. (3) TCPA 47 USC 227 SMS consent + CAN-SPAM email consent + GLBA financial-customer + HIPAA 45 CFR 164.502/504/514 health-customer + COPPA when child-customer + per-state biometric (BIPA + CUBI + Washington biometric) when face/voice. (4) EU AI Act Article 50 transparency when AI-ML identity-resolution + Article 13/14/15 + Annex III when AI-ML CRM record-creation drives customer-routing + Article 6/27 FRIA + DSA + DMA + GDPR Article 6/7/28/30 + LGPD + DPDP + PIPEDA + Quebec Law 25 + CCPA + CPRA + 18-state. (5) WCAG 2.2 AA + ARIA + EAA + ADA Title III + Section 508 + SOX 302/404/906 when public-company CRM material + COSO + Exchange Act 13(b)(2) + SEC Reg S-K. Audit writes a per-customer per-event WORM idempotent-record: event snapshot + idempotency-fingerprint + per-vendor Idempotency-Key + dedup-window match + identity-graph resolution + per-anchor gate-pass + AI-ML provenance + EU AI Act FRIA. Retention: 7-year FTC + 7-year IRS + 7-year HIPAA + 7-year GLBA + 7-year state bar + 6-year SEC + 3-year FINRA + 7-year SOX + GDPR Article 30 + EU AI Act Article 12 + SOC 2 CC7/CC8.

What does this skill connect to on the crm-record-creation agent and across the swarm?

On the crm-record-creation agent: per-customer omnichannel identity-resolution + per-customer event-stream ingestion + per-vendor CRM record-upsert. Across the swarm: per-vertical catalog schema validation (#597 same per-vendor Idempotency-Key + Outbox Pattern substrate) + integration-drift-monitor agent (#562 + #569 + #570 same per-vendor schema-drift substrate) + per-state-overlay-composer (#599 UPSTREAM canonical for TCPA + CAN-SPAM + GLBA + HIPAA per-state overlays) + governance-decision-router five-destination routing + tiered pre-filter deterministic gates + per-vertical compliance overlay. Commercial-pillar parent: /customer-identity-resolution.

What does the 6-workstream pre-engagement-baseline reporting cycle look like for this skill?

Every two weeks during the Tier 3 Fractional CMO with AI Swarm engagement, six workstreams report against the pre-engagement baseline. Workstream 1: per-customer per-event idempotent-record-creation coverage — customers monitored + events fingerprinted + vendors integrated. Workstream 2: Fingerprint per-event idempotency-fingerprint flow — canonical-event-hash + per-vendor Idempotency-Key generation + CloudEvents envelope + Lamport/HLC ordering. Workstream 3: Dedupe per-event dedup-window matching flow — W-minute/W-hour/W-day/W-week sliding window + per-vendor upsert-window + per-vendor 4xx/5xx retry result. Workstream 4: Stitch per-customer omnichannel identity-stitching flow — per-CDP identity-graph + per-location-aware identity-resolution + omnichannel stitching keys + merge-rule resolution. Workstream 5: Regulatory-defense audit coverage — per-vendor Idempotency-Key semantics + per-CDP identity-graph + omnichannel customer-stitching + TCPA + CAN-SPAM + GLBA + HIPAA + per-state biometric + EU AI Act Article 50 + SOX. Workstream 6: FBC feedback-loop pattern-learning — per-customer realized-vs-predicted dedup-rate + per-vendor Idempotency-Key conflict retrospective + per-CDP identity-graph merge retrospective.

Related reading

• Parent commercial pillar: customer identity resolution
• Sibling build-pillar: per-vertical catalog schema validation (#597 same per-vendor Idempotency-Key + Outbox Pattern substrate)
• Sibling build-pillar: per-state overlay configuration (#599 UPSTREAM canonical for TCPA + CAN-SPAM + GLBA + HIPAA per-state overlays)
• Fractional CMO with AI Swarm
• AI Readiness Assessment