Clay automation skill

Clay Automation

Overview

Clay is the orchestration layer where enrichment, scoring, and routing converge. Used correctly, it's a GTM multiplier. Used incorrectly, it's a credit-burning machine producing data nobody trusts.

This skill covers when and how to roll out Clay automation — data quality, table separation, rollout phases, and n8n handoff.

Playbook index: references/automation-playbook-index.md — all 38 automation + tool + gtm-ops playbooks.

Automation strategy (before tool config): Jen Igartua (Go Nimbly) — data before AI, maturity levels 0–4, human+machine division. Canonical → references/gtm-automation-expert-playbook.md (Pattern 30). Outbound copy/infra → Eric Nowoslawski / Pat Spielmann / Justin Michael — not this skill.

Tool implementation (table columns, LeadMagic waterfalls, loops) lives in tools/:

• clay-toolkit (tools/clay-toolkit) — tables and waterfalls
• clay-loops-toolkit (tools/clay-loops-toolkit) — signal loops

When to Use

• "Build a Clay enrichment workflow"
• "Set up a Clay table for our prospecting"
• "Configure a waterfall in Clay"
• "Use Claygent for research"
• "Push enriched data from Clay to HubSpot"
• "Optimize our Clay credit usage"
• "Design a Clay automation pipeline"

Authoritative Foundations

Clay workflow design follows patterns from DAMA-DMBOK data-quality dimensions, Ziellab (3 separate waterfalls: company, email, phone), and GTME Pulse (10 production templates tested at $5M-$100M ARR companies).

The core principle: Clay is a routing engine, not a CRM. Enriched data lives in your CRM; Clay processes it en route.

• Eric Nowoslawski — Crawl Walk Run (Growth Engine X). Roll out Clay automation in phases: Crawl — manual campaigns for 5–10 companies (no AI); Walk — system prompt with business context + manual examples, review first 50 outputs; Run — automate in Clay → Smartlead/SEP with Supabase for block lists. Default campaign type: Creative Ideas (3 constrained capabilities per prospect). Playbook → references/eric-nowoslawski-outbound.md.

Prerequisites

• Clay account (Pro plan or above for waterfall features and conditional columns)
• API keys for enrichment providers (Apollo, ZoomInfo, PDL, etc.)
• CRM connected (HubSpot, Salesforce, or Attio)
• ICP defined and documented

Step-by-Step Process

Phase 1: Table Architecture

Rule: separate company and person tables.

Company table: one row per domain. Enriches firmographics, tech stack, company-level qualification once.

Person table: one row per contact. References company data via domain lookup. Enriches email, phone, LinkedIn, role-level qualification.

This separation prevents credit waste from re-enriching the same company data across every contact row.

Phase 2: Waterfall Configuration

For each data field, configure a conditional waterfall:

Field	Primary	Fallback 1	Fallback 2
Email	LeadMagic Email Finder	Apollo	Hunter
Phone	Apollo	Cognism	ContactOut
Company data	Clay native	Apollo Company	Clearbit

Set conditions: each fallback only fires when the previous step returns empty or error. Use Clay's conditional logic: "Only run if [previous column] is blank."

Phase 3: Claygent Configuration

Claygent is AI-powered web research. Configure prompts explicitly:

Good prompt: "Find the work email for [name] at [company]. Search the company's team page, LinkedIn profile, and press releases. Return the email AND the source URL. Do NOT guess or construct emails from patterns. If no verified source, return empty."

Bad prompt (don't use): "Find me their email."

Critical rules:

• Always require source URL citation
• Explicitly prohibit pattern-guessing
• Set credit cap per Claygent call (5-10 credits)
• Use only for the 10-15% that structured providers miss

Phase 4: CRM Push

Push enriched data to CRM with a clay_status property:

clay_status	Meaning	Action
pending	In enrichment	Hold — not ready
enriched	Enrichment complete	Ready for verification
verified	Verified and safe	Can enter sequences
exported	Pushed to CRM	Done, archive in Clay

Only contacts with clay_status = verified enter sequences.

Phase 5: Credit Optimization

1. Qualify first, enrich deep later. Run ICP filters before expensive contact enrichment. Cuts costs 30-40%.

2. Credit caps per row. Set max 5-6 credits per row. If a contact is that hard to find, they're probably not a good fit.

3. Native integrations over HTTP API. Clay's native provider integrations are rate-limited and credit-billed correctly. HTTP API calls bypass Clay's cache and often double-charge.

4. Batch overnight. Claygent is 15-40s per row. Run large Claygent batches during off-hours.

Output Format

Clay workflow document with table architecture diagram, provider waterfall configuration, Claygent prompt templates, CRM push rules, credit budget, and maintenance schedule.

Quality Check

• Company and person tables separated
• Waterfalls configured with conditional fallback logic
• Claygent prompts explicitly prohibit guessing
• CRM push uses clay_status property gating
• Credit caps set per row
• Test batch (50 rows) validated before scaling

Common Pitfalls

1. One giant table. Combining company and person data wastes credits and makes re-enrichment impossible. Separate always.

2. Enriching before filtering. Running $0.15-0.40/contact enrichment on non-ICP records wastes budget. Filter on cheap data first.

3. Claygent guessing emails. Without explicit "do not guess" instructions, Claygent constructs pattern-based emails that bounce at 40-60% rates.

4. Two-way CRM sync. Clay-to-CRM should be one direction. CRM-to-Clay sync creates data conflicts. Push only.

5. Clay as permanent storage. Clay is a workspace. Push to CRM, archive or delete rows in Clay. Old rows decay just like anywhere else.

6. No credit caps. Without caps, a single row can chew through 15+ credits. Cap at 5-6 per row.

Execution Artifacts

• references/framework-notes.md — named frameworks, citation anchors, and operating assumptions
• templates/output-template.md — copy-paste deliverable structure for the user
• scripts/check-output.py — local checklist validator for required sections This skill includes lightweight artifacts the agent can load on demand:
• references/gtm-automation-expert-playbook.md — Jen Igartua RevOps automation strategy (Pattern 30)
• references/eric-nowoslawski-outbound.md — Crawl Walk Run, Creative Ideas, GEX stack (Eric Nowoslawski) Use the artifacts when the user asks for an implementation-ready deliverable, a repeatable workflow, or a quality check rather than generic advice.

Related Skills

• waterfall-enrichment: Deep waterfall architecture
• lead-enrichment: Enrichment execution patterns
• list-building: List building workflows in Clay
• clay-toolkit: tools/clay-toolkit — table blueprints, LeadMagic waterfalls
• clay-loops-toolkit: tools/clay-loops-toolkit — signal loops and routing
• ai-prompts-toolkit: Claygent and LLM prompt library
• n8n-automation: n8n as Clay export for complex cases
• crm-integration: CRM configuration for Clay data