Waterfall enrichment skill

Waterfall Enrichment

Overview

No single B2B data provider covers more than 60-75% of contacts. A multi-provider waterfall chains providers in sequence, each firing only when the previous returns empty. The result: 85-92% coverage at optimized cost.

This skill covers complete waterfall architecture: 3 separate waterfalls for company data, email, and phone — each with independently optimized provider ordering and verification integration.

Authoritative Foundations

• DAMA-DMBOK — Data quality dimensions — Score each provider on accuracy, completeness, validity, and timeliness, not coverage alone.
• Clay — Waterfall enrichment — Query providers in sequence and stop at the first validated result to control cost.
• Google and Yahoo — Bulk sender requirements (2024) — Validate every email the waterfall returns; bounces and complaints from bad data damage sender reputation.

When to Use

• "Build an enrichment waterfall"
• "Improve our data coverage"
• "Chain multiple enrichment providers"
• "Design a cost-optimized enrichment pipeline"
• "Set up waterfall enrichment in Clay"

Step-by-Step Process

Phase 1: Three Separate Waterfalls

Build independent waterfalls for different data types:

Company Waterfall:

1. Clay native / Clearbit (cheap, broad)
2. Apollo Company (mid-cost, good SMB coverage)
3. ZoomInfo (expensive, enterprise depth)
4. Claygent AI research (unstructured, last resort)

Email Waterfall:

1. LeadMagic Email Finder (verified results, pay-per-result)
2. Apollo (270M+ contacts, subscription)
3. Hunter.io (domain-pattern matching)
4. People Data Labs (alternative sourcing)
5. Claygent (AI web research)

Phone Waterfall:

1. Apollo (mobile, included in subscription)
2. Cognism (strong EU mobile coverage)
3. ContactOut (alternative sourcing)
4. People Data Labs (broad but thinner)

Phase 2: Provider Ordering

Sort by cost-per-hit, not cost-per-attempt. A cheap provider with 20% hit rate costs more per successful lookup than a moderate provider with 70% hit rate.

Formula: Cost-per-hit = Price per attempt ÷ Hit rate.

Phase 3: Conditional Fallback Logic

• Each fallback only fires when the previous step returns empty or error
• COALESCE logic: prefer high-confidence results even from later steps
• Credit caps: max 5-6 credits per row per waterfall
• Verification step after email waterfall — always

Phase 4: Provider Performance Monitoring

Track per provider: hit rate, cost per successful lookup, incremental coverage. Re-test provider ordering quarterly — provider performance changes over time. Drop providers contributing less than 5% incremental coverage.

Output Format

Waterfall architecture document with provider ordering, conditions, credit costs, expected coverage per stage, and verification integration.

Quality Check

Before delivering, verify:

• All required sections are complete
• Output matches the user's stated need
• Named frameworks are cited for key recommendations
• No vague claims — every recommendation has a specific action
• Deliverable is ready for operational use, not just conceptual

Common Pitfalls

1. One giant waterfall for everything. Different data types need different providers. Company data, email, and phone are separate waterfalls.

2. Wrong provider order. Sort by cost-per-hit, not coverage percentage. A cheap provider with low hit rate is expensive per result.

3. Skipping verification. Every waterfall returns 5-15% stale emails. Verification catches them before they damage sender reputation.

4. No credit caps. Without caps, a single stubborn row can consume 15+ credits across the full chain. Cap at 5-6 per row.

5. Not re-testing provider order. Provider performance changes quarterly. The waterfall that worked in January may be suboptimal by April.

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: 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

• lead-enrichment: Execute enrichment on a specific list
• clay-automation: Clay-specific waterfall configuration
• email-finding: Email-specific waterfall patterns
• contact-verification: Verification step after waterfall