clay-mastery

Clay Mastery

Clay stays your orchestration canvas; this skill re-expresses a standard enrichment table and its waterfalls as ColdIQ marketplace endpoint calls, so the same recipe runs with one ColdIQ key whether you build it in Clay or call the API directly. Clay-specific UI steps (run conditions, formula writer, Claygent) are kept as guidance.

ColdIQ Marketplace Endpoints

Task	Method	Path	Credits	Endpoint ID	Notes
Company enrichment	POST	/v1/limadata/enrich/company	1	limadata.enrich.company	Industry, headcount, revenue, funding, tech
Email — Prospeo (step 1)	POST	/v1/prospeo/enrich-person	unknown	prospeo.enrich_person	First in waterfall
Email — work (step 2)	POST	/v1/limadata/find/work-email	1	limadata.find.work_email	Name + domain
Email — Findymail (step 3)	POST	/v1/findymail/search/business-profile	1	findymail.search.business_profile	LinkedIn URL
Email — FullEnrich (step 4)	POST	/v1/fullenrich/contact/enrich/bulk	per result	fullenrich.contact.enrich_bulk	Nuclear, async
Email verification	POST	/v1/findymail/verify	1	findymail.verify	Replaces Debounce
Contact discovery (employees)	POST	/v1/limadata/prospect/employees	25	limadata.prospect.employees	Find people at a company
Tech stack	POST	/v1/builtwith/domain	flat	builtwith.domain	For tech-stack columns
News / signal	POST	/v1/serper/news	unknown	serper.news	Cheap news lookups

Standard Table Layout (9 phases, left → right)

1. Input — Company Name, Domain, LinkedIn URL
2. Company Enrichment — Industry, Employee Count, Revenue, Funding, Tech Stack → POST /v1/limadata/enrich/company · 1 cr · limadata.enrich.company
3. Signal Detection — Hiring, News, LinkedIn Ads, Growth (see signal-detection)
4. Contact Discovery — Full Name, Title, LinkedIn URL → POST /v1/limadata/prospect/employees · 25 cr · limadata.prospect.employees
5. Email Waterfall — see below
6. Email Verification → POST /v1/findymail/verify · 1 cr · findymail.verify
7. AI Qualification — AI Company Tier, AI Contact Tier, AI ICP Score (Claygent)
8. AI Personalization — AI Custom Signal, AI Opening Line, AI Value Prop (Claygent)
9. Output — Final Email, Sequence Tag, Campaign Assignment

Naming: prefix sources (LI - Employee Count, Apollo - Revenue), prefix AI columns (AI - Opening Line), Title Case, column order = execution order.

Email Waterfall (cheapest-first, "only run when previous is empty")

1. → POST /v1/prospeo/enrich-person · ? cr · prospeo.enrich_person (unverified) — always first
2. → POST /v1/limadata/find/work-email · 1 cr · limadata.find.work_email — only if step 1 empty
3. → POST /v1/findymail/search/business-profile · 1 cr · findymail.search.business_profile — only if above empty
4. → POST /v1/fullenrich/contact/enrich/bulk · per result · fullenrich.contact.enrich_bulk — nuclear, async
5. → POST /v1/findymail/verify · 1 cr · findymail.verify — on ALL found emails

In Clay, set each column's run condition to "Only run when [Previous Email Column] is empty". Final-email formula (pick first non-empty):

{{Prospeo - Work Email}} || {{ColdIQ - Work Email}} || {{Findymail - Work Email}} || {{FullEnrich - Work Email}} || ""

Copy-paste Clayscript formulas

// First name cleaning
{{First Name}}?.replace(/[^a-zA-Z- ]/g, "").trim().split(' ').map(w => w.charAt(0).toUpperCase()+w.slice(1).toLowerCase()).join(' ').replace(/\s.*/,'')

// Domain from URL
{{Website}}?.replace(/^https?:\/\/(www\.)?/, '').replace(/\/.*/, '').toLowerCase()

// Title seniority check
/^(C[A-Z]O|Chief|VP|Vice President|Director|Head of|SVP|EVP|President|Founder|Co-Founder|Partner|Owner|Managing Director)/i.test({{Title}} || '')

// Employee count bucket
(n => n <= 10 ? '1-10' : n <= 50 ? '11-50' : n <= 200 ? '51-200' : n <= 500 ? '201-500' : n <= 1000 ? '501-1000' : '1000+')({{Employee Count}} || 0)

Claygent prompts that work

• Company tier: "Assign Tier 1/2/3 … Return ONLY 'Tier 1/2/3' + one-sentence reason."
• Opening line: "Write a cold email opening (1 sentence, <20 words). Reference {{Signal}}. Sound human. Don't pitch. Don't start with 'I noticed'. Make it about THEM."

Default Claygent to GPT-4o-mini; escalate only when it fails.

Best practices & debugging

Pre-filter before enrichment; cheapest tool first; test 50–100 rows; cache; column order = execution order. When enrichments fail, check: input populated? input format (LinkedIn full path, domain no protocol)? API key active? run condition correct? provider has data → move to next in waterfall. See resources/credit-optimization.md.