Clay is the table-shaped enrichment workspace where AI agents run prospecting steps as columns: find the company, find the contact, classify the buying signal, draft the opener. Claygent is the research agent inside it. The workflow that holds up: the agent enriches and writes, the SDR validates the top of the list, and the enrichment trail persists past the moment the CRM row gets updated.
The Clay enrichment workflow, step by step
1. Source the list. Pull from LinkedIn Sales Navigator, Apollo, or ZoomInfo into a Clay table. Use waterfall enrichment: Apollo first for email, ZoomInfo for mobile. Each column records which provider answered, so you know whose data you are trusting.
2. Add the signal column. Job-change triggers, funding, hiring spikes, tech-stack changes. Clay integrates Common Room and 6sense for intent; a Claygent column can browse the company blog and pull the most recent product announcement.
3. Score against ICP. A Claygent or ChatGPT column reads the row against your ICP definition and returns a tier plus a one-sentence reason. The reason is what matters; without it, the score is a number nobody trusts.
4. Draft the opener. Another AI column takes the signal, the ICP reason, and the contact's recent LinkedIn post, and writes two sentences. Apollo, Outreach, Smartlead, or Instantly takes it from there.
5. Push to the CRM. Sync validated rows to Salesforce or HubSpot. The fields that go: account, contact, stage, owner. The fields dropped on the floor: the signal that triggered the row, the Claygent search trail, the rejected drafts, the SDR's edit.
Worked example: an SDR working a job-change list
Maya runs a Clay table of newly promoted VPs of RevOps. A Claygent column visits the new employer's careers page, counts open RevOps roles, and reads the last earnings transcript for any quoted RevOps priority. ChatGPT writes a three-line opener tying the promotion to one specific build-out. Maya reviews thirty rows, kills four, edits six, and sends the rest through Smartlead. Reply rate clears 9%. Two weeks later her AE asks why a specific account was prioritized and the rationale is gone.
Where the enrichment trail dies
The CRM stores account, contact, and stage. The sales-engagement tool stores the sequence step and the reply. Neither stores why the prospect was on the list, which signal fired, which Claygent search produced the rationale, or which version of the opener the rep actually sent. Six weeks later the AE running discovery cannot reconstruct the reason the prospect agreed. The opening question lands flat.
One way to solve this is a workspace like Dock that holds the research brief, the triggering signal, the Claygent trail, and the personalization rationale next to a salesforce_account_id pointer back to the CRM. Salesforce and Outreach stay the system of record. Dock holds the agent-interpreted context so the next rep picks up the thread. Pair it with a clear agent identity for enrichment runs.
Why this matters
Clay made enrichment cheap. The new bottleneck is not finding the data, it is keeping the reason for the outreach attached to the account after the row leaves Clay. Teams that hold both ship clean lists and clean CRM updates; teams that only hold the row ship pollution.
Start with the pillar on AI sales prospecting that compounds.
FAQ
Is Claygent the same as ChatGPT in a Clay column? No. Claygent is Clay's research agent that browses public pages and navigates gated forms. A ChatGPT or Claude column reads row context and writes. Production tables use both.
Does Clay replace Apollo or ZoomInfo? No. Clay sits on top, calling Apollo, ZoomInfo, and 50+ providers as waterfall steps. You still pay them; you orchestrate in one table.
How do I keep enrichment from polluting the CRM? Validate before sync. Treat Clay as staging. Push only rows a human reviewed, and store rationale somewhere durable, not a CRM note field nobody reads.
What is signal-based selling? Outreach triggered by real-time buyer behavior (job change, funding, hiring, product use) rather than a quarterly list refresh. Common Room and 6sense are common signal sources.