Building a usable prospect list in Clay: found data, inference, message

Start from a segment you can write as criteria

Clay describes itself as a data enrichment and outreach platform that combines several data sources into prospect profiles. Its homepage lists account sourcing, lead scoring and CRM enrichment among the plays teams build on it. The practical starting point is therefore not a tool setting but a segment you can express as rules: industry, tech stack, location, headcount, a trigger such as hiring.

Pump shows what this looks like. Its case study describes codifying ideal-customer criteria around tech stack, matching companies to them, then enriching each account with stakeholder contacts so new reps received a ready territory list. Pump reports that ramp time fell from 90 to 14 days. Like every figure in this article, that is the vendor's and customer's published account, not an independent measurement.

Merge took a broader route: it enriched and categorised more than 50,000 Salesforce accounts, tagging industry, sub-industry and a suggested use case for each. That last field is a model's suggestion, not a looked-up fact, and the first sign that one table mixes different kinds of content.

1345

Enrichment: where credits and actions are spent

The pricing page separates two meters. Data credits are spent when you buy data from the marketplace, such as emails, phone numbers or company details. Actions measure orchestration: running an enrichment, calling an AI model, sending data to another system or exporting. If an enrichment returns nothing, neither is charged. With your own provider API keys, a run costs one action and no credits.

Limits shape table design. The Free plan allows 200 rows per table with 100 data credits and 500 actions a month; Launch allows 50,000 rows with 2,500 credits and 15,000 actions; Growth lists 6,000 credits and 40,000 actions. Actions reset every cycle, while credits roll over up to twice the monthly amount on Launch and Growth. The sources also differ on headline monthly prices, so confirm the current figure at purchase.

Editorial suggestion, not a vendor feature: put cheap filters first, run paid lookups on a small sample of rows, and add the heaviest AI columns last, so a mistake in segment definition costs a few rows rather than a monthly allowance. Token-heavy AI models are priced variably and show an estimate per row marked with a tilde, which is worth reading before a full run.

23

Found, inferred, drafted: three labels for every column

AlertMedia's case study makes the distinction visible. Its scoring depends on location counts, and Owen Chandler, the GTM engineer, built a workflow that waterfalls several enrichment providers for international locations. That is found data, with a provider behind it. Industry recategorisation, by contrast, uses Claygent, Clay's AI research agent, which is an inference you should treat as a classification to sample, not as a record.

The same company parses Gong call transcripts against the MEDDPICC framework and stores each extracted value together with supporting evidence from the call, then writes it to the Salesforce opportunity. Reps review and confirm rather than type. Keeping the evidence beside the value is the habit worth copying: an inferred field without its source cannot be audited. AlertMedia reports a 22% higher close rate for top-tier enterprise accounts, tied to its scoring setup.

A-LIGN adds a warning about found data itself. Starting from 11,000 target contacts, it compared provider locations with LinkedIn and reports that 1,500 of 6,500 existing records had the wrong location, about 23%. Provider output is not automatically true; it is only sourced.

67

CRM sync: plan, write-back rules and duplicates

On the pricing page, CRM auto-sync and enrichment first appear in Growth, alongside HTTP API integrations and webhooks; Enterprise adds data warehouse syncs and Clay API access. The product listing names Salesforce, HubSpot, Dynamics 365 CRM and Pipedrive among integrations, and CSV export is also noted. A team on Free or Launch should plan for manual export or import.

Duplicates are a design question, and two customers show different answers. Merge's event workflow picks up a name posted in Slack, enriches it, checks whether the person already exists in Salesforce and only creates a record if not. AlertMedia matches inbound leads to existing accounts so about 15% go straight to the assigned rep. Before switching on write-back, decide which fields Clay may overwrite and which it may only fill when empty; that rule is our recommendation, since the sources do not prescribe it.

2356

Write the message from found fields, hedge the inferences

Pump's messaging illustrates the safe pattern. Templates referenced provider-specific services, so an AWS user saw EC2 and RDS while a Google Cloud user saw Cloud Run, and hiring signals decided timing. The facts driving the copy were looked-up attributes. Pump reports a 25% better response rate from cloud-provider personalisation alone.

Merge's account agent drafts an email for the account owner when it detects a launch or partnership. At Lovable, recruiters review Clay's research and make every outreach decision themselves. In both, a person sits between inference and send. Our suggestion: let model-written text assert only found fields, and phrase anything inferred as a guess the sender can verify.

458

Checks before the table goes wide

Clay's trust page lists SOC 2 Type II, ISO 27001 and ISO 42001, and a data processing agreement is available on request by email. Its subprocessor list includes an LLM provider, which matters if prospect data goes into AI prompts. Data-source counts also vary across Clay's own pages (100+, 150+ and 200+), so judge coverage on your segment rather than the headline.

What the Free plan can settle: whether your criteria select the right accounts, whether found fields match a trusted source on a small enrichment sample, and whether a handful of AI answers read as sound. It cannot settle everything. Paid lookups draw on only 100 credits, so a full 200-row run may not fit, and write-back to the CRM needs Growth or a manual CSV import.

1239