Clay

Solution

Clay built Claygent with GPT-4 to turn web research and sales-enrichment work into an agentic workflow. Instead of forcing teams to juggle multiple data providers and manual research tasks, Claygent visits websites, finds relevant information, summarizes it, and feeds those insights into outreach workflows much like a sales researcher would. To keep the system efficient, Clay uses GPT-4 to reason about where on a site the needed information is likely to appear before scraping, then narrows the search space with a binary-search pattern instead of sending entire sites through the model. The broader Clay platform also assigns different models to different tasks: smaller, cheaper models support tools like AI Formula Generator for plain-English data transformations, while GPT-4 Turbo is reserved for more complex workflows. Cross-verification against multiple data sources and models helps assess trustworthiness so small GTM teams and “Claygencies” can operate at agency scale with only one or two people.

Data flow

Lead and website targets enter Claygent, which uses GPT-4 to decide what sections to scrape and how to summarize findings. Enriched data is then checked against other providers and models, and the results feed personalized outreach and GTM workflows inside the Clay platform.