Achieving 10x growth with agentic sales prospecting

Clay built Claygent, an AI web scraper that uses GPT‑4 to understand and extract highly specific information from websites. To make Claygent as efficient as possible, Clay optimizes the number of tokens passed to GPT‑4 and chooses the appropriate model for each use case. 

When scraping a website, it would be inefficient to send the whole site to GPT‑4. Instead, Claygent asks GPT‑4 which section of the website is most likely to contain the desired information. For example, GPT‑4 might indicate that SOC-2 compliance information is generally found in the footer. Claygent can then specifically scrape the footer instead of the whole website. 

Claygent also uses a binary search approach, where it takes part of a website, checks if the required data is there, and if not, moves to another part. This approach progressively narrows the search space until the required information is found. 

Clay selects the most appropriate AI model for each product task to optimize cost. For instance, their AI Formula Generator tool, which allows non-technical users to transform data using plain English instructions and examples, runs on cheaper, smaller versions of models. Clay then reserves the speed and intelligence of GPT‑4 Turbo for other, more complicated tasks.

In the future, Clay plans to fine-tune their AI agent for specific verticals, such as finding local fitness gyms, doctors, or nurses. They want Claygent to handle complex requests like “find me all companies that have hired 10 engineers in the last month.” Beyond that, Clay is exploring ways to use signals and triggers to make Claygent more proactive, like detecting when a customer visits a website and generating personalized outreach messages or alerting sales teams to potential opportunities in real time.