Case Study: Checkr achieves 86% Salesforce answer accuracy with Credal's MCP

Checkr boosts AI-generated Salesforce answer accuracy from 13% to 86% with Credal.ai

Checkr, a trust and safety platform for background checks, faced a significant challenge when providing its revenue team with AI access to Salesforce data. Direct access to raw data resulted in AI models generating answers that sounded confident but were only 13% accurate, a problem compounded by Checkr's use of three different Salesforce instances with conflicting data definitions. They partnered with the vendor Credal to solve this problem using Credal's MCP (Model Context Protocol) product.

Credal's solution involved creating three separate, governed MCP servers, one for each Salesforce instance, which were grounded in specific business context and documentation. This approach prevented the AI from guessing by asking clarifying questions for ambiguous queries. The result was a dramatic jump in data accuracy from 13% to 86%, allowing Checkr's team to reliably query Salesforce in natural language across all instances within tools like Claude and Cursor.

Checkr