Salesforce has introduced Koa Decisions, the next evolution of its HyperClassifier model, designed to enhance decision-making across its platforms. Built on Google’s Gemma model, Koa Decisions offers multimodal support for images and audio, enabling faster and more efficient workflows. The model reads a request and a predefined list of categories, then makes a decision in milliseconds, significantly reducing the need for larger, more expensive AI models for routine tasks.
Decision models like Koa Decisions are optimized for tasks such as routing customer inquiries, assigning scores, or selecting workflows. For instance, when a customer asks about a lost package, the model quickly directs the request to the appropriate team. Salesforce’s data science team found that decision models process requests in about 26 milliseconds, compared to 1.4 seconds for general-purpose LLMs, making them ideal for enterprise applications.
Koa Decisions is part of the broader Koa family of models, which also includes a reasoning model built with NVIDIA. The new model will be integrated into various Salesforce applications, including flows, Apex, and prompt templates, as well as an API for developers. This expansion allows for more versatile decision-making across different workflows and applications.
Customers can test Koa Decisions through a free Trailhead account, where they can experiment with routing support requests and adding custom categories. The model’s efficiency is highlighted by its ability to classify requests without extensive retraining, reducing the time from idea to production from months to hours. This makes it a valuable tool for businesses looking to streamline their processes and improve operational efficiency.