From Requests to Decisions: What Koa Decisions Brings to Customers

When customers contact a business, they expect their requests to reach the right team without unnecessary delays. Whether they need help with an order, an account issue, or a service enquiry, every interaction involves a decision about what should happen next.

Salesforce has already introduced Koa, its CRM reasoning model developed with NVIDIA to help Agentforce handle complex, multi-step business tasks. Now, Salesforce is expanding the Koa family with Koa Decisions, a new model announced on November 6, 2026, that focuses on making quick, everyday decisions.

Rather than generating detailed responses or working through complex tasks, this new model is designed to make a specific choice in milliseconds. In this blog, we will look at what Koa Decisions does, how it works with Salesforce Agentforce, and what its introduction could mean for businesses and their customers.

What Is Koa Decisions and How Does It Work?

Koa Decisions is Salesforce’s next version of its HyperClassifier decision model. It reads a request, compares it with a list of categories defined by the business, and selects the most suitable option in milliseconds.

Unlike a general-purpose AI model that generates detailed responses, an AI decision model focuses on making a specific choice. It can return a category, a yes-or-no answer, or a score that another system can use to take the next step.

How Does Koa Decisions Make Decisions?

Think about a customer who contacts a company because they cannot access their account. The request could belong to the accounts, sales, or engineering team. Someone needs to identify the right destination before the issue can be handled.

Salesforce Koa Decisions can read the message and select the relevant category in a single step. The process works like this:

As Salesforce explains, the decision model returns a choice rather than generating a long response. This makes it useful for tasks where software needs a clear answer to continue working.

Why Does This Matter for Customer Requests?

A general-purpose language model can understand a customer message, but it may take longer to produce an answer when the task only requires selecting one option. A dedicated decision model is designed for that specific job.

Salesforce measured routing requests in Agentforce and found that its decision model took approximately 26 milliseconds, compared with around 1.4 seconds for a general-purpose LLM. That is roughly 98% lower response time in this particular comparison.

For customers, faster routing can mean less waiting before a request reaches the right team. It also helps businesses keep routine processes moving without using a larger model for every small decision.

What Makes Koa Decisions Different?

The new model builds on HyperClassifier and expands how decision models can be used across Salesforce. A few capabilities make this particularly relevant for businesses handling different types of customer requests.

1. Decisions in milliseconds

Koa Decisions handles single-step decisions in milliseconds. Instead of generating a detailed explanation, it identifies the required option and returns a result that other software can act on.

This is useful when a business receives a high volume of requests and needs to route them without adding unnecessary processing time.

2. Support for text, images, and audio

Customer requests do not always arrive as written messages. Someone might share an image of a damaged pfroduct or send an audio message explaining an issue.

Salesforce Koa Decisions adds multimodal support for images and audio, extending the types of inputs the model can handle. This gives businesses more flexibility when working with different customer interactions.

3. New categories without retraining

Businesses often introduce new services, departments, or request types. Traditionally, building a predictive machine learning model for a new category could involve collecting and labelling data, training the model, and checking its performance.

With Koa Decisions, teams can add a new choice to the available categories without going through that entire process. In Salesforce’s demonstration, the model selected a newly added Returns option with “no retraining.”

The demonstration also showed confidence scores of 95% and 99.5% for two example decisions. These are scores from the demo, not a general accuracy guarantee.

4. More ways to use decision models across Salesforce

Salesforce plans to extend the model beyond request routing. The announced areas include:

These are planned extensions, so their availability should be checked as Salesforce releases further updates.

How Koa Decisions Fits into Salesforce Agentforce

An AI agent does more than respond to questions. It also needs to decide which action to take, which team should handle a request, or what information should be shared next.

This is where an AI decision model can support Salesforce Agentforce. It can handle the smaller choices that keep an agent’s workflow moving, while a larger language model can focus on tasks that need reasoning, explanations, or content generation.

For example, when a sales deal starts showing signs of risk:

This approach gives each model a more focused role. As Salesforce puts it, the goal is “Less Reasoning. More Doing.” Routine decisions can happen quickly, while more complex work gets the attention it needs.

What This Means for Businesses and Their Customers

The value of Koa Decisions is not just about how quickly a model returns an answer. It is about what happens after that decision and how smoothly the customer journey continues.

For businesses, this can support:

For customers, these improvements can mean fewer unnecessary handoffs, quicker responses, and a smoother experience when interacting with a business.

Conclusion

As businesses handle more customer requests through AI-powered systems, getting each request to the right place becomes just as important as generating a response. Koa Decisions brings a faster way to handle these everyday choices.

For customers, the benefit comes down to a smoother experience. Requests can be directed more quickly, workflows can continue with fewer delays, and teams can spend more time resolving issues instead of managing routine decisions.