Intent Classification

Intent classification is a natural language processing (NLP) technique that identifies the underlying goal or purpose behind a piece of text or speech, then assigns it to one of a set of predefined categories. Instead of reading words literally, the model answers a more useful question: *what does this person actually want to do?* Common labels might be "request a refund," "check order status," or "escalate to a human."

How it works

A model is trained on many labeled examples so it learns the patterns, phrasing, and signals that map a message to an intent. Traditional approaches used rule-based keyword matching or statistical classifiers; modern systems use neural models and large language models (LLMs) that capture meaning rather than exact words. When new text arrives, the model returns the most likely intent (often with a confidence score), and downstream software decides what happens next—route a ticket, trigger an automation, or fetch the right knowledge.

In Salesforce / a GPTfy BYOM context

Inside Salesforce, intent classification is the routing brain that turns messy inbound text into the right action. With GPTfy's Bring Your Own Model (BYOM) approach, an admin can point their chosen LLM—Claude, GPT, or Gemini—at a Case description or inbound email *inside the org*, with PII masking applied before anything leaves Salesforce.

Concrete example: A new Case comes in reading "Charged twice this month, fix it now." GPTfy passes the masked text to the configured model, which classifies the intent as `billing_dispute` with high confidence. That label can then set a field, assign the Case to the billing queue, and prime a grounded reply using the customer's actual record. No Data Cloud required—the classification and the action both run on the Salesforce data already in the org.

This makes intent classification a foundation for record-level automation: it decides *which* workflow, prompt, or retrieval step fires next.

FAQ

Is intent classification the same as sentiment analysis? No. Sentiment analysis measures tone (positive, negative, neutral). Intent classification measures *purpose*—what the user wants to accomplish. They are often used together.

Do I need to train a custom model to use intent classification in Salesforce? Not necessarily. With a BYOM setup like GPTfy, a well-designed prompt plus a defined list of intent labels lets a general-purpose LLM classify reliably, so you can start without building a dedicated classifier.

How is intent classification different from an Agentforce agent? Intent classification is one building block—it labels the request. An agentic system may use that label to decide which action or tool to run. GPTfy lets you run this on your own model inside Salesforce as an Agentforce alternative.