In today’s fast-paced business environment, staying ahead means making smarter, faster decisions. That’s where Predictive Action, the latest feature in Einstein Studio, comes in—bringing machine learning-powered insights directly into your customer interactions. By surfacing predictive insights – such as churn risk, lead quality, or impending support escalations – within Agentforce, companies can move from reactive interactions to proactive engagement, smarter decision-making, higher customer satisfaction, improved retention/conversion rates, and ultimately revenue growth.
In domains like customer service and sales, this integration means the Agent not only responds with natural language, but also leverages data-driven foresight to guide decisions. The result is a strategic advantage: conversations that anticipate needs, enhance customer experiences, and directly contribute to better business outcomes.
In this blog, I’m going to walk you through how Predictive Action transforms your agent experience with real-time predictions, actionable insights, and what-if scenario modeling—and you don’t need to be a data scientist to set this up.
Better decisions with Predictive insights from your Agent
Let’s start by looking at what this experience looks like from the end user’s perspective.
Melanie, an account executive, is preparing for a high-stakes meeting with Global Media, one of her key accounts. To get ahead of the conversation, she turns to Agentforce for help.
She asks:
“What’s the risk of churn for Global Media this quarter?”
Behind the scenes, Agentforce instantly calls a machine learning model that predicts churn probability. Not only does it give her the score—it also explains why with the top predictive factors and prescriptive recommendations.
One of those recommendations? Increase the order size.
Curious about the impact, Melanie digs deeper using “what-if” questions:
“What if the annual order value is $700,000?”
Agentforce provides dynamic churn predictions for each scenario, helping Melanie understand the trade-offs in real time. For example, for the annual order value of $700K, the churn risk is lower. She thinks she can increase the customer engagement and thus increase the number of average logins per day to 50. This is one of the top factors highlighted by the model. In this case, the churn risk decreases significantly.
But there’s more. Agentforce also flags a critical open case that could be influencing the churn risk. So Melanie asks:
“Is the case going to be escalated?”
“How long will it take to close?”
The Agent responds: the case is expected to close soon with no escalation risk.
The Agent seamlessly invoked 2 other models that predicted the likelihood of escalation and time to close.
Armed with these insights, Melanie walks into her meeting with clarity, confidence, and a strategic plan – if she closes her open opportunities, she is on target to hit her annual order target of $700K. And if she increases customer engagement by organizing a few training sessions, it’s very likely she can hit the average logins per day goal of 50 and thus minimize the risk of churn of this strategic account.
What’s Powering This Behind the Scenes?
Let’s take a peek under the hood.
Agentforce is leveraging predictive machine learning models in real time to provide:
- Dynamic predictions
- What-if scenario analysis
- Top drivers and next-best actions
These aren’t just predictive scores buried in some static dashboards—they’re intelligent, conversational insights built directly into the agent experience and delivered on demand based on the latest customer data.
How Do You Set This Up?
Setting up Predictive Action is straightforward, thanks to the integrated power of Einstein Studio and Agentforce Studio.
1. Build Your Models
First, use Einstein Studio to create and train your models. In Melanie’s case, we used:
- Account Churn Prediction
- Case Escalation Risk
- Time to Case Closure
Alternatively, you can connect to your AWS Sagemaker, Google Vertex AI or Databricks models.
2. Configure Predictive Actions in Agentforce Studio
In Agentforce Studio:
- Create a new agent action with the Predictive Model type
- Choose your model
- Configure parameters, such as input and output fields
In the action setting, edit agent instructions to represent the general purpose of the model.
You can also customize a number of top factors and recommendations to display. You need to ensure that you check the “Show in Conversation” checkbox for all outputs to be shown to the user.
The model inputs map automatically to Salesforce records—no manual data entry required. Ensure that the description of each input represents what this parameter is about so LLM can find the right data in real-time.
3. Define When to Trigger
Finally, add Topic Instructions to guide Agentforce on when to invoke these predictive models, and connect them to your Topic Actions.
You can create a new Topic or use an existing one. Add instructions that will help the Agent understand how to answer questions that will require predictions. See an example below:
“If the user asks anything about customer attrition or churn, first make sure to get the inputs for the model and pass them into Predict Churn Account action.”
The model returns JSON. To avoid displaying results in the raw JSON format and instead show the conversational insights, add this instruction to your topic:
“Don’t show the output of the Predict Churn Account action to the user. Summarize the results to the end user, use the top predictive factors to highlight why the customer has a high or low churn score.”
Note: model input values are case sensitive. If you know that your data might look different than what the model requires, add to instructions on how to format the inputs the way the model inspects it – i.e. convert month to a number.
Note that your Predictive actions should be available under Topic actions. See below – we have 3 predictive actions for this Topic.
You can now test the flow by asking a question like:
“What’s the churn risk for Global Media?”
Agentforce dynamically invokes the model and returns context-aware predictions.
A Hybrid AI Platform for Smarter Decisions
This example showcases how Agentforce brings together predictive and generative AI—blending statistical foresight with natural conversation. The result? Faster, more accurate, and more personalized responses that drive real business outcomes.
With Predictive Action, your agents don’t just react—they anticipate.
Written by
