ASIntegration · AI and Machine Learning

Connect AWS SageMaker model predictions to Salesforce Agentforce

AWS SageMaker is the machine learning platform that data science and engineering teams use to build, train, and deploy predictive models at scale on AWS. Salesforce Agentforce is the AI agent platform for commercial and service workflows. For organisations with investment in both AWS and Salesforce, the opportunity is to make the predictive models deployed in SageMaker available to the commercial team without the data science team having to build a bespoke integration for each use case. A churn prediction model in SageMaker should be invocable by an Agentforce agent reviewing account health. A lead scoring model in SageMaker should inform how Agentforce routes and prioritises opportunities. When Emerge Digital connects AWS SageMaker to Agentforce, model endpoints deployed in SageMaker are available as tools in Agentforce's reasoning and action workflow — so data science predictions drive commercial actions rather than sitting in notebooks and dashboards that the commercial team does not read.

All integrations

What this unlocks

SageMaker model endpoint invocation during live commercial conversations: Agentforce agents can call SageMaker real-time inference endpoints as a tool during account reviews, lead qualification, and service interactions — the model receives the Salesforce record context as input and returns a prediction or classification that informs the agent's next action.
Churn risk predictions surface in Salesforce account health workflows: a churn risk model deployed on SageMaker can be invoked by Agentforce during a periodic account health review — high-risk accounts receive Salesforce tasks for the customer success team, with the prediction score and contributing features as context, before the customer signals churn intent.
Lead scoring and intent classification from SageMaker models: a lead scoring model in SageMaker can be invoked when a new lead is created in Salesforce — the model scores the lead based on firmographic and behavioural features, and Agentforce routes the lead to the appropriate sales queue, velocity, and outreach sequence based on the score.
Batch prediction results imported to Salesforce fields: SageMaker Batch Transform can run predictions across large datasets — Agentforce can coordinate the batch job trigger, read the results, and write the prediction scores back to the relevant Salesforce account, contact, or opportunity fields, so the predictions are visible in the commercial record rather than in an S3 bucket.

In the customer journey

Churn risk model invoked during quarterly account health review

An Agentforce agent runs a quarterly account health review across the enterprise customer portfolio. For each account, the agent invokes the SageMaker churn risk endpoint with the account's engagement metrics — login frequency, support ticket rate, feature adoption, and contract renewal window. The model returns a probability score and the top contributing features. Accounts above 70% churn risk receive a Salesforce task for the customer success manager with the prediction context. Seventeen accounts are flagged for intervention — only three would have been identified by the manual account review process.

Inbound lead scored by SageMaker model and routed to the appropriate queue

A new lead is created in Salesforce from an inbound form submission. The Agentforce agent invokes the SageMaker lead scoring endpoint with the lead's firmographic data — industry, company size, technology stack, and the content they downloaded. The model returns a score of 82 out of 100. The agent routes the lead to the enterprise sales queue, sets the follow-up SLA to four hours, and flags it for the next available enterprise AE. The lead routing takes seconds rather than waiting for a daily batch scoring process.

Batch product recommendation predictions written to Salesforce account fields

An Agentforce agent triggers a SageMaker Batch Transform job to generate product expansion recommendations for all active accounts in the mid-market tier. The job runs overnight. The next morning, the agent reads the batch output from S3 and writes the top recommended product expansion to each Salesforce account's next-best-product field. The commercial team's account review the following week shows which products each account is most likely to buy next — computed from the actual usage and engagement signals.

Why not using SageMaker and Salesforce independently?

Data science teams deploy models in SageMaker; commercial teams work in Salesforce. The typical gap is that SageMaker predictions live in dashboards or S3 exports that the commercial team does not access in their daily workflow. The commercial team makes decisions without the model's input; the data science team has no feedback loop from commercial actions. Emerge Digital builds the tool-call and batch coordination layer that makes SageMaker predictions available to Agentforce agents as a first-class input to commercial decisions — so the data science investment drives commercial outcomes rather than informing a dashboard nobody reads.

AWS SageMaker's Agentforce integration is relevant wherever predictive model outputs should inform commercial decisions — lead qualification, account health, renewal risk, product expansion, and service classification. It is most valuable for organisations with data science teams that have already built and deployed models in SageMaker, where the gap is the last-mile commercial integration rather than the model development.

How Emerge integrates AWS SageMaker

Emerge Digital designs the SageMaker + Agentforce integration as a consulting engagement. We assess which deployed SageMaker endpoints or batch jobs are relevant to the commercial workflows in scope, design the tool-call layer that gives Agentforce agents access to model predictions with the correct Salesforce context as input, build the batch prediction import workflow for writing scores back to Salesforce fields, and test the end-to-end prediction and action workflows. The integration is designed in collaboration with your data science team to ensure the model input schema and output interpretation are correct.

How we structure an engagement

Related integrations

FAQ

Can Agentforce agents trigger SageMaker training jobs or re-train models?

No. Model training and retraining stay with the data science team's ML pipeline. Agentforce agents invoke inference endpoints and batch transform jobs; they do not trigger model training or modify model configurations.

We use Databricks ML or Azure ML rather than SageMaker — can the same integration pattern work?

Yes. The tool-call pattern — an Agentforce agent invoking an external ML model endpoint — works with model serving platforms other than SageMaker. Emerge builds to the ML serving platform your data science team uses. The integration design is adapted to the serving platform's API and authentication model.

How do we handle model latency for real-time endpoint calls within a live conversation?

SageMaker real-time inference endpoints typically respond in milliseconds to low hundreds of milliseconds for most model types. The integration is designed with latency expectations and timeout handling built in — where a model response takes too long, the agent falls back to a default routing decision rather than blocking the conversation. Latency requirements are validated during the engagement design.

How long does an AWS SageMaker + Agentforce integration take?

A focused engagement typically runs six to ten weeks: scoping which model endpoints and batch jobs are in scope, designing the tool-call layer with the correct input schema and Salesforce context, building real-time and batch integration workflows, validating prediction accuracy and latency in the Agentforce context, and testing end-to-end commercial decision scenarios. Engagements with multiple model endpoints may require additional time for model-by-model validation.

Ground your agents in AWS SageMaker.

Tell us what your agents need to read and write in AWS SageMaker, and we'll design the integration and the governance around it.

Talk to the practice

Prefer email? Write to the practice instead.