How Forecast's AI Features Use Your Data

This article explains how Forecast's AI features use your data.

Jump to... 

• Which parts of Forecast use AI?
• Generative AI features
  • What happens to the prompts and responses I send to the generative AI features? Are they used to train models, retained, or seen by the model provider?
  • Do you enable Bedrock's request and response logging?
  • Do your own records of AI interactions contain my data, and are they used for training?
• Predictive models
  • Is my data used to train the predictive models?
  • Is there a shared model across customers, and is my data kept separate from other companies?
• What data the AI features use
  • Do the AI features use my email, calendar, or CRM data?
  • Do you use files I upload to train the models?
• De-identification and deletion
  • Is any de-identification or removal of personal information applied before my data is used by the models or written to logs?
  • When an account is deleted, does any retained data continue to feed the AI training process?

Which parts of Forecast use AI?

Forecast uses AI in two different ways, and they handle your data differently.

• Predictive models help you plan work. These include Task Overrun Alerts, Task Management Suggestions (suggested roles, assignees, and work estimates), Find Available Team Member, and Nova Insights. They are trained models that learn from your delivery data.
• Generative AI features are built on large language models hosted by Amazon Bedrock. These include Nova agents, the AI Report Builder, and the MCP server (both the in-app Ask AI experience and the bring-your-own-model connection). They generate text and take actions in response to your prompts.

The sections below cover each group, then the questions that apply to both.

Generative AI features

What happens to the prompts and responses I send to the generative AI features? Are they used to train models, retained, or seen by the model provider?

When you use Forecast's generative AI features, your prompts and the model's responses are processed by foundation models hosted on Amazon Bedrock within AWS.

They are not used to train or fine-tune any model, are not retained after your request has been processed, and are not shared with or visible to the model provider.

Bedrock runs each provider's model in an isolated deployment that the provider cannot access, and our Bedrock configuration is set to zero data retention.

Note: AWS explains its data handling at Learn how Amazon Bedrock uses model input and output data.

Do you enable Bedrock's request and response logging?

We do not currently enable Amazon Bedrock's built-in request and response logging.

Do your own records of AI interactions contain my data, and are they used for training?

Yes, our own monitoring and diagnostic records contain your data. To operate the generative features reliably, we keep traces of AI interactions in our own AWS environment. These traces include identifiers such as your company name, company ID, and account ID, along with the content of your interactions with the AI features.

We use these records to monitor performance, diagnose problems, and improve the reliability of the features. They are not currently used to train, tune, or evaluate any model.

Note: This is separate from Bedrock. Bedrock does not retain your data for these purposes; these are our own operational records, held in our environment and under our control.

Predictive models

Is my data used to train the predictive models?

Yes. Each model is built in two layers.

1. A base model is trained on aggregated, historical delivery data so it understands general patterns, such as how long a type of task usually takes.
2. That base model is then refined on your own company's data so its predictions reflect how your teams actually work.

The models learn from your operational data, such as your tasks, time registrations, roles, skills, and resourcing information. They are refreshed on a regular cadence, currently around every two weeks, so they stay current as your data changes.

Is there a shared model across customers, and is my data kept separate from other companies?

The base model is trained on aggregated historical data drawn from across Forecast customers, so every customer benefits from broad patterns rather than starting from nothing. This aggregated data is used to learn general patterns, not to expose any individual customer's records.

The layer that is specific to you, the refinement trained on your company's data, is tied to your account. It is not shared with, exposed to, or reused for any other customer, and no other customer's refined model is ever applied to your data.

Note: Forecast's models are also kept entirely separate from any other product we operate. Your data is not pooled or combined across products.

What data the AI features use

Do the AI features use my email, calendar, or CRM data?

The predictive models use only the operational data you manage in Forecast, such as task details, time registrations, project information, roles, skills, teams, and departments. They do not read your email, your calendar, or any CRM or sales data.

Do you use files I upload to train the models?

No. We are building some agentic features that use data from attached files as part of a workflow, but that data isn't used to train our models.

De-identification and deletion

Is any de-identification or removal of personal information applied before my data is used by the models or written to logs?

No PII is used in our training data.

When an account is deleted, does any retained data continue to feed the AI training process?

Your company-specific model layer is tied to your account, and Forecast's models are not combined with any other product.

Keep exploring

• Forecast MCP Server — An introduction to Forecast MCP and what it can do for you. 
• Forecast AI: Getting Started — Get started with Ask AI directly within Forecast, or connect your preferred AI assistant (such as Claude or ChatGPT) and make your first request. You can also learn more about the safeguards in place.
• Forecast AI: Capabilities and Prompt Examples — Explore the full list of questions you can ask and actions you can take, with example prompts.