If you haven’t already, check out Arun Ulag’s hero blog “FabCon and SQLCon 2026: Unifying databases and Fabric on a single, complete platform” for a complete look at all of our FabCon and SQLCon announcements across both Fabric and our database offerings.
AI teams need to move quickly, but the reality is that most of the data they need simply isn’t ready for AI. In fact, Gartner predicts that through 2026, 60% of AI projects will be abandoned because they lack AI-ready data.
A large portion of enterprise information exists in unstructured text. This includes support tickets, contracts, call transcripts, documentation, and internal communications. These sources often contain information that can help train or evaluate AI systems. However, they also include sensitive data such as personal identifiers, financial details, or confidential business information. Because of this, access to these datasets is often restricted.
Tonic Textual (Generally Available) as a workload in the Microsoft Fabric Workload Hub. The workload helps teams detect sensitive information in text and prepare datasets that can be used in AI development workflows inside Fabric. For the official announcement, read the press release - Tonic.ai Announces General Availability of Tonic Textual for Microsoft Fabric.
Unlocking unstructured data for AI development
Many organizations already manage structured data for analytics in Fabric. Unstructured text can also provide useful information for AI systems.For example, in healthcare, information such as clinical notes, discharge summaries, physician documentation, and patient communications often contain context that structured datasets do not capture. These sources can support use cases such as knowledge retrieval, clinical documentation analysis, or internal search tools. However, they may include protected health information, which limits how the data can be used in development environments. Preparing this data before it is used in AI systems allows organizations to reduce privacy risk while still using the information contained in these documents.
Tonic Textual workload provides tools to identify and transform sensitive information so that unstructured datasets can be used more safely in AI workflows within Fabric.
Preparing unstructured data inside Fabric
Tonic Textual runs as a workload within Microsoft Fabric. It scans unstructured files stored in Microsoft OneLake and detects sensitive entities such as names, identifiers, or financial information.Organizations can configure how these entities are handled, options include:
- Redacting sensitive values
- Masking specific identifiers
- Replacing values with synthetic data
- Applying custom rules for specific entity types
Running these steps inside Fabric allows teams to prepare text data without moving it outside the platform.
Preparing text for AI workflows
When sensitive entities are transformed, it is important that the text remains usable for downstream tasks. Tonic Textual preserves document structure and surrounding context when transformations are applied.For example:
- Support conversations can keep their dialogue format.
- Contracts can keep their structure and clauses.
- Documentation can retain technical context.
Applying the same transformation rules across datasets also helps organizations use consistent privacy policies across development workflows.
Example workflow in Fabric
Teams can use Tonic Textual in Fabric as part of their existing data workflows.1. Add the workload: Install Tonic Textual from the Microsoft Fabric Workload Hub.
2. Select data in OneLake: Choose document collections, transcripts, or other text datasets stored in OneLake.
3. Select an output location: Choose where the prepared dataset should be written in OneLake. This allows teams to keep the original source data unchanged while creating a separate AI-ready dataset.
4. Detect sensitive entities: Tonic Textual scans the dataset and identifies sensitive information.
5. Configure transformations: Select how each entity type should be handled.
Figure: Configure de-identification preference.
6. Use the prepared data: The resulting dataset remains in Fabric and can be used in downstream pipelines such as training or retrieval systems.
This approach allows teams to prepare text datasets before they are used in prompts or models. For a quick end‑to‑end walkthrough of this flow in action, check out the Tonic Textual on Microsoft Fabric video.