Google Cloud Introduces Object Enrichments to Turn Unstructured Data into AI-Ready Context

Google Cloud has introduced Object Enrichments, a new capability for Cloud Storage designed to transform unstructured files into information that artificial intelligence systems can understand and use.

The feature uses AI models to automatically extract contextual metadata from stored objects, including images and PDF documents. The extracted information is saved alongside the original files, allowing AI agents and applications to access relevant context without repeatedly processing large volumes of raw data.

The launch addresses a growing enterprise challenge: organisations store enormous amounts of information in images, documents, videos and audio files, but making that information useful for AI applications can be expensive and operationally complex.

Why Unstructured Data Creates Challenges for AI

Enterprise information is not always stored in organised databases. Important business knowledge often exists in documents, photographs, recordings, videos and other files held in object storage.

AI agents need meaningful information from these files to answer questions and complete business tasks. However, processing thousands or millions of raw files whenever a user makes a request can increase costs and response times.

Businesses have traditionally built custom pipelines to extract information, process files and store metadata separately. These systems require infrastructure management, error handling, scheduling and ongoing maintenance.

Google Cloud's Object Enrichments aims to simplify this process by bringing AI-based metadata extraction directly into Cloud Storage.

How Google Cloud Object Enrichments Works

The new capability automatically processes selected Cloud Storage objects using supported AI models and stores the resulting metadata in a feature called Object Context.

The setup involves three main steps:

  1. Define the data scope: Select the Cloud Storage buckets containing the files that need processing and specify the processing location.

  2. Choose AI models: Select one or more of the available first-party models for images and PDFs.

  3. Grant permissions: Provide the required Identity and Access Management (IAM) permissions for the service to access the selected buckets.

The managed service handles model inference, response processing, retries and throttling. This reduces the need for businesses to build and maintain separate processing pipelines.

Lower Costs for Large-Scale AI Processing

Cost reduction is one of the main benefits highlighted by Google.

According to the company's published pricing, enriching 10 million images or PDFs costs approximately $700. Google says its image annotation models provide savings of around 75% compared with similar services from other cloud providers.

Supported image capabilities include object detection, face detection, image labelling and explicit content detection. The service also includes a PDF classification model that can organise documents into categories such as Finance, Games, News and Health.

Google also cited early customer results. L'Oréal reported labelling its creative asset repository 100 times faster, while Hyperscience reported a 50% or greater improvement in document classification accuracy alongside a two-to-three-times reduction in classification inference costs.

These are customer-reported outcomes rather than independently verified industry-wide benchmarks.

Object Context Keeps Metadata Alongside Original Files

A key feature of Object Enrichments is that extracted information is stored alongside the original object instead of being maintained in a separate metadata database.

Google says this approach offers several operational advantages:

  • Simplified management: Context follows files when they are copied, moved or replicated and is removed when the original object is deleted.

  • Access control: IAM permissions allow administrators to control who can read, add, update or remove object context.

  • Faster discovery: Applications can filter objects using contextual key-value information.

  • BigQuery integration: Object context can flow into BigQuery through Storage Insights datasets, allowing teams to analyse file insights alongside structured business data.

This can reduce the need for separate storage systems and custom synchronisation processes.

Supporting AI Agents and Enterprise Applications

Google identifies several potential use cases for Object Enrichments.

Marketing and media organisations can use image annotations to find specific creative assets across large archives. Online marketplaces and digital platforms can use content detection to support trust and safety operations.

Autonomous vehicle, robotics and smart-home companies can also use extracted metadata to identify specific images or sensor records for AI model training.

In addition, organisations can combine insights from documents and images with structured business information to support analytics, personalisation and fraud detection.

Google also plans to support richer multimodal processing through Gemini for tasks such as summarisation, entity extraction, transcription and translation. This functionality is described as upcoming rather than generally available.

Availability and What Comes Next

Object Enrichments is available in preview through the Google Cloud Storage console and Cloud Storage API.

The initial offering includes five first-party models for image and PDF processing. Google says the service can scale to millions or billions of objects, with BigQuery integration and connections to AI agents through the Cloud Storage MCP server.

For enterprises managing large unstructured datasets, the launch represents another step towards making existing cloud storage more useful for AI applications without requiring extensive custom infrastructure.