AI Structured Responses early-access

Last updated:
Edit on GitHub

Structured Responses enable AI to return data in predefined, validated formats instead of free-form text. This allows you to integrate AI output directly into your business workflows, databases, and applications with confidence in data quality and consistency.

Business value

Reliable, predictable data

Instead of parsing free-text AI responses that vary in format and structure, you receive data in consistent, predefined schemas. This eliminates ambiguity and ensures every AI response contains exactly the fields you need.

Seamless system integration

Structured data from AI can be directly inserted into databases, passed to APIs, or used in business logic without manual transformation. This accelerates development and reduces integration complexity.

Reduced errors and rework

Validated structured responses eliminate common issues with AI-generated content:

  • Missing required information
  • Inconsistent data formats
  • Incorrect data types
  • Unparseable responses

Your development team spends less time handling edge cases and more time building features.

Enable data-driven processes

Structured Responses allow AI to participate in data-intensive workflows where free-text is insufficient:

  • Extract product attributes for catalog import
  • Generate structured reports and analytics
  • Classify customer inquiries for routing
  • Create validated forms and documents

When to use Structured Responses

Use Structured Responses when AI output must feed into systems, processes, or databases that require specific data fields:

Product data extraction and enrichment

Scenario: You receive product information from suppliers in various formats such as PDFs, emails, spreadsheets, and free-text descriptions. You need to extract standardized product attributes for your catalog.

Without Structured Responses: AI generates text descriptions. Your team manually parses the text to identify SKU, price, dimensions, materials, and other attributes. Inconsistent formats lead to errors.

With Structured Responses: AI extracts data and returns it in a predefined structure:

{
  "sku": "PROD-12345",
  "name": "Industrial Pump Model X",
  "price": 1299.99,
  "material": "stainless steel",
  "dimensions": {"length": 50, "width": 30, "height": 40},
  "weight_kg": 15.5,
  "category": "industrial-equipment"
}

Business impact: Faster catalog updates, consistent data quality, reduced manual data entry, fewer import errors.

Customer inquiry classification

Scenario: Customer service receives hundreds of inquiries daily. Each must be classified by type, priority, and routed to the appropriate team.

Without Structured Responses: AI generates text like “This seems like a high-priority billing question.” Your system cannot automatically route based on text.

With Structured Responses: AI returns structured classification:

{
  "category": "billing",
  "subcategory": "payment_issue",
  "priority": "high",
  "requires_human_review": true,
  "sentiment": "negative",
  "estimated_resolution_time_minutes": 30
}

Business impact: Automatic routing, faster response times, better resource allocation, improved SLA compliance.

Content generation with metadata

Scenario: Marketing team needs product descriptions with SEO metadata for thousands of products. Content must follow brand guidelines and include specific fields.

Without Structured Responses: AI generates product descriptions as text. Your team manually extracts title, meta description, keywords, and formats content for CMS.

With Structured Responses: AI returns complete structured content:

{
  "title": "Industrial Pump Model X - High-Efficiency 50L/min",
  "description": "Premium stainless steel pump ideal for...",
  "meta_description": "Shop Industrial Pump Model X...",
  "keywords": ["industrial pump", "stainless steel", "high efficiency"],
  "target_audience": "B2B industrial buyers",
  "tone": "professional",
  "word_count": 250
}

Business impact: Faster content production, consistent SEO optimization, ready for direct CMS import, scalable content operations.

Form and document generation

Scenario: Generate structured documents like quotes, invoices, or specification sheets based on customer conversations or data.

Without Structured Responses: AI generates formatted text that requires manual reformatting and validation before use.

With Structured Responses: AI produces validated document data ready for template rendering or PDF generation.

Business impact: Automated document creation, reduced manual work, consistent formatting, fewer errors in formal documents.

How it works

When requesting information from AI, you specify the expected data structure. The AI provider validates its response against this schema before returning it to you. If the response does not match the schema, the system automatically retries until a valid response is generated or the retry limit is reached.

Structured response flow

  1. Define data schema: You specify what fields and data types you expect from AI.

  2. AI processes request: AI understands both your question and the required output format.

  3. AI generates structured data: AI produces a response that matches your schema.

  4. Validation: The system validates the response against your schema.

  5. Retry if needed: If validation fails, the system automatically retries.

  6. Return validated data: You receive data guaranteed to match your schema.

Schema definition

Schemas are defined using Spryker Transfer objects, which are XML-defined data structures your development team creates. These Transfer objects specify:

  • Field names
  • Data types (string, integer, float, boolean, arrays, nested objects)
  • Required vs. optional fields
  • Nested structures
  • Descriptions for each field

Key capabilities

Type safety

All data returned matches the specified types:

  • Strings are always strings
  • Numbers are always numbers
  • Dates follow specified formats
  • Booleans are true or false, never “yes” or “maybe”

This eliminates type coercion errors and unexpected data formats.

Required field enforcement

Mark fields as required in your schema. AI responses must include all required fields, or the system retries automatically.

Nested data structures

Define complex, hierarchical data:

Product
  - name
  - price
  - specifications (nested object)
    - dimensions
      - length
      - width
      - height
    - weight
    - material
  - reviews (array of objects)
    - rating
    - comment
    - date

Automatic validation and retry

If AI returns invalid data:

  1. System detects validation failure
  2. System automatically retries the request
  3. Process repeats up to configured retry limit
  4. You receive either valid data or clear error indication

No manual error handling needed in most cases.

Multiple output formats supported

Use Structured Responses for various scenarios:

  • Single records (one product)
  • Collections (list of products)
  • Hierarchical data (order with line items)
  • Mixed types (text with metadata)

Business scenarios

Supplier data integration

Challenge: Your company works with hundreds of suppliers who provide product information in inconsistent formats. Importing this data into your catalog requires extensive manual processing.

Solution: AI with Structured Responses processes supplier data:

  • Reads PDFs, emails, or Excel files from suppliers
  • Extracts relevant product information
  • Returns data in your standard catalog import format
  • Validates completeness before import

Result: 80% reduction in manual data entry, faster time-to-market for new products, improved data quality, scalable supplier onboarding.

Automated content operations

Challenge: E-commerce team needs to generate product descriptions, category pages, and SEO metadata for thousands of SKUs. Manual content creation is slow and inconsistent.

Solution: AI content generator with Structured Responses:

  • Generates product descriptions based on specifications
  • Creates SEO-optimized titles and meta descriptions
  • Suggests relevant keywords and categories
  • Returns structured content ready for CMS import

Result: 90% faster content production, consistent brand voice, better SEO performance, ability to scale to new markets and languages.

Customer service automation

Challenge: Customer service receives diverse inquiries that must be classified, prioritized, and routed. Manual classification is slow and inconsistent.

Solution: AI classifier with Structured Responses:

  • Analyzes customer inquiry text
  • Classifies by category, priority, and sentiment
  • Determines appropriate department and estimated resolution time
  • Returns structured classification for automatic routing

Result: Instant inquiry routing, improved SLA compliance, better resource utilization, higher customer satisfaction.

Data quality and validation

Challenge: Your systems contain product data with missing information, inconsistent formats, and quality issues. Manual data cleanup is time-consuming.

Solution: AI data validator with Structured Responses:

  • Reviews existing product records
  • Identifies missing or inconsistent information
  • Suggests corrections and improvements
  • Returns structured validation report with specific issues and recommendations

Result: Improved data quality, faster data cleanup, consistent data standards, reduced downstream errors.

Comparison to alternatives

Free-text AI responses

Free-text approach: AI generates narrative responses. Your application uses regular expressions, parsing logic, or additional AI calls to extract structured information.

Structured Responses: AI generates validated, structured data directly. No parsing needed.

Traditional form-based systems

Traditional approach: Users manually fill structured forms. Slow, error-prone, requires exact knowledge of required fields.

Structured Responses: AI extracts or generates structured data from natural language or unstructured input. Fast, intelligent, user-friendly.

Manual data entry

Manual approach: Staff read documents, emails, or descriptions and manually enter data into structured fields.

Structured Responses: AI processes unstructured input and produces structured output automatically.