Columns

Creation methods

You'll see two different creation methods:

• Start from scratch (manual): Opens the workflow builder directly with a fresh canvas. Perfect for custom workflows and advanced users.
• Template library (automatic): Browse pre-built templates for common tasks. Select a template to automatically configure the workflow builder.

Template library

Ready-made templates provide standardized rules for many different column types:

• Product content: Product descriptions, meta descriptions, meta titles, bullet points, FAQ sections.
• Marketing: Newsletter content, social media captions, ad copy, email campaigns.
• SEO: Title optimization, keyword extraction, alt image text, category descriptions.
• Business operations: Company research, lead qualification, data formatting, translations.

Template categories

Navigate multiple sectors including:

• E-commerce: Product-focused templates for online stores.
• General: Universal templates for any industry.
• HR & recruiting: Templates for talent acquisition and management.
• Insurance: Specialized templates for insurance content.
• And more: Additional sectors available based on your data.

Key canvas features

• Real-time validation: Highlights errors and suggests improvements as you build.
• Zoom and pan controls: Navigate large, complex workflows.
• Undo/Redo functionality: Full history tracking (Ctrl+Z/Ctrl+Y).
• Block grouping: Organize complex workflows into collapsible sections.
• Collaboration tools: Allow multiple team members to edit simultaneously.
• Template saving: Reuse successful workflow patterns.

How it works behind the scenes

When you process a row of data, the Context Window doesn't just send raw spreadsheet columns:

• 1. Pulls data from your selected columns for that specific row.
• 2. Formats it as clean key-value pairs like 'Product: Wireless Headphones' and 'Price: $99.99'.
• 3. Passes only this structured context to the AI, along with your prompts.

The three types of context

Every context window can include up to three different types of data, each serving a specific purpose:

• Output context (primary content): The main data the AI uses to generate content. For a product description field, you might include title, features, specifications, category, and target audience information.
• Search context (research data): Used specifically when your workflow needs to look up external information. This might include brand and model numbers for finding reviews, product IDs for competitor research, or company names for market analysis.
• SEO context (optimization guidelines): Provides metadata guidance and constraints without becoming part of the main content. Perfect for keywords that should be naturally integrated, brand voice guidelines to maintain consistency, and target audience preferences.

Why context windows matter

• Precision: Only relevant data reaches the AI.
• Consistency: The same context ensures similar results for similar inputs.
• Performance: Less data means faster processing and lower costs.
• Quality: Reduced hallucinations and more accurate outputs.

Advanced context features

For power users, Context Windows offer sophisticated capabilities like multi-tag support (one column can serve multiple contexts) and priority weighting (telling the system which columns are most important for search or validation). The system even validates your selections in real-time to prevent conflicts and optimize performance.

Generative blocks

At the heart of most workflows are the generative blocks that actually create content using AI:

• Generative output block: The workhorse. It calls AI models to create text based on your instructions, supporting everything from product descriptions to marketing copy. You write natural language prompts with variable references like 'Write a compelling description for {{Title}} that highlights {{Key_Features}} and appeals to {{Target_Audience}}.' Behind the scenes, it connects to major AI models (GPT-4, Claude, etc.) with full control over temperature, token limits, and other parameters.
• Image output block: Generates visuals using AI image models. Describe what you want - 'Create a professional product photo of {{Title}} in a modern {{Style}} setting' - and it handles the rest, integrating with services like DALL-E and Stable Diffusion.
• Translation block: Provides high-accuracy language translation with cultural adaptation. Unlike generic AI translation, this block understands context and can adapt content for different markets while maintaining brand voice.

Logic & control blocks

Once you have content creation covered, you'll need blocks to make decisions and control workflow flow:

• Conditional block: Handles straightforward logic based on your data. Set up rules like 'IF {{Price}} > 100 THEN create premium description ELSE create standard description' to route different products through different content paths. It's perfect for category-based content variations or tiered pricing strategies.
• LLM conditional block: For more sophisticated decision-making, this lets the AI itself make judgments. Instead of rigid rules, you can write natural language conditions like 'IF this product appears luxurious or high-end THEN use formal, elegant language.' This handles the nuance and context that traditional rule-based systems miss - understanding that 'premium' can mean different things in different contexts.

Static & formatting blocks

Not every part of your workflow needs AI generation. Sometimes you need to add structure, formatting, or consistent elements without calling expensive AI models:

• Static output block: Perfect for adding fixed text, HTML templates, or standard content without any AI processing. Use it for headers, footers, disclaimers, or formatting wrappers. Since it doesn't make AI calls, it executes instantly and costs nothing.
• General rule block: Sets instructions that apply to all AI blocks in your workflow. Place one at the top with rules like 'Always use British English spelling' or 'Maintain a professional, consultative tone' and every generative block will follow these guidelines.
• Section blocks: Complex workflows can become unwieldy, so these let you group related blocks into logical units with descriptive names. Think of them as folders in your workflow - you can collapse sections to focus on specific parts and create reusable workflow components.

Data & integration blocks

To create truly powerful workflows, you'll often need to pull in data from other sources or perform calculations across multiple records:

• Reference block: Connects your workflow to other datasets in your workspace or organization. Need supplier information for your products? Or customer order history for personalized emails? This block pulls that data using matching criteria, enriching your current dataset with related information.
• Aggregation block: For calculations across multiple rows, this performs operations like sum revenues, calculate averages, find maximum values, or group data by categories. Perfect for creating summary reports or conditional logic based on aggregated data.

Fine-tuning your blocks

Every block comes with configuration options to customize its behavior:

• Layout & display: Choose horizontal or vertical text orientation, set multi-line text areas for complex prompts, add HTML markup hints for content structure.
• Execution & performance: Configure error handling and fallback behaviors, set timeout limits for long-running operations, enable retry logic for transient failures.
• Advanced capabilities: Create branching paths with conditional logic, implement looping for batch processing scenarios, enable caching to reuse results and improve performance.

Prompt-to-workflow

This is where the magic happens. Instead of manually building workflows, you simply describe what you want in plain English.

• How it works: Imagine you need to 'generate product descriptions, check if they're positive, and create social media posts for the good ones.' Instead of dragging blocks and writing prompts, you just type that sentence.
• The system: Analyzes your request, understands your data structure, and automatically builds the workflow for you. It applies intelligence to select the right blocks, write effective prompts, and connect them in the optimal sequence.
• Result: A complete, working workflow in seconds rather than hours. This approach is perfect for new users, quick prototyping, or common business scenarios. You get professional-grade workflows without needing to understand the technical details.

Column templates

For common business scenarios, why reinvent the wheel? Cension provides a library of professionally-designed workflow templates that solve specific problems.

• Battle-tested workflows: These aren't generic examples - they're proven implementations created from successful real-world use cases.
• Business function categories: Content marketing (blog posts, social media captions, email campaigns, product descriptions), Sales & outreach (lead qualification, personalized emails, company research), SEO & optimization (meta titles, keyword extraction, content optimization), Data operations (translation, formatting, validation, duplicate detection).
• Intelligent recommendations: Based on your dataset's content, Cension automatically suggests relevant templates (e.g., import product data and it suggests product description and social media templates; upload company information and it recommends research and qualification workflows).
• Optimized performance: Each template comes pre-configured with optimal settings, proven prompts, and efficient context window setups. They're not just faster to implement - they're more cost-effective and reliable than building from scratch.

From scratch

For maximum control and customization, you can build workflows entirely from scratch using the visual canvas.

• Complete control: Over prompts, logic, data flow, and performance optimizations. You have access to advanced features like complex conditional branching, custom integration blocks, and sophisticated error handling.
• Blank canvas approach: Start with a blank canvas and drag blocks from the library to create exactly the workflow you envision.
• Ideal for: Organizations with unique requirements that don't fit existing templates, or for technical users who want to innovate beyond standard patterns.

Hybrid

Many successful implementations start with a template and then customize extensively. This approach combines speed with flexibility.

• Template foundation: Begin with a proven template that solves 80% of your problem, then modify the prompts, add custom logic, and integrate additional data sources to address your specific needs.
• Perfect balance: Reliable (thanks to the proven foundation) and perfectly tailored to business needs.
• Common pattern: Standard processes with unique requirements are often best served by this approach.

Phase 1: Planning & analysis

Before you start building, take time to plan your workflow carefully. Good planning prevents costly mistakes and ensures your workflow delivers real value:

• Define objectives: What should the final content look like? Who will use it? What metrics matter most?
• Inventory data: Review every column in your dataset and understand what information is available. Look for relationships between columns that you can leverage. Identify any gaps where you might need external research.
• Choose approach: Consider your timeline, technical expertise, and how unique your requirements are. New to workflows? Start with prompt-to-workflow. Need proven solutions? Use templates. Have complex requirements? Build from scratch or use the hybrid approach.

Phase 2: Context window configuration

Context windows are the foundation of effective workflows. This is where you tell the AI exactly what data to consider when generating content:

• Output context: Select columns that provide rich, descriptive data about your subject. For product descriptions, you might choose title, features, category, and specifications - all the information that helps the AI understand what it's describing.
• Search context: If your workflow needs external research, configure this to tell the system which columns to use when searching the web or external databases. Choose identifying information that creates effective search queries.
• SEO context: For optimization tasks, add columns that provide guidance and constraints without becoming part of the main content. Keywords to incorporate, target audience preferences, and brand voice guidelines all belong here.

Phase 3: Building the workflow

• Step 7: Add core generation blocks: Start with your primary generative output block. Write clear, specific prompts using variable references. Example: 'Write a compelling product description for {{Title}} that highlights {{Key_Features}} and appeals to {{Target_Audience}}.'
• Step 8: Add logic & branching: Insert conditional or LLM conditional blocks for decision-making. Create different paths for different scenarios. Example: Different description styles based on product category or price point.
• Step 9: Include supporting blocks: Add general rule blocks for consistent tone and style. Include static output blocks for formatting and structure. Add reference blocks to pull in related data from other datasets.
• Step 10: Implement error handling: Plan for edge cases and potential failures. Add fallback content for when research fails. Include validation steps to ensure quality.

Phase 4: Testing & optimization

• Step 11: Test individual blocks: Use the preview feature to test each block in isolation. Verify that context window data is passed correctly. Check that variable references work as expected.
• Step 12: Test full workflow: Run the complete workflow on sample data. Verify that branching logic works correctly. Check that all blocks execute in the right order.
• Step 13: Performance optimization: Review token usage and execution times. Optimize context window selections to reduce unnecessary data. Consider caching for frequently used external data.
• Step 14: Quality assurance: Test edge cases and unusual data scenarios. Verify output quality meets your standards. Check for hallucinations or incorrect information.

Phase 5: Deployment & monitoring

• Step 15: Configure processing settings: Set up auto-process rules for new data. Configure audit mode if human review is required. Set up daily processing limits for cost control.
• Step 16: Monitor & iterate: Track workflow performance in the operations tab. Monitor credit usage and identify optimization opportunities. Gather user feedback and iterate on the workflow.
• Step 17: Scale & maintain: Save successful workflows as templates for reuse. Train team members on workflow usage. Regularly review and update based on changing needs.

Best practices throughout

• Context window best practices: Be selective - include only relevant columns. Balance breadth and focus. Test iteratively.
• Block organization best practices: Use clear, descriptive names. Add comments to complex logic. Group related blocks into sections.
• Performance best practices: Test with small datasets first, then scale up. Monitor credit consumption regularly. Use caching for expensive external API calls.
• Maintenance best practices: Document your workflows. Version control important workflows. Regularly review and optimize.

E-commerce product description generator

Imagine you run an online store with products ranging from budget accessories to luxury items. You want compelling descriptions that not only convert but also match each product's price positioning.

• The solution: A smart workflow that analyzes product data and generates tailored descriptions automatically.
• Context setup: Output context includes title, features, specifications, category, and price range - all the information the AI needs to understand the product. Search context uses brand, model number, and GTIN for competitor research. SEO context incorporates target keywords and brand voice guidelines.
• Workflow structure: General rule block sets the overall tone. LLM conditional block routes premium/mid-range/budget products differently. Reference block pulls competitor ratings. Static output block wraps everything in proper HTML formatting.
• Result: Every product gets a perfectly positioned description that converts better and appeals to the right audience, with real competitor insights built right in.

Social media content pipeline

Content marketers know the struggle: you publish a great blog post, but then need to create separate content for Twitter, LinkedIn, Instagram, and more. Each platform has different requirements, character limits, and audiences.

• The solution: One blog post becomes platform-optimized content for all your channels.
• Context setup: Output context contains the blog title, content, key points, and target platform. Search context for topic categories and author information. SEO context with brand hashtags and content pillar guidelines.
• Workflow structure: General rule establishes the overall approach. Conditional block routes content based on target platform (Twitter threads, LinkedIn posts, Instagram captions). Aggregation block tracks performance across all platforms.
• Result: Publish once, distribute everywhere. One blog post becomes multiple pieces of social content, each perfectly optimized for its platform's audience and format.

Lead qualification & outreach system

Automatically qualify leads, research companies, and generate personalized outreach sequences.

• Context setup: Company_Name, Industry, Company_Size, Lead_Source (Output); Company_Website, Company_Name, Key_Contacts (Search); Target_Persona, Value_Proposition (SEO).
• Workflow structure: General rule for sales development approach. LLM conditional scores leads as high/medium/low priority. Reference block pulls CRM interaction history. Generative block creates highly relevant outreach with company insights.
• Result: Leads are automatically scored, researched, and receive personalized outreach sequences that reference specific company details.

Multi-language product catalog generator

Generate product catalogs in multiple languages with culturally appropriate content.

• Context setup: Product_Name, Product_Description, Features, Specifications (Output); Product_ID, Category (Search); Target_Market, Cultural_Notes, Local_Preferences (SEO).
• Workflow structure: General rule for localization expertise. Conditional block routes by language (Spanish, German, Japanese branches with cultural adaptation). Reference block pulls local market preferences. Static output generates complete catalog pages.
• Result: Single product entries automatically generate complete, culturally-adapted catalogs for multiple international markets.