AI Enrichment

What is AI Enrichment?

AI enrichment uses large language models (LLMs) to analyze lead data and extract structured insights. Unlike traditional enrichment that relies on fixed data sources, AI enrichment can:

• Analyze company websites to extract funding stage, employee count, or services offered
• Generate personalized icebreakers based on prospect data
• Classify leads by fit score or buying intent
• Research pain points or use cases from public information
• Extract any custom data you define in your prompts

Supported AI Models

OpenAI

• GPT-4o: Highest quality, best for complex analysis
• GPT-4o Mini: Default — fast, affordable, handles most prompts well
• GPT-4 Turbo / GPT-4 / GPT-3.5 Turbo: Stable older models for specific use cases
• O1, O1 Mini: Reasoning models for analytical tasks

Google Gemini

• Gemini Pro: Google's flagship, competitive with GPT-4
• Gemini 1.5: Latest version with extended context window

Creating AI Enrichment Columns

Using Templates

Enrichabl provides pre-built templates for common enrichment tasks:

• Company Funding Stage: Extract funding round from website
• Personalized Icebreaker: Generate custom opening lines
• Lead Fit Score: Rate how well the company matches your ICP
• Pain Point Research: Identify challenges from website content
• Company Size Estimate: Determine employee count

Writing Custom Prompts

Create custom AI columns by writing prompts that reference lead fields with double curly braces:

# Extract company funding stage
Based on the company website {{website}}, determine their funding stage.
Return only one of: Bootstrapped, Seed, Series A, Series B, Series C+, or Unknown.

# Generate personalized icebreaker
Create a 1-sentence personalized icebreaker for {{first_name}} at {{company}}.
Use insights from their website {{website}} to make it relevant.
Keep it professional and conversational.

# Classify lead by fit
On a scale of 1-10, rate how well {{company}} fits our ICP:
- Target: B2B SaaS companies with 10-50 employees
- Based on: {{website}} and {{title}}
Return only the number (1-10).

# Research pain points
Visit {{website}} and identify 2-3 pain points that {{company}} likely faces
in their industry. Format as a comma-separated list.

Referencing Lead Fields

Use double curly braces to reference any field from your leads:

• {{email}} — Email address
• {{first_name}} — First name
• {{last_name}} — Last name
• {{company}} — Company name
• {{website}} — Company website
• {{title}} — Job title
• {{custom_field}} — Any custom field you imported

Chaining AI Columns

Create powerful workflows by using one AI column's output as input to another:

Example: Multi-Step Enrichment

1. Column 1 — Extract Services:
   text

   Based on {{website}}, list the main services {{company}} offers.

2. Column 2 — Match to Your Solution:
   text

   Given that {{company}} offers these services: {{services}}
   And our product is a [describe your product],
   explain in one sentence how we could help them.

3. Column 3 — Generate Icebreaker:
   text

   Write a personalized cold email opening line for {{first_name}} at {{company}}.
   Reference this value prop: {{value_prop}}

Cost Optimization

Choose the Right Model

• GPT-4o: Best quality, higher cost — use for complex analysis
• GPT-4o Mini: Best balance, default for most tasks
• GPT-3.5 Turbo: Lowest cost — use for simple classification
• Gemini Pro: Competitive alternative to GPT models

Reduce Token Usage

• Keep prompts concise and specific
• Request structured outputs (e.g., "1-10" instead of paragraphs)
• Use GPT-4o Mini or GPT-3.5 for simple tasks
• Use Multi AI columns to extract multiple fields in a single call

Validate Emails First

Always validate emails before AI enrichment to avoid wasting credits on invalid leads.

Best Practices

Prompt Engineering Tips

• Be specific: Clear instructions get better results
• Request format: Specify output format (number, yes/no, comma-separated, etc.)
• Provide context: Explain what you're trying to achieve
• Test first: Run on 10-20 leads before full batch

Handling AI Errors

If AI enrichment fails, common causes include:

• Missing website: AI can't analyze if no URL provided
• Rate limits: OpenAI has rate limits — spread processing over time
• Invalid prompts: Check that field references are correct
• Timeout: Simplify complex prompts or use a faster model

Advanced Techniques

Few-Shot Examples

Improve output quality by providing examples in your prompt:

Classify {{company}} by industry.

Examples:
- Stripe → Fintech
- Salesforce → B2B SaaS
- Tesla → Automotive

{{company}} → ?

Multiple Outputs in One Prompt

Extract multiple data points in a single AI call to save costs (use a Multi AI column):

Analyze {{website}} and return a JSON object with:
- funding_stage: (Seed/Series A/Series B/etc.)
- employee_count: (estimated number)
- tech_stack: (comma-separated list of 3 main technologies)

Return only valid JSON, no additional text.

Troubleshooting

Inconsistent Output Format

Problem: AI returns different formats for the same prompt

Solution: Be explicit about format. Use "Return ONLY [format]" in the prompt.

Slow Processing

Problem: AI enrichment takes too long

Solution: Use GPT-4o Mini for faster processing, or batch smaller groups.

API Key Errors

Problem: "Invalid API key" or rate limit errors

Solution: Check your API key in Settings, ensure sufficient credits, and verify rate limits.

Next Steps

• Set up OpenAI integration
• Combine with web scraping for richer data
• Learn about batch processing