AC
AutoClaygent
Lesson 4 of 850% complete
15 min read

Prompt Patterns

9 production-ready patterns AutoClaygent uses for common use cases

Why These Patterns Are Different

These aren't generic "find company info" patterns. They're battle-tested approaches from real GTM engineering projects that achieve 85%+ accuracy. Each pattern solves a specific problem that native integrations can't handle well.

ℹ️

These patterns use techniques like URL subdomain detection (95% accuracy vs 70% for text scraping) and decision trees with explicit branching logic.

How AutoClaygent Handles This

AutoClaygent knows all 9 patterns and chooses the right one based on your description. It customizes each pattern for your specific use case—adding the right fallbacks, sources, and output fields automatically.

Pattern 1: Business Model Classification

Use case: Classify service businesses by revenue model (subscription vs transactional vs hybrid)

Why it's valuable: Most CRM data about "business type" is wrong. Explicit decision trees beat generic prompts.

Given the company domain: {{domain}} Classify this business by their revenue/pricing model. **Research Steps:** 1. Navigate to {{domain}} and look for: - /pricing, /plans, /membership, or /subscribe pages - Explicit dollar amounts with time periods ($/month, $/year) 2. Classify based on what you find: SUBSCRIPTION = Recurring membership model - Look for: "monthly", "annual", "per seat", "subscription" - Customer pays regularly (weekly/monthly/yearly) - Examples: "$99/month", "Starting at $49/user/month" TRANSACTIONAL = Per-use or one-time payment - Look for: "per project", "one-time", "pay as you go" - Customer pays for each service/product - Examples: "$500 per audit", "Quote-based pricing" HYBRID = Offers BOTH recurring AND one-time options - Has subscription tiers AND pay-per-use options - Examples: "Base plan $49/mo + $5 per additional user" ENTERPRISE = Custom/contact sales pricing - No public pricing, "Contact us", "Request demo" 3. If no pricing page exists: - Check /about or homepage for business model clues - Look at how they describe their service delivery **Output as JSON:** { "business_model": "subscription" | "transactional" | "hybrid" | "enterprise" | null, "pricing_found": true | false, "pricing_details": "Brief description of what you found" | null, "evidence_url": "URL where you found pricing info", "confidence": "high" | "medium" | "low" }

Pattern 2: Platform Detection via URL Patterns

Use case: Detect what SaaS platforms a company uses by analyzing portal URLs

Why it's valuable: URL patterns are 95%+ accurate. Companies use vendor-hosted portals that reveal their tech stack.

Given the company domain: {{domain}} Detect what platforms this company uses by analyzing their portal/login URLs. **Research Steps:** 1. Look for customer-facing portals: - Check {{domain}}/login, {{domain}}/portal, {{domain}}/app - Look for "Client Login", "Patient Portal", "Customer Portal" links - Check the footer for portal links 2. When you find a portal link, analyze the URL pattern: **Subdomain Patterns (highest confidence):** - {company}.platformname.com → Uses Platform Name - app.{company}.com redirecting to vendor → Uses that vendor - portal.vendorname.com/{company} → Uses Vendor Name **Common Platform URL Patterns:** - *.salesforce.com → Salesforce - *.hubspot.com → HubSpot - *.zendesk.com → Zendesk - *.calendly.com → Calendly - *.webflow.io → Webflow 3. Check for scheduling/booking tools: - Look for "Book Now", "Schedule", "Appointments" - These often reveal Calendly, Acuity, Cal.com, etc. **Output as JSON:** { "platforms_detected": [ { "platform_name": "Name of platform", "platform_category": "CRM | Support | Scheduling | Chat | Website", "detection_method": "subdomain | redirect | widget | integration_page", "evidence_url": "URL that revealed this" } ], "portal_url": "Customer portal URL if found" | null, "confidence": "high" | "medium" | "low" } **Why this works:** URL subdomain patterns are 95%+ accurate vs text scraping at 70%. DNS doesn't lie.

Pattern 3: CRM Validation & Mismatch Detection

Use case: Find where your existing CRM data is WRONG by comparing to website reality

Why it's valuable: Finding 5,000 wrong records is more valuable than enriching 50,000 records.

Given: - Company domain: {{domain}} - CRM field to validate: {{field_name}} - Current CRM value: {{crm_value}} Validate if the CRM value matches what's actually on the company's website. **Research Steps:** 1. Navigate to {{domain}} and find information related to {{field_name}} 2. Look in these locations based on field type: - Company size/employees → /about, /team, /careers - Industry/vertical → Homepage, /about, meta description - Location/HQ → /contact, /about, footer - Business type → /pricing, /about, homepage 3. Compare what you find to the CRM value: MATCH = Website confirms CRM value MISMATCH = Website contradicts CRM value (specify what you found) STALE = CRM value appears outdated UNVERIFIABLE = Cannot confirm or deny from website **Output as JSON:** { "field_validated": "{{field_name}}", "crm_value": "{{crm_value}}", "website_value": "What you found on website" | null, "validation_status": "match" | "mismatch" | "stale" | "unverifiable", "evidence": "Specific text proving your finding", "recommendation": "keep" | "update" | "review", "confidence": "high" | "medium" | "low" } **Why this works:** Finding errors in existing data is more valuable than adding new fields.

More Patterns (Full Course)

The complete course includes 6 additional production patterns:

PatternUse CaseKey Innovation
Digital Maturity ProfilingScore prospects by online sophisticationEmail domain as proxy for website existence
B2B vs B2C DetectionDetermine target marketNavigation menu as primary signal
Corporate Structure DetectionFind parent companies, subsidiariesIdentify actual decision-maker level
Buying Intent DetectionFind companies ready to buy NOWFunding + leadership timing signals
Multi-Axis ClassificationClassify along 5 dimensionsPer-dimension confidence scoring
SMB Owner DiscoveryFind small business owner names5-stage waterfall with confidence stacking

Pattern Combinations

For comprehensive enrichment, chain patterns in sequence:

Sales Timing Optimization

  1. Buying Intent Detection → Is now a good time?
  2. CRM Validation → Is our data still accurate?
  3. Corporate Structure → Who actually decides?

SMB Outreach Pipeline

  1. Digital Maturity → Segment before enrichment (personal email = different approach)
  2. SMB Owner Discovery → Find the decision-maker via 5-stage waterfall
  3. Business Model → Understand how they make money

Tech Stack Intelligence

  1. Platform Detection → What do they use? (URL patterns)
  2. Multi-Axis Classification → Tech sophistication level
  3. Buying Intent → Are they likely to switch vendors?
💡
Pro Tip

Each pattern should be its own Claygent column in Clay. Don't combine them into one mega-prompt—that violates the 3-task rule and tanks accuracy.

Key Principles

These patterns work because they follow these principles:

  • URL patterns beat text parsing — 95% accuracy vs 70%
  • Navigation signals come first — Check menu before body content
  • Null is a feature — Returning null when unsure beats wrong data
  • Confidence by dimension — Not one overall score, but per-field
  • Mismatch detection is valuable — Finding errors in existing data
  • Decision trees beat "find X" — Explicit branching logic

Ready to build Claygents that actually work?

Get the complete course with interactive playground and all 9 prompt patterns.