Personalization Engine

Pipeline contract

Stage 13 of 20. Reads 12-segmented.csv, 09-signals.csv. Writes 13-personalized.csv (adds angle_source, situation_line, value_line, cta_soft).

• HOW to run it at volume: references/subagent-pattern.md — 1 lead, then batches of 10 with approval, then fan-out on Haiku sub-agents. Do not fan out before the human approves a batch.
• Angle order: a dated signal from stage 9, then audience evidence (which page they follow, what they commented on), then segment-level bucket copy. Record which one was used in angle_source.
• No evidence = bucket copy. Never invent a fact about a person or company.
• Merge company_name_clean and first_name_clean from stage 8, never the raw provider names. Stages 14–16 inherit this.

Files live in runs/<run>/. Column definitions: /prospecting-system. Standalone use: ignore this block.

When to Use

• Building personalization workflows for outbound campaigns
• Training SDRs or AI systems on what "good personalization" looks like
• Designing variable strategies for email templates
• Upgrading campaigns from generic to signal-based personalization

Framework

The Personalization Hierarchy

Not all personalization is equal. Higher levels convert better but take more effort. Match your level to your volume and deal size.

Rule of thumb:

• High-value accounts (>$50K ACV): Level 4-5 personalization
• Mid-market (10K-50K): Level 3-4
• SMB at scale (<10K): Level 2-3 with automation

6 Signal Categories

Every personalization starts with a signal — something you observed about the prospect or their company. Here are the six signal categories ranked by conversion impact:

1. Hiring Signals (Highest Intent)

What to look for: Job postings that indicate a need your product solves.

Data sources: LinkedIn Jobs, Indeed, company careers page, Otta, Wellfound

2. Funding Signals

What to look for: Recent capital raises that trigger growth mode.

Data sources: Crunchbase, PitchBook, TechCrunch, LinkedIn announcements

3. Technology Signals

What to look for: Tech stack changes that indicate shifting priorities.

Data sources: BuiltWith, Wappalyzer, SimilarTech, G2 reviews, job descriptions (tech requirements)

4. Content Signals

What to look for: What the prospect is publishing, sharing, or engaging with.

Data sources: LinkedIn feed, podcast directories, company blog, event speaker lists

5. Company Event Signals

What to look for: Organizational changes that create new needs.

6. Performance Signals

What to look for: Publicly visible indicators of business performance.

The Personalization Formula

Every personalized email element follows this 3-part structure:

[OBSERVATION] + [IMPLICATION] + [BRIDGE]

Bad personalization: "Hi Sarah, I see you work at Acme Corp. We help companies like Acme..." Good personalization: "Sarah — noticed Acme just posted 3 SDR roles in Austin. Scaling the outbound team usually means the founder-led selling phase worked, but the playbook isn't documented yet."

The difference: bad personalization names facts. Good personalization draws insights from facts.

Variable Architecture

Design your email templates with a layered variable system:

Tier 1: Auto-Populated (No Manual Work)

These come straight from your lead list:

Tier 2: Enrichment-Derived (Automated Research)

These require data enrichment but can be automated:

Tier 3: Research-Derived (Manual or AI-Assisted)

These require reading/analyzing content:

Personalization at Scale: The Bucket Strategy

For high-volume campaigns (1,000+ leads), you can't write individual emails. Instead, create personalization "buckets":

Step 1: Segment your list by signal type

List of 2,000 leads
├── Bucket A: Recently funded (400 leads)
├── Bucket B: Hiring sales roles (350 leads)
├── Bucket C: Tech stack change (250 leads)
├── Bucket D: Industry-specific pain (600 leads)
└── Bucket E: No strong signal (400 leads)

Step 2: Write bucket-specific opening lines Each bucket gets its own personalized opener that feels individual but applies to the whole segment:

Step 3: Layer in Tier 1 variables for the personal touch

The result: every lead gets an email that feels researched, but you wrote 5 versions, not 2,000.

Personalization Quality Scoring

Rate every personalized email element on this scale before sending:

Minimum threshold: Score 3+ for mid-market, Score 4+ for enterprise.

Templates

Signal Research Template

For each lead, capture:

Company: {{companyName}}
Contact: {{firstName}} {{lastName}}, {{title}}

Signal Scan:
- [ ] Hiring signals: ___
- [ ] Funding signals: ___
- [ ] Tech signals: ___
- [ ] Content signals: ___
- [ ] Company events: ___
- [ ] Performance signals: ___

Strongest signal: ___
Personalization angle: ___
Opening line draft: ___
Quality score (1-5): ___

Personalization Brief (For AI or SDR)

Client: {{clientName}}
Target Persona: {{persona}}
Campaign Angle: {{angle}}

Personalization Requirements:
- Minimum quality score: {{minScore}}
- Required signal types: {{signalTypes}}
- Variables available: {{variableList}}

Bucket Definitions:
- Bucket A ({{bucketName}}): {{criteria}} → {{openingApproach}}
- Bucket B ({{bucketName}}): {{criteria}} → {{openingApproach}}
- Bucket C ({{bucketName}}): {{criteria}} → {{openingApproach}}

Tips

• The best personalization references something the prospect DID, not something they ARE. "I saw you posted about X" beats "I see you're a VP of Sales" every time.
• Don't over-personalize Step 1 if it comes at the cost of volume. A Level 3 email sent to 500 people beats a Level 5 email sent to 50 — unless your ACV justifies the time.
• Keep a running database of which signal types drive the highest positive reply rates. After 3 months, you'll know exactly which signals to prioritize.
• When in doubt, lead with the hiring signal. It's the most reliable indicator of active buying intent.
• AI can handle Tier 1-2 personalization at scale. Reserve human effort for Tier 3 (content-based insights) on your highest-value targets.
• Test "no personalization" as a control variant. Sometimes a strong offer with zero personalization beats weak personalization — and it tells you if your copy is doing the heavy lifting.

Progressive disclosure: load signal-specific research playbooks and enrichment tool integrations only when building personalization for a specific campaign.