Cold Email Personalization

Personalization is the difference between an email that earns a reply and one that gets deleted. But personalization doesn't mean "insert {first_name} and {company}." Merge tags are not personalization. Personalization is referencing something specific about the recipient that proves the email was written for them, not batch-sent to 500 people.

The principle: personalization should answer one question for the recipient: "Why me, why now?" If the personalized element doesn't answer that question, it's decoration, not personalization. A compliment about their LinkedIn banner photo is decoration. A reference to the RevOps hire they posted last week is personalization.

The Personalization Spectrum

Which level to use

The 7 Personalization Token Types

A personalization token is a single, specific data point about the prospect that you insert into a templated email. Each token type answers "why you, why now" differently.

Token 1: Signal Token

Reference a specific, recent event at their company.

Signal token rules:

• Must be from the last 90 days. Older signals are history, not triggers
• Must connect to a problem. "Congrats on the Series B" alone is a pleasantry. "Congrats on the Series B. The scaling chaos usually hits around month 3" is personalization with purpose

Token 2: Content Token

Reference something the prospect published or said publicly.

Content token rules:

• Reference their specific take, not just the topic. "Loved your LinkedIn post" is generic. "Your point about attribution being a data problem" is specific
• Only use content from the last 30 days unless it's evergreen (a major conference talk, a widely shared article)

Token 3: Role Token

Reference something specific about their job function, title, or responsibility.

Role token rules:

• Infer, don't state the obvious. "I see you're a VP of Sales" wastes the token. "Starting the VP Sales role in Q1 usually means..." adds insight to the data point

Token 4: Company Token

Reference something about their company's situation, stage, or trajectory.

Company token rules:

• Combine 2 company data points for depth. "Series B" alone is thin. "Series B with a 12-person sales team scaling to 30" is a situation
• Don't repeat what's on their LinkedIn headline. They already know their company size. Add an implication

Token 5: Metric Token

Reference a specific number or benchmark relevant to their situation.

Metric token rules:

• The metric must be relevant to their specific situation, not a generic stat
• Back it up. If they ask "where did you get that number?" you should have an answer

Token 6: Mutual Token

Reference a shared connection, community, or experience.

Mutual token rules:

• The connection must be real. Fabricating a mutual connection destroys trust permanently
• Shared customers should be similar to them (same stage, vertical, or size), not just a big logo

Token 7: Stack Token

Reference a specific tool or technology they use.

Stack token rules:

• Reference the tool only if you have a take on it. "I see you use HubSpot" is observation. "Most HubSpot teams at your scale outgrow the native sequencing" is insight
• Stack data from job postings is more reliable than stack data from BuiltWith (which only captures web-facing tools)

Where Tokens Go in the Email

Token placement map

[Subject line] — optional token (signal reference, topic)

[OPENER: Token goes here] — 15-25 words
  The personalized first line. This is where most tokens land.
  Pattern: {token} + {implication}

[BODY: Template] — 30-50 words
  The value prop, proof point, or credibility bridge.
  This section is the same for all prospects in the batch.
  (Some teams add a second token here for Tier 1 ABM)

[ASK: Template] — 10-15 words
  The CTA. Same for all prospects.
  "Worth a 15-minute look?" / "Open to a quick teardown?"

[SIGNATURE]

Placement rules

• One token in the opener is the minimum. This is the Level 2 standard. One personalized data point in the first line
• Two tokens (opener + body) is Level 3. The opener references their situation. The body references a peer or metric specific to their segment. This is the sweet spot for $30-100K ACV deals
• Don't personalize the CTA. The ask should be consistent and tested. Personalizing "Would you prefer a teardown of your specific outbound setup?" sounds nice but makes A/B testing impossible and doesn't measurably improve reply rates
• Don't personalize the subject line with merge tags. "{first_name}, quick question" is the most ignored subject line pattern. Subject lines should be topic-driven, not name-driven. See cold-email-subject-lines skill

Personalization That Backfires

Not all personalization is good personalization. Some types actively hurt reply rates.

The creep test

Before including a personalization token, ask: "Would the prospect feel surveilled or flattered?" If the answer is "surveilled," don't include it. Professional data (posts, talks, company news, hiring) is fair game. Personal data (education, family, hobbies, activity on other people's posts) is not.

Personalization at Scale

The batch personalization workflow

1. Build prospect list (200 prospects)
2. Segment by shared attribute (stage, vertical, signal type)
3. Write one template per segment (Level 1 personalization)
4. For each prospect, add one token (Level 2)
   - Research source: LinkedIn profile + one additional source
   - Time budget: 2-3 minutes per prospect
   - If no good token found: use segment-level token (Level 1 fallback)
5. QA: spot-check 10-20% for accuracy and tone
6. Load into sequencing tool

Scaling rules

• Level 2 at 200 prospects = 7-10 hours of research + personalization. This is approximately 1 SDR-day. Decide if that's worth 1.5-2x the reply rate vs Level 1
• AI personalization (see bulk-personalization-with-ai skill) can compress Level 2 to 75 minutes per 200 prospects. Use it when available
• Don't try to do Level 3 (custom line) at 200 prospects. At 5-8 min each, that's 17-27 hours. Either reduce the list to 50-100 or use AI to generate custom lines at scale
• Level 1 is always the fallback. If research doesn't turn up a usable token for a prospect, they get the segment-level template. A clean Level 1 email beats a forced, weak Level 2 token

Personalization by Email Position

Position rules

• Email 1 carries the personalization weight. This is where the prospect decides if the sequence is real outreach or a blast. Invest the token here
• Email 2 should use a different token type. If Email 1 used a signal token, Email 2 uses a metric or company token. Variety signals real thought, not a one-trick template
• Breakup emails don't need personalization. The brevity (30 words) is the hook. Adding a personalized line to a breakup makes it longer and weaker

Measuring Personalization Impact

A/B testing

Measurement rules

• Minimum 100 prospects per variant. Below 100, results are noise
• Measure positive reply rate, not just total reply rate. Aggressive personalization can increase replies but also increase "stop emailing me" responses
• Track cost per meeting by personalization level. Level 3 costs 3x more rep time than Level 2. Does it produce 3x the meetings? If not, Level 2 is the better ROI

Pre-Send Checklist

• [ ] Every email has at least one personalized token (Level 2+)
• [ ] The token answers "why you, why now" for the specific recipient
• [ ] The token is based on verifiable, professional data (not personal)
• [ ] The token is from the last 90 days (signals) or 30 days (content)
• [ ] No fabricated or guessed data in the token
• [ ] The personalization level matches the ACV and prospect count (see selection table)
• [ ] Each email in the sequence uses a different token type
• [ ] The personalized line is 15-25 words (not a paragraph)
• [ ] The personalization connects to the email's purpose (not decoration)
• [ ] Level 2+ emails have been spot-checked for accuracy (10-20% of batch)

Anti-Pattern Check

• Merge tags are called "personalization." {first_name} and {company} are not personalization. They're mail merge. Every cold email tool does this. Real personalization references something specific about the recipient
• Same token type on every email. Signal reference on Email 1, signal reference on Email 2, signal reference on Email 3. Vary the token types. Different data point, different perspective per email
• Personalization that's just a compliment. "Great post on LinkedIn!" is a compliment. "Your point about attribution being a data problem matches what we see at Series B teams" is personalization that connects to a problem. Connect, don't compliment
• Spending 15 minutes per email on a $10K ACV prospect. The math doesn't work. At $10K ACV and 15% close rate, that prospect is worth $1,500 in expected value. Don't spend $20 of rep time personalizing one email. Use Level 2 (2-3 min) and save Level 4 for $100K+ deals
• Personal data in cold outbound. Education, family, hobbies, liked posts, personal milestones. This feels invasive, not thoughtful. Stick to professional signals: posts they wrote, company news, job postings, conference talks
• Forced personalization when there's nothing to personalize. Sometimes the prospect has no LinkedIn posts, no recent company news, and no detectable signals. Use a segment-level template (Level 1) rather than forcing a weak token. "I see you're in sales at a SaaS company" adds nothing
• Over-personalized emails. A 90-word email where 60 words are personalization and 30 words are the pitch. The personalization overwhelms the message. One line (15-25 words) is enough. The rest of the email is the value prop and the ask