name: email-sequences description: Design automated email sequences and drip campaigns. Use when building onboarding flows, nurture sequences, re-engagement campaigns, or follow-up automation. license: MIT

Email Sequences

Design and run automated email sequences (drip campaigns) that convert without fatiguing recipients or damaging sender reputation.

When to use this skill

• Building an onboarding, nurture, re-engagement, or winback sequence
• Deciding how many emails to include and how far apart to space them
• Setting up entry triggers and exit conditions for automated flows
• Adding branching logic based on opens, clicks, replies, or behavior
• Running A/B tests within a sequence (subject lines, content, timing)
• Diagnosing why a sequence has declining engagement or rising unsubscribes
• Preventing overlap between multiple sequences hitting the same contact

Related skills

• onboarding-emails - deep dive on welcome and activation sequences specifically
• cold-outreach - cold email follow-up sequences (different rules, different infrastructure)
• email-copywriting - writing emails people actually read
• ab-testing - testing methodology beyond sequence-specific tests
• suppression-lists - managing bounces, complaints, and opt-outs
• bounce-handling - processing delivery failures and retry strategies
• rate-limiting - volume controls that protect reputation
• email-compliance - CAN-SPAM, GDPR, CASL, unsubscribe requirements
• sender-reputation - monitoring and recovering reputation

Sequence types and recommended structure

Different goals require different sequence shapes. Here are the common types with proven structures.

Onboarding / welcome

Goal: get a new user to their first success moment.

Length: 3-5 emails over 7-10 days. Welcome emails get 60%+ open rates - the rest of the sequence won't match that. Front-load your most important content.

Exit when: user completes the activation milestone (not just opens an email).

Lead nurture

Goal: move a prospect from awareness to purchase readiness.

Length: 5-8 emails over 2-4 weeks. Space emails 2-4 days apart. Never more than 3 emails per week.

Exit when: prospect converts (signs up, books demo, makes purchase) or replies.

Re-engagement

Goal: revive contacts who stopped opening or clicking.

Length: 2-3 emails over 10-14 days. Shorter is better - if 3 emails don't re-engage them, more won't either.

Exit when: contact engages (opens, clicks), or after the final email. If no engagement after the sequence, move to suppression or reduce to quarterly cadence.

Winback

Goal: recover a cancelled customer or lost deal.

Length: 3-4 emails over 30 days. Wider spacing - they just left, so don't be aggressive.

Exit when: customer re-activates, replies, or explicitly declines.

Upsell / cross-sell

Goal: expand an existing customer relationship.

Length: 2-3 emails. Only trigger when usage data actually supports the upsell. Untargeted upsells annoy people fast.

Exit when: customer upgrades, dismisses, or replies.

Timing and cadence

Spacing between emails

The right gap depends on urgency and sequence type:

Never send more than 3 emails per week to the same contact across all sequences combined. This is the single most important cadence rule. Exceeding this drives unsubscribes and spam complaints regardless of how good the content is.

Send timing

• Weekdays outperform weekends for B2B. Tuesday, Wednesday, and Thursday are the strongest days.
• B2C is more flexible - weekends can work for consumer products, especially Saturday morning.
• Send during business hours in the recipient's timezone. 9 AM - 3 PM local time gets the best open rates.
• Avoid Monday morning and Friday afternoon. Inboxes are either overloaded or already mentally checked out.

Fatigue scoring

Track engagement signals per contact and adjust cadence dynamically. A simple fatigue model:

Fatigue score components:
- Send frequency (0-30 points): >5/week = 30, >3/week = 20, >1/week = 10
- Monthly volume (0-15 points): >20/month = 15, >10/month = 10, >5/month = 5
- Bounces (0-20 points): each bounce = 10 points (cap at 20)
- Complaints (0-25 points): each complaint = 15 points (cap at 25)
- Engagement decay (0-10 points): >30 days since last engagement = 10

Thresholds:
- Score >= 70: stop sending
- Score >= 40: reduce frequency
- Score < 40: safe to send

When the fatigue score hits "reduce frequency," double the gap between sequence emails. When it hits "stop sending," pause the sequence and move the contact to a re-engagement flow instead.

Entry triggers

Event-based triggers (best)

Start a sequence when a specific event occurs:

• Signup completed - onboarding sequence
• Trial started - trial nurture sequence
• Cart abandoned - recovery sequence (send within 1 hour)
• Feature milestone reached - upsell sequence
• Subscription cancelled - winback sequence
• Inactivity threshold - re-engagement sequence (e.g., no login for 14 days)

Event triggers are the most reliable because they're based on something the contact actually did (or stopped doing).

Segment-based triggers

Enroll contacts when they match specific criteria:

Segment: "Trial users who used Feature X but not Feature Y"
Filter:
  - lifecycle_stage = "trial"
  - AND event_count("feature_x_used", last 7 days) > 0
  - AND event_count("feature_y_used", last 7 days) = 0

Segment-based triggers are powerful for targeting specific user profiles but require clean data. Evaluate segments on a schedule (daily or hourly), not continuously, to avoid race conditions.

Manual enrollment

For sales-driven sequences where a human decides to enroll a prospect. Always check suppression status before enrollment.

Trigger rules

• One trigger per sequence. If multiple events should start the same email flow, route them through a single entry point that deduplicates.
• Deduplicate enrollments. A contact should only have one active run per sequence. If the trigger fires again while they're already in the sequence, ignore it.
• Check suppressions at enrollment. Don't enroll contacts who have unsubscribed, complained, or hard-bounced.

Exit conditions

Exit conditions determine when to stop sending to a contact before the sequence finishes naturally. Get these wrong and you'll send irrelevant emails that damage trust and deliverability.

Required exit conditions

Every sequence needs these:

1. Goal achieved. The contact did the thing the sequence was designed to produce (purchased, activated, booked a demo). This is the happy path exit.
2. Explicit opt-out. The contact unsubscribed or replied asking to stop. Honor immediately - not after the next scheduled email.
3. Hard bounce. The email address doesn't exist. Remove from the sequence and suppress.
4. Spam complaint. Stop all email to this contact, not just the current sequence.
5. Reply received. In most cases, a reply means the conversation should move to a human or a different flow. Continuing the automated sequence after a reply looks robotic and damages trust.

Recommended exit conditions

1. Entered a higher-priority sequence. If a lead nurture contact starts a free trial, they should exit the nurture sequence and enter the onboarding sequence instead.
2. Fatigue threshold crossed. If the contact's engagement has been declining across all email, pause rather than keep sending.
3. Negative signal detected. If inbound reply classification detects intent like "objection" or "not_now," exit the sequence and route appropriately.

Implementing exit conditions

Check exit conditions at two points:

• At enrollment - don't start a sequence for a suppressed contact
• Before each step executes - re-evaluate conditions before every send, not just at enrollment

This matters because a contact might reply between step 2 and step 3. If you only check conditions at enrollment, step 3 fires anyway.

Before executing step N:
1. Is the contact suppressed? -> exit
2. Has the contact achieved the goal? -> exit
3. Has the contact replied? -> exit (route to human/different flow)
4. Is the contact in a higher-priority sequence? -> exit
5. Does the fatigue score say "stop"? -> exit
6. Has the contact complained about any email? -> exit
All clear -> execute step N

Branching logic

Branching transforms a linear sequence into an adaptive flow that responds to what each contact does.

Behavioral branches

Branch based on what the contact did (or didn't do) in previous steps:

{
  "type": "branch",
  "config": {
    "conditions": [
      { "field": "lastLoginDaysAgo", "operator": "lt", "value": "7" }
    ],
    "onMatch": { "nextStep": 5 },
    "onNoMatch": { "nextStep": 4 }
  }
}

Common behavioral branches:

Segment-based branches

Branch based on contact attributes, not just behavior:

If contact.lifecycle_stage == "enterprise":
  -> send enterprise case study
Else:
  -> send SMB case study

Time-based branches

Branch based on when the contact entered or how long they've been in the sequence:

If days_since_enrollment > 30 and no_engagement:
  -> move to re-engagement track
Else:
  -> continue nurture

Keep branching simple

Every branch doubles the paths you need to test and maintain. In practice:

• 1-2 branch points per sequence works well
• 3+ branch points creates complexity that rarely improves results enough to justify the maintenance cost
• If you need heavy branching, you probably need separate sequences for separate segments instead

A/B testing within sequences

What to test

Test one variable at a time within a single step. The most impactful variables, in order:

1. Subject line - highest impact, easiest to test
2. Send time - morning vs. afternoon, different days
3. CTA - button text, placement, number of CTAs
4. Content length - short vs. long
5. Content approach - educational vs. social proof vs. direct pitch

How to test

For each step you want to test, create variants with different weights:

Step 3 - Feature highlight:
  Variant A (50%): "3 ways to use [feature]" (educational)
  Variant B (50%): "How [company] increased revenue 40% with [feature]" (social proof)

Use deterministic assignment - the same contact should always see the same variant if re-evaluated. Hash-based bucketing (hash of experiment ID + contact email) ensures consistency without storing assignments upfront.

Statistical significance

Don't call a winner too early. You need enough data:

• Minimum sample size: at least 200-300 sends per variant before drawing conclusions
• Significance threshold: p-value < 0.05 (95% confidence)
• Run time: let the test run for at least one full cycle through the step (all contacts in the current cohort should have received it)

A two-proportion z-test works for comparing conversion rates between variants. If your control converts at 5% and the variant converts at 7%, you need roughly 1,500 sends per variant to detect that difference with 95% confidence.

What to measure

Don't optimize for opens alone. Measure by step position:

Sequence performance metrics

Per-step metrics

Track these for every step in the sequence:

Sequence-level metrics

Drop-off analysis

The most actionable sequence metric is where people stop engaging. Plot open/click rates by step:

Step 1: 45% open, 8% click
Step 2: 38% open, 5% click
Step 3: 35% open, 4% click   <- normal decay
Step 4: 18% open, 1% click   <- problem step - content, timing, or fatigue
Step 5: 15% open, 0.8% click

A steep drop between specific steps means something is wrong with that email or the gap before it. A gradual decline across all steps means the sequence is too long.

Attribution

Tie sequence sends to business outcomes. Each email in the sequence is a touchpoint, and when a contact converts, attribute the conversion to the steps that preceded it.

Common attribution models for sequences:

• Last touch: credit the final email before conversion. Simple but undervalues earlier nurturing steps.
• First touch: credit the first email. Useful for measuring which sequences initiate journeys that eventually convert.
• Linear: equal credit to every step the contact received. Best default for sequence optimization.
• Time decay: more credit to recent touches. Good for long sequences where later steps are more directly influential.

Preventing sequence overlap

When a contact is eligible for multiple sequences, you need rules to prevent them from getting buried in email.

Priority system

Rank your sequences by priority:

Priority 1: Transactional (receipts, password resets) - always send
Priority 2: Onboarding (new user activation) - high priority
Priority 3: Trial expiration (time-sensitive) - high priority
Priority 4: Nurture (education, relationship) - medium priority
Priority 5: Re-engagement (reviving inactive) - low priority
Priority 6: Upsell/cross-sell - low priority
Priority 7: Marketing newsletter - lowest priority

When a contact qualifies for a higher-priority sequence, either:

• Pause lower-priority sequences (resume when the higher-priority one finishes)
• Exit lower-priority sequences (re-evaluate enrollment later)

Global send budget

Regardless of how many sequences a contact is in, cap the total sends per contact:

• Maximum 3 emails per week across all sequences combined
• Maximum 10 emails per month across all sequences combined
• Minimum 24-hour gap between any two emails to the same contact

When a sequence step is due but the contact has hit their send budget, delay it - don't skip it. Skipping creates holes in the sequence logic.

Cooldown enforcement

Enforce cooldowns at the infrastructure level, not in sequence logic. The sequence shouldn't need to know about other sequences - it just sends, and the policy layer blocks if cooldown hasn't elapsed.

{
  "status": "blocked",
  "reason": "cooldown",
  "detail": "Contact received a message 18 hours ago. Cooldown is 48h.",
  "nextEligibleAt": "2026-03-31T08:00:00Z"
}

The sequence engine reschedules the step for nextEligibleAt and continues normally.

Sequence architecture

State management

Each sequence run needs to track:

• Run ID - unique identifier for this contact's run through this sequence
• Current step - which step is next
• Status - active, paused, completed, exited
• Context - data collected during the run (which branches taken, engagement data)
• Enrollment time - when the contact entered

The key architectural decision: the sequence engine should be stateful, but email sending should be stateless. The engine tracks where each contact is in the sequence. Each individual send goes through the same policy evaluation as any other email - deduplication, suppression, rate limiting, cooldown.

Step types

A well-designed sequence engine supports these step types:

Example journey definition:

{
  "name": "Trial nurture",
  "triggerEvent": "trial.started",
  "steps": [
    {
      "type": "send",
      "position": 1,
      "config": {
        "templateId": "trial-welcome",
        "payload": { "subject": "Welcome to your trial" }
      }
    },
    {
      "type": "delay",
      "position": 2,
      "config": { "delayMinutes": 4320 }
    },
    {
      "type": "branch",
      "position": 3,
      "config": {
        "conditions": [
          { "field": "hasCompletedSetup", "operator": "eq", "value": true }
        ],
        "onMatch": { "nextStep": 5 },
        "onNoMatch": { "nextStep": 4 }
      }
    },
    {
      "type": "send",
      "position": 4,
      "config": {
        "templateId": "trial-setup-help",
        "payload": { "subject": "Need help getting started?" }
      }
    },
    {
      "type": "send",
      "position": 5,
      "config": {
        "templateId": "trial-power-features",
        "payload": { "subject": "3 features most teams discover in week 2" }
      }
    },
    {
      "type": "delay",
      "position": 6,
      "config": { "delayMinutes": 7200 }
    },
    {
      "type": "send",
      "position": 7,
      "config": {
        "templateId": "trial-ending-soon",
        "payload": { "subject": "Your trial ends in 3 days" }
      }
    },
    {
      "type": "end",
      "position": 8
    }
  ]
}

Deduplication

Each contact should only have one active run per sequence. If the trigger event fires again while a run is active, the second run should be rejected. This prevents the most common sequence failure: a customer getting duplicate emails because multiple instances of an automation detected the same condition.

Use a dedupe key composed of journeyId + contactEmail and check for active runs before creating a new one.

Reply handling

When a contact replies to a sequence email, the reply should be classified by intent and routed accordingly:

Continuing to send automated emails after someone has replied is the fastest way to get spam complaints. Even if the reply is just "thanks," pause the sequence and evaluate.

Common mistakes

1. No exit conditions beyond sequence completion

The sequence has 7 steps, so every contact gets all 7 emails regardless of what happens. A contact who purchased after step 2 still gets step 3-7 ("here's why you should buy"). This is the most common sequence mistake and the most damaging to trust.

Fix: Implement goal-based exits. Check before every step whether the contact has already achieved the sequence goal.

2. Ignoring replies

Contact replies "Not interested right now" and still gets the next 4 emails on schedule. Nothing says "automated" louder than ignoring a direct response.

Fix: Classify inbound replies by intent and exit or pause the sequence when a reply is received.

3. No cross-sequence coordination

A contact is in the onboarding sequence, the trial expiration sequence, AND the feature education sequence simultaneously. They get 3 emails on Tuesday.

Fix: Implement a global send budget per contact. Cap at 3 emails/week across all sequences. Use sequence priority to determine which emails get delayed when the budget is hit.

4. Testing on opens instead of conversions

You A/B test subject lines and pick the variant with higher opens. But the high-open variant had clickbait subjects that led to lower conversions. Opens are a proxy metric, not the goal.

Fix: Measure the metric that matters for the sequence goal - conversion rate, activation rate, revenue per contact.

5. Sequences that are too long

A 12-email nurture sequence running over 8 weeks. By step 8, open rates are 5% and you're just training spam filters. Engagement data consistently shows that most reply/conversion value comes from the first 4-5 emails.

Fix: Start with 3-5 emails. Add steps only when data shows contacts are still engaging at that point in the sequence.

6. Same content to everyone

A single nurture sequence for all leads regardless of industry, company size, or stated interest. The content is generic enough to be irrelevant to everyone.

Fix: Use segment-based branching or separate sequences for meaningfully different audiences. Two well-targeted 4-email sequences beat one generic 8-email sequence.

7. No warmup for sequence volume

You build a 5-step sequence and enroll 10,000 contacts on day one. Even if the emails are great, sending 10,000 emails from a new template in the first hour triggers rate limits and spam filters.

Fix: Ramp enrollment gradually. Start with 100-200 contacts, monitor delivery and engagement, then increase by 2x every few days. See the email-warmup skill.

8. Sending during cooldown windows

The sequence engine doesn't know about the cooldown from yesterday's transactional email, so it fires step 3 six hours after a receipt email. The contact gets two emails in half a day.

Fix: Enforce cooldowns at the infrastructure level, not in the sequence. Every send - whether from a sequence, a transactional trigger, or a one-off campaign - goes through the same policy engine. The sequence should handle "blocked: cooldown" responses by rescheduling, not by skipping.

Checklist: launching a new sequence

• Sequence has a clear, measurable goal (not "engagement" - a specific conversion event)
• Entry trigger is defined and deduplication is in place
• Exit conditions cover: goal achieved, reply received, unsubscribe, bounce, complaint
• Each step has a minimum delay of 24 hours from the previous step
• Total emails per week per contact won't exceed 3 across all active sequences
• Suppression list is checked at enrollment AND before each step
• Reply handling is configured - replies exit or pause the sequence
• Fatigue scoring is active - contacts with high fatigue get reduced frequency or are paused
• A/B tests (if any) have enough expected volume for statistical significance
• Initial enrollment is ramped gradually, not all at once
• Step-level metrics are being tracked (delivery, open, click, unsubscribe, complaint)
• Drop-off analysis is set up to identify problem steps
• Sequence priority is set relative to other active sequences

References

• Mailchimp - What Is an Email Sequence - fundamentals and examples
• MailerLite - Email Cadence & Frequency Best Practices - data-backed timing guidance
• Omnisend - Email Automation - automation benchmarks (automations earn 16x more per send than broadcast)
• Moosend - Email Fatigue - fatigue signals and prevention
• Instantly - Email Sequence Troubleshooting - common technical failures
• ActiveCampaign - Email Marketing Benchmarks 2025 - industry benchmark data
• RFC 8058 - One-Click Unsubscribe (required for marketing sequences)
• Google Email Sender Guidelines - bulk sender requirements
• M3AAWG Best Practices - industry standards for responsible sending