agentleFS
manypicommanypicom/sales-skills/skills/outbound-campaign-ops/SKILL.md

outbound-campaign-ops

Skill1 starsChanged 2 months agoMITView on GitHub

Plan, launch and run cold email campaigns — structure, send windows, limits, testing and the launch checklist. Use whenever the user mentions running a campaign, launching outreach, campaign setup, send windows, daily limits, A/B testing emails, campaign structure, or scaling outbound. Also use before any campaign goes live, and when someone is running several campaigns and losing track of what's working.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 7 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Never mix segments in a campaign.

    line 29
  2. Distinctive:

    The inbox cap is the ceiling, and it's warmup-aware: If three campaigns share an inbox, their combined volume must respect the inbox's cap, …

    line 52
  3. Distinctive:

    Run this every time. It takes fifteen minutes and it's the difference between a campaign and an incident.

    line 58
  4. Distinctive:

    Never launch a campaign to the full list.

    line 94
  5. Distinctive:

    50 leads: Check deliverability, bounces and rendering.

    line 96
  6. Distinctive:

    Scale: if the numbers hold. Add inboxes rather than raising per-inbox volume.

    line 98
  7. Very specific:

    Read ten actual replies. Metrics tell you something changed.

    line 128
  8. Distinctive:

    When a campaign works, scale in this order:

    line 132
  9. Very specific:

    Don't keep sending through a bounce spike: Stop, fix, resume slowly. See bounce-recovery.

    line 153

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Outbound Campaign Operations

The writing and the list get the attention. The operations decide whether the programme survives its third month.

Most outbound failures at scale are operational: a campaign launched to an unverified list, a limit raised because a quarter was ending, three campaigns sharing an inbox and blowing its cap, or a test that ran with two variables changed at once and taught nobody anything.

The four objects

Keeping these separate is what makes outbound manageable past one campaign.

Inbox       A mailbox you send through. Carries the daily cap,
            the warmup schedule and the health.
Sequence    The email steps and timing. Reusable.
Campaign    Binds one sequence to one inbox. Carries the send
            window, the daily limit and the tracking settings.
Enrollment  A lead placed into a campaign.

The value of the split is that a sequence that works can be reused across inboxes, segments and clients without being rewritten, and an inbox's health is a property of the inbox rather than of whatever campaign happens to be using it.

One campaign per segment

Never mix segments in a campaign. It's tempting because it's fewer things to manage, and it destroys your ability to learn anything.

Campaign: DACH SaaS 20-200 — VP Marketing — hiring signal
Campaign: DACH SaaS 20-200 — Founder — funding signal
Campaign: UK agencies 5-50 — Owner — no signal

Each has its own sequence, its own angle, and its own metrics. When one produces a 6% reply rate and another produces 0.8%, you know something. A blended campaign at 2.4% tells you nothing.

Send windows and limits

SettingSensible default
DaysTuesday to Thursday, sometimes Monday and Friday
Hours08:00 to 17:00 in the recipient's timezone
Per-inbox daily cap150 at full warmup, lower during ramp
Per-campaign daily capSet below the inbox cap when sharing
StaggerRandomised gaps, never a burst
New leads per daySteady. Avoid spikes

Two rules that matter more than the numbers:

The inbox cap is the ceiling, and it's warmup-aware. If three campaigns share an inbox, their combined volume must respect the inbox's cap, not each get the full allowance. Systems that don't enforce this quietly burn mailboxes.

Recipient timezone, not yours. Sending to New York at 09:00 CET means arriving at 03:00 local, which lands under thirty overnight emails by the time they wake.

The launch checklist

Run this every time. It takes fifteen minutes and it's the difference between a campaign and an incident.

LIST
[ ] Verified — invalid, no_mx and disposable removed
[ ] Deduplicated against existing leads, customers and open deals
[ ] Checked against the suppression list
[ ] 20 random records manually checked for ICP fit
[ ] Role accounts decided on (in, out, or separate campaign)

COPY
[ ] Test-rendered on 20 real leads, all 20 read
[ ] Every merge variable has a natural fallback
[ ] Links checked, and kept to one or none in step 1
[ ] Unsubscribe mechanism present and working
[ ] Read aloud

INFRASTRUCTURE
[ ] SPF, DKIM, DMARC passing on the sending domain
[ ] Inbox warmed, or warmup schedule active
[ ] Custom tracking domain, not a shared one
[ ] Send window set to recipient timezone
[ ] Daily caps set below the inbox ceiling

PROCESS
[ ] A named person owns the replies
[ ] Replies land somewhere monitored
[ ] Reply-stop tested with a real address
[ ] Bounce threshold and auto-pause configured
[ ] Success metric agreed before launch

The reply-stop test is the one people skip. Send yourself a step-one email, reply to it, and confirm step two never arrives.

Launch small

Never launch a campaign to the full list.

  1. 50 leads. Check deliverability, bounces and rendering. Stop if bounces exceed 2%.
  2. 200 leads over a few days. Now the reply rate means something.
  3. Scale if the numbers hold. Add inboxes rather than raising per-inbox volume.

Most problems are visible in the first 50 and expensive at 5,000.

Testing properly

The rules are unremarkable and routinely broken.

  • One variable at a time. Subject or first line or CTA. Changing two teaches nothing.
  • Enough volume. Reply-rate differences need several hundred per variant to distinguish from noise. At 100 sends per variant, a 2% versus 3% result is a coin flip.
  • Test one thing at a time across the programme, not five tests running simultaneously on overlapping lists.
  • Test the big things first. Audience, then offer, then first line, then subject. Subject-line tests are the most run and the least consequential.
  • Give it a full cycle. A test stopped after step one measures the opener, not the sequence.

What's worth testing, roughly in order of effect: the segment, the angle, the researched personalisation, the CTA, the sequence length, then the subject line.

The weekly review

Twenty minutes, same time each week, per campaign:

CheckThreshold
Bounce rateOver 2%, pause and investigate
Complaint rateOver 0.1%, pause immediately
Open rate trendSudden drop means deliverability, not copy
Reply rateCompare against the other campaigns, not an industry number
Positive reply shareThe metric that actually matters
Inbox healthAny mailbox degrading
Replies unansweredShould be zero

Read ten actual replies. Metrics tell you something changed; replies tell you what. See outbound-metrics-benchmarks.

Scaling

When a campaign works, scale in this order:

  1. More leads in the same segment, if the segment is deep enough.
  2. More inboxes, keeping per-inbox volume constant. See inbox-rotation-scaling.
  3. A new segment, with its own campaign and its own angle.
  4. Never: raise per-inbox volume above the safe ceiling because a target is close.

That last line is where most programmes break. The pressure to hit a number at quarter end produces a volume increase, the volume increase produces a reputation problem, and the reputation problem costs the next quarter.

Running multiple clients or brands

For agencies, separation is the whole discipline: separate sending domains, separate inboxes, separate suppression lists, separate everything. One client's bounce problem must not touch another's deliverability, and one client's leads must never appear in another's campaign. See agency-multi-client-outbound.

Don't do these

  • Don't launch on a Friday. Nobody's watching the bounce rate on Saturday.
  • Don't mix segments in one campaign.
  • Don't run several campaigns through one inbox without a combined cap.
  • Don't raise limits to hit a number.
  • Don't test two variables at once.
  • Don't launch without a named reply owner.
  • Don't keep sending through a bounce spike. Stop, fix, resume slowly. See bounce-recovery.

Running it

Most of this checklist is only reliable when the platform enforces it rather than the operator remembering. The failures happen on the busy weeks.

ManyPI separates inbox, sequence, campaign and enrollment as distinct objects, and its scheduler checks every condition before each send: campaign active and inside the send window, inbox under its warmup-aware cap, campaign under its own daily limit, address not suppressed or undeliverable, and the lead hasn't replied. Sends are staggered rather than fired in a burst, and a bounce guard pauses sending automatically if the 24-hour bounce rate crosses the threshold. Campaigns, sequences, enrollment and stats are all available over MCP, so a campaign can be built and launched in conversation:

claude mcp add --transport http manypi https://mcp.manypi.com/mcp -s user

The launch checklist still applies. No scheduler checks whether the list is the right people.

Related skills

email-sequence-design, email-deliverability, outbound-metrics-benchmarks, inbox-rotation-scaling, reply-handling, agency-multi-client-outbound, bounce-recovery

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.