compare-campaigns
Skill8 starsChanged 6 months agoMITView on GitHub
A/B compare two or more Smartlead campaigns side-by-side — sends, open rate, reply rate, positive reply rate, bounce rate, per-step performance. Identifies which campaign is winning and why. Use when the user wants to compare campaign versions, see if V2 beats V1, evaluate copy variants, or asks "which one is performing better".
In this repository11 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- audit-mailbox-configSkill
- check-deliverabilitySkill
- clone-campaignSkill
- create-campaignSkill
- launch-campaignSkill
- smartlead-onboardingSkill
- spam-testSkill
- upload-leadsSkill
- warmup-mailboxesSkill
- weekly-reportSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 89
If bounce rates differ significantly, that often masks a copy-quality conclusion.
- Very specific: line 99
If V1 bounces 8% on @google.com and V2 bounces 4% on @google.com using the same lead list, something is different about how V2's senders …
- Distinctive: line 143
If sample sizes are too small or differences are marginal:
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.
Files
Before running: verify exact flag names with
smartlead <command> --help. Flag shapes shown below are illustrative — confirm against the actual CLI.
/compare-campaigns
You're doing a head-to-head comparison of 2+ Smartlead campaigns to find the winner.
What you need from the user
- Campaign IDs to compare (typically 2, can be 3-4)
- Comparison axis — what variable is different? (copy, audience, sender pool, schedule)
- Optional: minimum sample size for statistical significance
Step 1 — Pull analytics for each campaign
for cid in $CAMP1 $CAMP2; do
smartlead campaigns analytics $cid --output json > "analytics-$cid.json"
doneConfirm both have meaningful sample sizes. Anything below ~200 sends per arm is too early to call.
Step 2 — Side-by-side headline metrics
HEAD-TO-HEAD
Campaign A (V1) Campaign B (V2)
─────────────────────────────────────────────────────────────────
Sends 1,287 1,295
Delivered 1,221 (94.9%) 1,243 (96.0%) ← B
Bounces 66 (5.1%) 52 (4.0%) ← B
Opens 412 (33.7%) 497 (40.0%) ← B
Replies 28 (2.29%) 41 (3.30%) ← B
Positive replies 7 (0.57%) 16 (1.29%) ← B ⭐
Meetings booked 2 5 ← B
─────────────────────────────────────────────────────────────────
Winner: Campaign B (V2)Mark each row with which arm wins (or "tied").
Step 3 — Per-step performance
Where in the sequence are replies coming from?
smartlead campaigns sequences <id> --output json --with-statsPER-STEP REPLY ANALYSIS
V1 V2
───────────────────────────────────────────────
Step 1 12 replies 18 replies ← V2 better hook
Step 2 9 replies 15 replies ← V2 better follow-up
Step 3 5 replies 7 replies ← roughly tied
Step 4 2 replies 1 reply ← V1 slightly betterThis tells you WHERE the V2 win comes from. Step 1 hook? Step 2 reframe?
Step 4 — Reply categorization breakdown
smartlead inbox list --campaign-id $CAMP1 --output json | jq -r '.[].category_id' | sort | uniq -cREPLY CATEGORIES
V1 V2
─────────────────────────────────────────────
Interested (1) 4 10 ← V2
Meeting Request (2) 2 5 ← V2
Information Request (5) 1 4 ← V2
Not Interested (3) 15 12 ← V1 (more rejects)
Wrong Person (7) 4 2 ← V1
Out Of Office (6) 2 8 ← V2 (different audience activity)Note: "Not Interested" being lower in V2 means it's getting clearer aspositive-or-negative — fewer "polite no thanks" replies.
Step 5 — Bounce comparison
If bounce rates differ significantly, that often masks a copy-quality conclusion. Higher bounces = less reach = fewer chances to win.
# Are the bounces from different recipient segments?
for cid in $CAMP1 $CAMP2; do
smartlead leads list $cid --status BOUNCED --output json | \
jq -r '.[] | .email' | awk -F@ '{print $2}' | sort | uniq -c | sort -rn | head -5
doneIf V1 bounces 8% on @google.com and V2 bounces 4% on @google.com using the same lead list, something is different about how V2's senders are interacting with Google's ESP — probably the AI ESP matching setting (see docs/learnings/esp-matching-analysis.md).
Step 6 — Statistical significance check
For positive reply rates (the metric that matters), check if the difference is real or noise.
Rule of thumb at typical cold email sample sizes:
- Difference of <0.3pp with ~1,000 sends/arm → noise, don't conclude
- Difference of 0.3-0.7pp with ~1,000 sends/arm → suggestive, run more
- Difference of >0.7pp with ~1,000 sends/arm → likely real
- Anything with <300 sends/arm → too early, keep sending
Step 7 — Diagnose WHY one won
Look at differences:
| Variable | V1 | V2 |
| Subject line | "Quick question about X" | "{first_name} — is X still a priority?" |
| Email 1 length | 95 words | 62 words |
| Sender pool | Mixed (Gmail + Outlook) | Gmail only |
| AI ESP matching | ON | OFF |
| Send time | 9am ET | 7am ET |
Hypothesize: which variable change likely drove the result?
Step 8 — Recommendation
RECOMMENDATION
──────────────────────────────────────────────
✓ V2 wins clearly on positive reply rate (1.29% vs 0.57%, 2.3x)
✓ Sample size is sufficient (~1,300 per arm)
✓ Step 1 + Step 2 drove the win — the new hook is working
NEXT STEPS:
1. PAUSE V1, route remaining V1 leads to V2 sequence
2. Use V2's subject line as the template for future campaigns
3. Try V3 testing: V2 copy + 5am ET send time (earlier wins often)
4. /clone-campaign V2 for the next audience segmentWhen you can't conclude
If sample sizes are too small or differences are marginal:
INCONCLUSIVE
─────────────────────────────────
V2 leads V1 by 0.2pp on positive replies, but with only 280 sends per arm,
this is within noise. Keep both running for another week, then re-compare.Reference
docs/learnings/esp-matching-analysis.md— bounce-rate differences by ESP routingdocs/learnings/settings-api-quirks.md— reply category IDs
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
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.

