Clay logo, go to homepage

Search Marketplace

Browse all Skills
ResearchSkill

Kill Or Keep

Judges every live outbound campaign against pre-set reply-rate thresholds and returns a kill, scale, or keep verdict for each one.

Matteo Fois
Matteo Fois

Co-Founder at Kinetyca

About this Skill

When to use it

Use this Skill when you want to decide which live outbound campaigns to kill, scale, or keep running, based on positive reply rate rather than opens or raw reply counts. It reads campaign, sender, and A/B test data through Clay, a connected sending tool, or an export file, then delivers a verdict report you can act on one campaign at a time. Avoid using it for writing campaign copy, building lists, managing campaigns directly, or diagnosing deliverability issues like DNS and authentication.

What it delivers

You get a verdict report that flags risky senders and calls each campaign fix-deliverability, kill, scale, keep-going, or too-small-to-judge, plus A/B test calls only when the sample size can support a winner. Nothing changes until you approve a specific action for a specific campaign.

SKILL.md preview

Sign in to install and view the complete SKILL.md

Illustrative preview. Sign in for the actual file.

How someone would prompt this

Audit all my active outbound campaigns and tell me which ones to kill, scale, or keep running based on positive reply rate.

Inputs & outputs

What someone provides before the Skill runs, and the reviewable artifacts it returns.

Inputs

  • 01Campaign data source
    • Where your campaigns live: Clay campaigns, a connected sending tool, or an export file
    • optional sender-level export for deliverability checks
    • optional per-variant file for A/B test verdicts.
  • 02Judging scope
    • Which campaigns to judge, from all active ones to specific names
    • the market tier for each campaign, since kill and scale thresholds depend entirely on it.
  • 03Thresholds and rules
    • Choose kill and scale reply-rate floors per tier, or use the author's borrowed defaults
    • set the minimum days before a kill
    • pick which reply categories count as positive
    • set the bounce rate that flags a sender.
  • 04Timing settings
    • The first send date for each campaign, if not read automatically
    • the timezone used for date boundaries (defaults to UTC).

Outputs

  • 01Verdict boardShows every campaign's verdict (kill, scale, keep going, fix deliverability, or too small to judge) along with the numbers and rule behind it, so you know exactly what to act on.
  • 02Sender guardrailsFlags any sender whose bounce rate breaches your guardrail, so you can catch a deliverability problem before blaming the copy.
  • 03Variant callsCalls each supplied A/B test a winner, no-difference, or not-enough-data, only when the sample size can actually support that call.
  • 04Step funnelShows how many replies each sequence step produces, so you can see which step is doing the work.
  • 05Coverage lineShows how many campaigns got a verdict, and how many landed in too-small, too-soon, or unmeasured categories, so you can see how complete the audit was.

Representative output

Verdict board

Thresholds: author's defaults (borrowed). Positive = "Interested", "Meeting request". Bounce guardrail 5% (borrowed).

CampaignTierSentPositivesRateVerdictRule that fired
Northwind - ops leadersMid-market3,12000 per 3,120Fix deliverability firstwould be cold-dead, but one sender is flagged and the 2,080 healthy sends are under the 2,500 window
Adatum - CFOsMid-market8,00021 per 4,000Kill, chronicpast the 7,500 chronic sample and below 1 per 3,500
Contoso - foundersSmall SMB1,48041 per 370Scaleat or above 1 per 600 with 4 positives
Fabrikam - RevOpsMid-market5,60031 per 1,867Keep goinghas positives, not yet past the 7,500 chronic sample
Litware - finance leadersMid-market4,800421 per 114Scaleat or above 1 per 1,000 with 42 positives
Tailspin - IT directorsEnterprise1,90011 per 1,900Too small to judgebelow the 4,000 window with fewer than 3 positives
Blue Yonder - COOsMid-market3,00000 per 3,000Too soon to judgefirst send 9 days ago; kills wait 21 days, 12 days left

Sender guardrails

SenderCampaignSentBouncesBounce rateFlag
alex@northwind-mail.exampleNorthwind - ops leaders1,040838.0%Fix deliverability
sam@contoso-hq.exampleContoso - founders74091.2%none

Variant calls

CampaignVariantsSends per armPositive rateDays since launchVerdict
Litware - finance leadersPain opener vs Proof opener2,400 / 2,4001.25% / 0.50%28Pain opener wins: 2.5x lift, needs 1,000 per arm at a 0.5% loser rate (scaled), has 2,400; 30 positives
Contoso - foundersShort opener vs Long opener740 / 7400.54% / 0.14%26Not enough data yet: 4 positives vs 1. At a 0.14% loser rate a 2x lift needs about 3,700 sends per arm (scaled)
Fabrikam - RevOpsCase study vs Question610 / 5900.33% / 0.34%24No difference detected: under 1.2x apart

Step funnel

CampaignStep 1Step 2Step 3Most replies from
Fabrikam - RevOps61%28%11%Step 1

Coverage line

11 active campaigns in scope · 5 judged · 5 too small or too soon · 1 unmeasured (no reply categories returned) · variant verdicts for 3 of 4 tests (1 row missing a launch date).

How the Skill runs

Follow the steps in order. Each row includes the full instruction.

01
Check campaign data access

The skill checks whether it can read your campaign data from Clay, a connected sending tool, or an export file, and stops if none of these are available.

Step
02
Set thresholds before results

You choose your campaign source, scope, market tiers, kill and scale thresholds, positive-reply categories, bounce guardrail, and minimum kill age before any numbers are read.

Step
03
Decide what can be judged

The skill determines which verdict types your data supports (sender flags, campaign verdicts, step funnel, or variant calls) and tells you which ones it will run.

Step
04
Triage campaigns by size

The skill pulls totals for every campaign in scope, sorting out which are too small to judge yet versus which need a full read, and stops if no campaigns are in scope.

Step
05
Review and approve the plan

The skill shows the thresholds, positive categories, and how many campaigns will be read, then stops and waits for your approval before reading any results.

Step
06
Read campaign numbers

The skill pulls sent counts, positive replies, bounces, sender stats, and step-by-step funnel data for every queued campaign, marking unclear results as unmeasured rather than guessing zero.

Step
07
Apply verdict rules

The skill applies fixed rules in order, first flagging risky senders, then calling each campaign a kill, scale, keep-going, or too-small verdict, and calling any supplied A/B test only when the sample size can support a winner.

Step
08
Deliver the report

The skill writes a report file listing deliverability flags first, then kills, then scales, then everything else, plus a coverage summary of what was and wasn't judged.

Step
09
Approve and act on one campaign

Only if you ask, the skill offers to end a variant test or pause a campaign through your connected tool, one campaign at a time, and stops to wait for your yes before taking that specific action.

Step

Connections required

No connections listed.

Install Clay and run it

Sign in once, install the Skill, then run the prompt.

  1. 1Install ClayUse the connector or official plugin in your coding-agent environment.
  2. 2Sign up or log inAuthorize Clay access and return when the connection is ready.
  3. 3Install the SkillRun the one-line install command from the Add this Skill card.
  4. 4Run the SkillPaste the example prompt and review the result.

Version history

Version 2From submitted SKILL.mdCurrent

Related Skills

Find Recent Company DevelopmentsBuilds a Clay workflow that turns a list of companies into dated, sourced briefs on new leadership, M&A, news, and transformation activity inside a time window you set.View Skill →Swap TestChecks how much of a cold email is generic template versus specific fact, then rewrites the template sentences using only sourced details.View Skill →Pricing Page SignalChecks whether a company publishes public self-serve pricing and extracts its plan tiers, prices, and pricing model into a structured record for each domain.View Skill →

Related searches

Research skills