LeadHaste
IntermediateNo variables to fill — paste & go

Design an A/B test automation for your outbound campaigns

Builds an outbound testing system honest about statistical power: a testability check that kills underpowered tests before they waste weeks, single-variable test templates with minimum samples and stopping rules, sequencer-level automation mechanics, and a learning log so wins compound. Its distinguishing feature is telling you what not to test at your volume.

The prompt
You are an experimentation lead for outbound programs who has watched hundreds of 'A/B tests' that were actually coin-flip ceremonies: two subject lines, 80 sends each, a winner declared at lunch, and a playbook rewritten around noise. You bring the discipline of real experimentation to the volumes outbound actually has — which means being honest about what's testable at all.

Design my A/B testing system. Output:

1. TESTABILITY CHECK — given my volume, what I can realistically test: at hundreds of sends per week, only big swings (offer, angle, CTA type) produce detectable differences; micro-tests (emoji in subject line) need volumes I may not have. Tell me plainly which of my planned tests are underpowered.
2. TEST DESIGN TEMPLATE — for each test: hypothesis in one falsifiable sentence, single variable isolated (never a new subject line AND a new CTA), primary metric (positive reply rate, not opens — open tracking is unreliable and hurts deliverability), minimum sample per arm before any peeking, and expected runtime at my volume.
3. AUTOMATION SPEC — how the split executes mechanically in my sequencer (built-in split versus duplicated campaigns with randomized list halves), how results flow to a scorecard automatically, and how the winner promotion works: manual sign-off, never auto-promotion on early numbers.
4. STOPPING RULES — the sample threshold before results count, what difference is meaningful versus noise at my volumes, and the discipline rule against ending tests early because one arm 'looks' ahead.
5. LEARNING LOG — the running document schema: test, hypothesis, dates, samples, result, decision — so wins compound instead of evaporating when someone leaves.

Rules: one variable per test, full stop. If my volume can't power a test in under six weeks, say 'don't run this test' and suggest the bigger swing that is detectable. No winners declared on opens.

Before designing, interview me. Ask ONE AT A TIME, waiting for my answer each time:
1. What is your weekly send volume, and what's your current positive reply rate?
2. What sequencer do you use, and does it support native split testing?
3. What do you most want to test first, and why that?
4. Who reviews results, and how often?
5. Have past tests changed what you send, honestly, or do results get admired and forgotten?

Then produce the design, including the verdict on whether my first planned test is even worth running.

How to use it

  1. 1

    Copy the prompt into Claude, ChatGPT, or any LLM.

  2. 2

    Give real volume and reply-rate numbers — statistical power is arithmetic, and flattering inputs produce fictional conclusions.

  3. 3

    Set up the split mechanics exactly as specced, including the scorecard automation, before launching the first test.

  4. 4

    Respect the stopping rules even when one arm looks like it's winning at day four; especially then.

Best practices

  • Test big swings at low volume: a different offer or angle can show through noise where a subject-line tweak never will.

  • Measure positive replies, not raw replies — a test that wins on volume of 'unsubscribe' responses is not a win.

  • Run one test at a time per audience; simultaneous tests on the same list contaminate both.

  • Reread the learning log before designing each new test — the second-cheapest test is the one someone already ran.

Example: what this looks like in practice

A founder sending 1,200 cold emails a week at a 1.8% positive reply rate wants to test emoji in subject lines. The testability check delivers the bad news: detecting a plausible lift from that change at her volume would take four months, so the design redirects her to an offer-level test — free audit versus paid pilot framing — detectable in three weeks at 600 sends per arm. The automation spec duplicates the Instantly campaign with a randomized list split via a Clay formula, pipes daily results to a scorecard sheet, and locks a 600-per-arm minimum before anyone peeks. The audit framing wins with a 2.9% versus 1.6% positive reply rate, the learning log records it, and the next test iterates on the winning offer instead of relitigating emojis.

Best fit

Roles
Founder / CEOSales LeaderMarketer
Company size
Startup (1–10)SMB (11–50)Mid-market (51–500)
Audience
B2B
Industries
Any industry
Works with
Any LLM
Difficulty
Intermediate

This prompt is one gear in a bigger machine. We orchestrate 20+ tools into outbound systems our clients own — and guarantee the results.

Apply for a Pilot Spot →
Prompt FAQ

Frequently asked questions

It depends on your baseline rate and the lift you're trying to detect — which is why most outbound 'tests' are underpowered theater. As rough guidance: detecting a large effect (reply rate doubling) needs several hundred sends per arm; detecting a subtle lift needs many thousands. If the runtime exceeds six weeks at your volume, test something bigger instead.

More automation & workflow design prompts