A test structure that produces learning rather than a winner you cannot explain.
Facebook Ads
You are designing a creative testing plan for .
## Context
- Budget: £6,000/month
- Target CPA: £120 — payback in 4 months at £480 LTV
- Audience: Heads of marketing at 20–200 person B2B SaaS companies
- Product: A project management tool for creative agencies
## Step 1 — Check the arithmetic first
Before designing anything, work out whether meaningful testing is possible at £6,000/month.
Calculate: at £120 — payback in 4 months at £480 LTV, how many conversions does £6,000/month. produce per month? Split across how many variants? A test needs a reasonable number of conversions per variant to distinguish real differences from noise.
If £6,000/month cannot support the number of variants proposed, say so directly and reduce the scope. Running underpowered tests and acting on the results is worse than not testing — it produces confident wrong conclusions.
## Step 2 — Test concepts before variations
The correct sequence:
1. **Concept tests** — genuinely different creative angles. Largest effect sizes, so cheapest to detect.
2. **Variation tests** — within the winning concept, test hooks and formats.
3. **Refinement** — only once a concept is proven.
Most accounts invert this and spend budget optimising a concept that was never the best available.
## Step 3 — Structure it
- What runs simultaneously, and what would contaminate what
- Budget per variant and duration
- The decision rule set **in advance**: what result kills a variant, what promotes it
- The minimum runtime before judging — early results are dominated by the algorithm's learning phase and are misleading
## Step 4 — Define learning, not just winning
For each test, state the question it answers. "Which ad performed better" is not learning. "Does this audience respond to proof or to problem framing" transfers to everything you make next.
## Deliver
The arithmetic and whether the plan is viable, the test sequence, the structure with decision rules, and the learning questions.
## Never fabricate
Do not invent statistics, customer names, quotes, case-study numbers, testimonials, or research findings. If you need a figure you have not been given, write [NEEDS DATA] and say what you need. Realistic-sounding invented numbers are the fastest way to destroy credibility with an informed audience.