====== Why refund windows break test conclusions ======

===== The premature winner problem =====

Most split tests conclude before the refund window closes.

Variant A converts better.
Variant B underperforms.
A winner is declared.

But refund windows exist for a reason.

Revenue is not final at checkout.

CBSplit was built to evaluate tests after financial resolution, not before it.

===== Refund windows delay revenue truth =====

Many offers include:

  
  * 30-day refund periods
  * 60-day guarantees
  * Conditional return policies
  * Satisfaction guarantees

During this period, revenue is provisional.

Tests that end before refund windows close measure incomplete outcomes.

===== Impulse-driven variants often win early =====

Variants that use:

  
  * Urgency-based messaging
  * Scarcity pressure
  * Emotional framing
  * Simplified billing language

Frequently increase initial conversions.

They may also increase:

  
  * Buyer remorse
  * Post-purchase dissatisfaction
  * Refund submission rates

Refund windows expose this misalignment.

===== Refund timing reshapes net revenue =====

Refunds typically appear:

  
  * Days after purchase
  * After product use
  * After billing clarity
  * After upsell review

A front-end winner may look profitable before refunds are processed.

After refund reconciliation, net revenue can reverse.

CBSplit recalculates performance using refund-adjusted results.

===== Cohort-level refund behavior matters =====

Refund impact is not uniform.

Different cohorts may show:

  
  * Higher refund rates in specific geographies
  * Higher dissatisfaction from certain traffic sources
  * Refund clustering from aggressive angles

Short tests average this behavior out.

Refund windows reveal structural weaknesses.

===== Scaling before refund closure compounds risk =====

When a test winner is scaled prematurely:

  
  * Traffic volume increases
  * Refund exposure multiplies
  * Processor scrutiny rises
  * Net profitability shrinks

Refund windows often expose fragility after scaling decisions are made.

CBSplit aligns scaling decisions with refund maturity.

===== Subscription funnels amplify refund distortion =====

Subscription offers often combine:

  
  * Initial purchase
  * Trial periods
  * First rebill cycles
  * Refund eligibility windows

If tests conclude before these cycles complete, evaluation is incomplete.

CBSplit extends testing horizons beyond initial billing events.

===== Refund-adjusted metrics redefine winners =====

True evaluation requires:

  
  * Net revenue after refunds
  * Cohort-based refund ratios
  * Rebill-adjusted LTV
  * Processor-safe performance

A variant that wins before refund closure may lose afterward.

CBSplit prevents premature conclusions.