Offer testing Evidence

Two variants, one winner, called on deposits

Service 02 / 04

Almost every SMS test we inherit is measured on clicks, which is why almost every one of them is wrong. A bonus that pulls a big click rate and a small deposit rate is a worse offer that looks like a better one. We split on the segment, hold everything else still, and let the deposit column decide.

1 in 3

Tests that overturn the favourite

6 hrs

Median time to a call

9,000

Minimum arm size we will call on

How a job actually runs

01

Frame

We name the one thing being tested and what result would change your mind.

02

Split

The segment is halved on a hash of the player id, so the halves are comparable.

03

Watch

Deposits are counted live against each arm as the waves go out.

04

Roll

The winner takes the remainder of the list, usually the same evening.

What is in it

Everything below is included at the standard engagement. Nothing here is an upsell later on.

  • Matched halves drawn from one segment, not two similar ones
  • One variable changed at a time, so the result means something
  • The deciding metric agreed in writing before the send
  • Minimum sample enforced, so a lucky evening cannot call a test
  • Losing variants stopped mid-flight rather than left to finish
  • A running library of what has already been proven on your list

Get offer testing priced

Tell us the market, the segment and the offer. You get the plan and the number back, not a calendar invite.

Messages sent today

407,891