Sample floor
No page is judged before it reaches its minimum sample. An early lead on thin traffic is mostly noise.
FIELD REPORT 12 · CONVERSION RATE · MATCH GROUP
The landing page
tournament
About 100 landing pages, tested in rounds on paid traffic at Match Group. Each round cut the losers, and the winners' blocks were recombined into new challengers. The registration rate went from 12% to 24%.
2xregistration rate,
12% to 24%
12% to 24%
~100landing pages,
tested in rounds
tested in rounds
- ADVANCED
- NEW CHALLENGER
- CUT
- CHAMPION, A ROUND 3 CHALLENGER
01 THE BRIEF
88 of every 100 paid visitors left without registering.
At Match Group I was Senior Digital Marketing Manager, owning app and web acquisition: roughly $2,000,000 a month across Google, Meta and Apple Search Ads. Every paid web click landed on a landing page with one job, to get the visitor registered.
Those pages registered 12% of paid visitors. That rate sat under every web campaign. At the same cost per click, a page that registers twice as many visitors halves the cost of each registration.
Testing one page against another, one pair at a time, was too slow for the number of ideas we had. So I ran the landing pages as a tournament.
ROLESenior Digital Marketing Manager
SPEND~$2,000,000 a month, Google, Meta, Apple Search Ads
PAGESAbout 100, tested in rounds
WHENMatch Group (S&P 500), Nov 2019 to Mar 2024
02 THE MECHANISM
Losers came down. Winners were taken apart and recombined.
Run the rounds. Each read cuts the bottom three quarters of the live field, and every survivor gets a partner: a new challenger built from winning blocks. The round shapes are a sketch. The 12% start and the 24% finish are real.
● THE TOURNAMENTROUND 1 OF 5 · ILLUSTRATIVE ROUNDS
Round 1: 64 pages live, paid traffic split evenly. Run the round to read it.
ROUND 1 FIELD · 64 LIVE PAGESTRAFFIC SPLIT EVENLY
- KEPT FROM LAST ROUND
- NEW CHALLENGER
- ADVANCES
- CUT
PAGE TO BEATREGISTRATION RATE
- REAL: 12% BEFORE, 24% AFTER
- ILLUSTRATIVE STEPS
ROUND1 OF 5Paid traffic split evenly across the live field.
LIVE PAGES64The read cuts the bottom three quarters.
PAGES BUILT SO FAR64The sketch builds 94 by the final. The real program went through about 100.
PAGE TO BEAT12.0%Where it started. A real number.
Traffic was split evenly across the live pages in every round. A page came down only after it reached the sample floor and fell short of the confidence bar, so real rounds cut unevenly. The sketch cuts exactly three quarters to keep the arithmetic readable.
03 THE PAGE
Every page was five slots, filled differently.
One template, built for the phone first. A new page was a new pick in each slot, so a winner could be taken apart and its blocks reused. The options drawn here are illustrative.
- 01HEADLINEThe promise, a question, or a number.
- 02HERO IMAGEA couple, one person, or the app on a phone.
- 03PROOFStar ratings, press logos, or a member count.
- 04FORM POSITIONAt the top, under the image, or after the proof.
- 05CTA COPYAn action, a curiosity line, or a low commitment ask.
243
3 × 3 × 3 × 3 × 3, IN THIS SKETCH
possible pages from three options in five slots. Too many to test at once with a fair sample each, so the rounds searched them instead.
How a challenger was built
Two winners from the last read. Each slot on the new page came from one of them.
Blocks that kept winning spread through the field. Blocks that kept losing disappeared from it.
- FROM WINNER A
- FROM WINNER B
- NOT PASSED ON
04 THE RULES
No page came down on a hunch.
Three rules, written before round one and applied by the system the same way every round. The drawings are shapes, not data.
Confidence bar
A page comes down only when its whole range sits below the leader's. Overlap means the round keeps running.
Guardrail
Registrations had to turn into subscribers at a rate inside a band around the baseline from before the tournament. A page that won sign ups and lost subscribers was out.
05 WHAT I DID
I set the rules. The system ran the rounds.
PICKED ONE SCORE
Registrations per paid visitor, defined once and used in every round. Downstream subscription rate rode along as the guardrail, so a page could not win by pulling in sign ups who would never pay.
BUILT THE TEMPLATE
With design and development, I rebuilt the landing page as five swappable slots, designed for the phone first. A new page became a new combination of blocks we already had.
WROTE THE RULES FIRST
Sample floor, confidence bar and guardrail were written down before round one. The testing system applied them the same way every round, so a cut never depended on who was in the room.
RAN THE ROUNDS
Paid traffic split evenly across the live pages. At each read the losers came down, and the winners' blocks were recombined into the next round's challengers. About 100 pages went through the program.
Page speed was part of the program too: separately, I project managed an all time high in marketing dev work for SEO, including Core Web Vitals. I do not count that work in the 12% to 24%.
06 THE RESULT
Twice as many paid visitors registered.
The champion took all paid web traffic and registered 24% of visitors, up from 12% before the tournament. At the same cost per click, that halves the cost of each registration.
12%registration rate before the tournament
24%after, with the champion on all paid web traffic
~100landing pages tested in rounds
Same clicks, twice the registrations.
Each dot is one paid visitor. At a fixed cost per click, the cost of each registration falls by half.The 12%, the 24% and the page count are my own round numbers from reporting at the time. Round shapes, block options and the steps between 12% and 24% on this page are illustrative. Raw traffic and registration volumes stayed with Match Group.
07 WHAT CARRIES OVER
Any funnel with steady paid traffic can run a tournament.
01
Build pages from swappable blocks. A winner you can take apart shows which parts won, and those parts seed the next round.
02
Write the cut rules before the first visitor lands. A sample floor and a confidence bar settle the arguments about calling a test early.
03
Score one number per visitor and guard it with the next step in the funnel. Sign ups who never pay are a cost.
04
Keep the field full. Cutting alone runs out of pages. Recombining winners keeps producing better challengers.