Dynamic A/B Testing 

with AB/CD
Frontend Conference Zurich

September 2. 2016
Memi Beltrame
A/B Tests
Buy Now! Buy for $1’000
1. Divide users in 2 groups
A B
A/B Tests
Buy Now! Buy for $1’000
1. Divide users in 2 groups
2. Each group gets a button
A B
A/B Tests
Buy Now! Buy for $1’000
1. Divide users in 2 groups
2. Each group gets a button
3. See what works and what not
A B
Example with 1000 users
Buy Now! Buy for $1’000
500 500
Example with 1000 users
Buy Now! Buy for $1’000
3% Conversion 12% Conversion
500 500
Example with 1000 users
Buy Now! Buy for $1’000
3% Conversion 12% Conversion
15 Sales 60 Sales
500 500
Example with 1000 users
Buy Now! Buy for $1’000
3% Conversion 12% Conversion
15 Sales 60 Sales
$15’000 $60’000
500 500
Example with 1000 users
Buy Now! Buy for $1’000
3% Conversion 12% Conversion
15 Sales 60 Sales
$15’000 $60’000
500 500
A difference of $45’000
Finding out which
button works
just cost you $45’000
Is there a way to bring
down that number?
Dynamic A/B Tests
2 Rules
Buy Now! Buy for $1’000
2 Rules
Buy Now! Buy for $1’000
1. 10% get a button at random
2 Rules
Buy Now! Buy for $1’000
1. 10% get a button at random
2. 90% get the best button
Find the best button
By tracking 3 values for each one
Find the best button
1. Views
By tracking 3 values for each one
Find the best button
1. Views
2. Clicks
By tracking 3 values for each one
Find the best button
1. Views
2. Clicks
3. Ratio of Clicks / Views
By tracking 3 values for each one
How are Options Tracked?
Start of Test
How are Options Tracked?
After first user
How are Options Tracked?
After 1000 users
Resulting in $111’000 for the test
Finding out which
button works
just cost you $9’000
Finding out which
button works
just cost you $9’000
That’s $36’000 less.
AB/CD
A service to set up dynamic A/B
tests that handles the math behind it.
AB/CD
http://abcd.lessamess.ch
Setup
After Setup
1. Request an option via an API
After Setup
1. Request an option via an API
2. Store the option in a session
After Setup
1. Request an option via an API
2. Store the option in a session
3. Report back to AB/CD when
an option performs
API
What you get
Results in real time
Thanks!
Memi Beltrame
@bratwurstkomet

Dynamic A/B testing with AB/CD