+ All Categories
Home > Marketing > Is your A/B test result significant?

Is your A/B test result significant?

Date post: 07-Dec-2014
Category:
Upload: euronet-srl
View: 215 times
Download: 0 times
Share this document with a friend
Description:
A/B-testing analytics errors by Online Dialogue
52
< Have more fun A/B-testing > Ton Wesseling – CEO Testing.Agency #DDTT Amsterdam 2014/06/24 A/B-testing Errors
Transcript
Page 1: Is your A/B test result significant?

< H a v e m o r e f u n A / B - t e s t i n g >

To n W e s s e l i n g – C E O Te s t i n g . A g e n c y # D D T T A m s t e r d a m 2 0 1 4 / 0 6 / 2 4

A / B - t e s t i n g E r r o r s

Page 2: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Why me?

ü  10 years+ A/B-test experience

ü  I always loved the numbers side

ü  I’ve made all these mistakes

Page 3: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Page 4: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

We opt imize

Page 5: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Nature

Page 6: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

A better example

Page 7: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Groundhog Day – The A/B-test movie

Page 8: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Trying to get the gir l

Page 9: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Constant test ing environment

A / B Testing

Page 10: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Better Dialogue Optimizat ion

Winner!"

Variation B

Variation A

Default

Page 11: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

This is why you’re A/B-test ing

Small steps to learn how

to make the next big step

Page 12: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Test ing is not free

Use it to LEARN

Page 13: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

50% of the A/B-tests I get to see is wrong

Page 14: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Look at your A/B-test ing results!

+24% +52%

+83% +12%

+126% +16%

Page 15: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Did you push the winners l ive to your si te?

You must be rich now!

Page 16: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Or is this real i ty?

Page 17: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Or even worse:

A/B-test winners are letting you down!

Page 18: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Are these real winners?

+24% +52%

+83% +12%

+126% +16%

Page 19: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

20 reasons why

Your A/B-test analyzing sucks

Page 20: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Get your stat ist ics r ight!

Page 21: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

1: low signif icance levels

ü  Big chance the winner is not a real winner

Page 22: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

2: low power levels

ü  Big chance a real winner is not recognized

Page 23: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Example test set-up

ü  90% significance level

ü  100 conversions per variation

ü  Average conversion rate: 2%

ü  20 out of 100 ideas is a winner

ü  With an average uplift of 10%

Page 24: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Average: 75% signif icance & 40% power

Page 25: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

So:

ü  100 tests will give 10 false positives (90% significance)

ü  40% of the 20 winners are recognized (40% power)

ü  8 real winners & 10 false positives

ü  NOT 18 improvements of 15%+

Page 26: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

3: Not calculat ing traff ic up front

ü  You thought the page had thousands of

visitors

Page 27: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

4: Not calculat ing what you can improve

ü  You thought the page had lots and lots

conversions

Page 28: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

ABTestGuide.com/calc

Page 29: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

5: Not choosing your segments up front

ü  You just go and dig unitil you find something

(you will always find something that seems true BUT is not!)

Page 30: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

6: You stop the test once i t is signif icant

Page 31: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

6: You stop the test once i t is signif icant

Page 32: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

7: You just let i t run unt i l i t ’s signif icant

Page 33: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

7: You just let i t run unt i l i t ’s signif icant

Your test should have a fixed lenght!

Page 34: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

8: Your test takes too long

ü  Cookie deletion (you will loose 10% in two weeks)

They will re-enter the the and pollute your samples!

Page 35: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

9: You just choose a test length

ü  At least 1 full purchase cycle!

Page 36: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

10: You don’t test ful l weeks

ü  At least 1 full week (if your not testing for time of day / week effects)

Page 37: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

11: You think they always use one decive

ü Only test 1 device at a time!

Page 38: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

12: You’re test ing broken stuff

ü  Always check browser compatibility!

ü Dont’ break dynamic stuff!

Page 39: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

13: You have mult iple tests that overlap

ü  Your variation did not won (it was a combination of variations)

Page 40: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

14: You thought this page matters

ü  But it did not (so you decide to use some other metric to declare the winner)

Page 41: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

15: You forgot to leave out some visi tors

ü  Only test on those who can be changed (if you test a sales promotion – leave out your current clients who just logon)

Page 42: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

16: You’re not slowing the control down

ü  Add the same code to the control

Page 43: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

17: You’re not test ing on fresh people

ü  Start with control for a full purchase cycle (and then start sending traffic to your variations – setting the control to 0%)

Page 44: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

18: You forgot to give them t ime to buy

ü  Stop getting fresh visitors after your test time (send new traffic to the control, but give you tested people time to finish converting)

Page 45: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

19: You’re not diving into your analyt ics

ü  Always measure with analytics software (You will find conversions that are not supposed to be there!)

Page 46: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

20: You’re not measuring the r ight stuff

ü  Use unique visitors

ü  Against lifetime value predictors!

Page 47: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

I f you keep on making mistakes:

Stuff like this happens

Page 48: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Test ing is not free

Use it to LEARN

Page 49: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

Bonus: You did not re-test your winners

ü  Always re-test to be sure it was not a glitch

Page 50: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

So when you implement winners

This will happen!

Page 51: Is your A/B test result significant?

< H a v e m o r e f u n A / B - t e s t i n g >

To n W e s s e l i n g – Te s t W i n n e r # D D T T A m s t e r d a m 2 0 1 4 / 0 6 / 2 4

A / B - t e s t i n g E r r o r s

Page 52: Is your A/B test result significant?

Email: [email protected] Twitter: @TonW #DDTT

ABTestGuide.com/calc


Recommended