1 5 Kyle Rush Architecting Great Experiments MozCon 2014
-
Upload
ksampathme -
Category
Documents
-
view
218 -
download
0
Transcript of 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
1/61
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
2/61
MEANS
WRITE IT DOWNOR
TWEET IT
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
3/61
EXPERIMENT CONCEPTS
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
4/61
SIGNIFICANCEThe risk of encountering a false positive.
95% = out of 100 a/a tests, 5 will inaccurately report a difference.
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
5/61
POWERThe risk of encountering a false negative.
80% = out of 100 a/b tests, 20 winners will not be reported.
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
6/61
ONE-TAIL VS. TWO TAILOne tail = is the variation better?
Two tail = is the variation better OR worse?
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
7/61
MDEMinimum detectable effect
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
8/61
SAMPLE SIZEHow many subjects are in your experiment.
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
9/61
Always use a sample size calculator to calculate sample sizebefore starting an a/b test.
bit.ly/VUBti8
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
10/61
SAMPLE SIZE CALCULATOR
bit.ly/SWR3YC
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
11/61
EXAMPLEAbsolute lowest MDE
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
12/61
CONVERSION RATE: 4%
MDE: 1%
POWER: 80%SIGNIFICANCE: 5%
TAILS: ONE
2,972,435Visitors per branch
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
13/61
EXAMPLEFocus on time
1 month = 170,000 unique visitors
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
14/61
CONVERSION RATE: 4%
MDE: 6%
POWER: 80%SIGNIFICANCE: 5%
TAILS: ONE
83,230Visitors per branch
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
15/61
EXAMPLESmall startup
1 month = 3,000 unique visitors
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
16/61
CONVERSION RATE: 4%
MDE: 45%
POWER: 80%SIGNIFICANCE: 5%
TAILS: ONE
1,567Visitors per branch
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
17/61
Sample size calculators tell you how many subjects,but not which subjects should be in your experiment.
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
18/61
SAMPLINGcan be really hard
week day vs. weekend traffic
campaign vs. organic traffic
returning vs. new visitors
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
19/61
WHERE SHOULD I TEST?
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
20/61
EXAMPLEE-commerce website
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
21/61
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
22/61
RUN EXPERIMENT FOR 1 MONTH (100K VISITORS)
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
23/61
RUN EXPERIMENT FOR 1 MONTH (100K VISITORS)
CONVERSION RATE: 2.5%
POWER: 80%SIGNIFICANCE: 5%
TAILS: ONE
MDE: 10%
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
24/61
CHECKOUT PAGE
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
25/61
CHECKOUT PAGE
ASSUMPTIONS1. Lower traffic2. Relatively high conversion rate
Let's assume:
50% conversion rate
10,000 monthly unique visitors
RUN EXPERIMENT FOR 1 MONTH (10K VISITORS)
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
26/61
RUN EXPERIMENT FOR 1 MONTH (10K VISITORS)
CONVERSION RATE: 50%
POWER: 80%SIGNIFICANCE: 5%
TAILS: ONE
MDE: 5%
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
27/61
Start by focusing a/b tests on thelast step in your conversion funnel.
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
28/61
WHAT SHOULD I TEST?Depends on MDE and time.
LANDING PAGE
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
29/61
LANDING PAGE
OPTIMIZATION(MOZCON 2013)
bit.ly/1wkpgye
GOING BEYOND THE LOW
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
30/61
GOING BEYOND THE LOW
HANGING FRUIT(CONVERSION CONFERENCE 2014)
bit.ly/1kU4sZ0
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
31/61
BE FEARLESS
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
32/61
EXPERIMENTTerse vs verbose
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
33/61
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
34/61
RESULT+31% LEADS
99.9% CONFIDENCE
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
35/61
BEFORE YOU TEST
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
36/61
GOALSMeasure as many goals as possiblemicro: form field errors, time on pagemacro: purchase, revenue
Choose a primary goalDon't forget about down the funnel goals
Repeat purchase
Save payment information
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
37/61
EXAMPLESuccess vs. submit
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
38/61
13.72% CONVERSION RATE
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
39/61
17.15% CONVERSION RATE
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
40/61
25% DIFFERENCE
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
41/61
Measure as many goals as possible for every experiment.
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
42/61
QUALITY ASSURANCENo bugs in the variationNo bugs in the control
Tracking works correctly
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
43/61
ELIMINATING BIASDouble blind experiments
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
44/61
EXPERIMENT BRIEFHypothesisAudience descriptionGoals tracked
Stopping conditionsScreenshots
QA summary
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
45/61
EXPERIMENT BRIEF
TEMPLATEbit.ly/1nvNJjx
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
46/61
AFTER YOU TEST
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
47/61
STATISTICAL TIENot enough data to conclude that there is a difference.
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
48/61
The overwhelming majority of
experiments results are a statistical tie.
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
49/61
EXAMPLEStatistical tie
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
50/61
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
51/61
RESULT
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
52/61
RETESTING
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
53/61
FIRST TEST
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
54/61
+36% revenue
SECOND TEST
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
55/61
statistical tie
THIRD TEST
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
56/61
Statistical tie
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
57/61
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
58/61
Always, always record detailed
experiment results in an archive.
EXPERIMENT ARCHIVE
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
59/61
EXPERIMENT ARCHIVEexperiment date
audience/url
screenshots
hypothesisresults
link to experiment
link to result csv
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
60/61
EXPERIMENT ARCHIVE
TEMPLATEbit.ly/1q9tRWI
-
8/12/2019 1 5 Kyle Rush Architecting Great Experiments MozCon 2014
61/61