1A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
A/B Testing Ultimate Guideline
How to think, design, release and analyze A/B testing
and learn about them
30/05/2018
GPeC Summit - Bucharest
Ricardo Tayar - @rtayar
2A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Hello Bucharest!
4A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
¿What is A/B testing?
5A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
In web analytics, A/B testing (bucket tests or split-run testing) is a controlled
experiment with two variants, A and B.[1] [2] It is a form of statistical hypothesis
testing or "two-sample hypothesis testing" as used in the field of statistics. A/B testing
is a way to compare two versions of a single variable typically by testing a subject's
response to variable A against variable B, and determining which of the two
variables is more effective.
Wikipedia
6A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
7A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
8A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
9A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
10A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
11A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
12A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
13A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
¿Why A/B testing
(and personalization)
is so needful in a
digital business?
14A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
A/B testing is the main tool we have to prove that something we have built a
hypothesis around (action, strategy, change) will improve our results in our digital
business, starting from data.
A/B testing allow us to validate with quantitative data how useful and needful a
changue could be in our context in a specific moment.
It helps us to avoid guessing or self-reference design. Both things are great problems
in digital design processes today.
It helps us in our constant learning about how can we built a stronger and more
efficient digital product.
15A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
16A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Data put design in
business context
17A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Goals
18A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Goals
Revenue
Leads
Increasing %
Engagement
19A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Data
20A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Goals Data
Facturación
GA / Digital
Analytics
Leads Qualitative data
Incrementos % Logs
Engagement CRM / LTV
User Research
21A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
22A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
23A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
24A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
25A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
26A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
27A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Hypothesis
28A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Goals Data Hypothesis
Facturación
GA / Digital
Analytics
Brainstorming
Leads
Datos
cualitativos
5 why rule
Incrementos % Logs
Covariation
hypothesis
Engagement CRM / LTV
Correlation /
Cause-effect
User Research FUDs
29A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
A hypothesis should be a statement, a proposition that you make to describe what
will happen in a system given specific circumstances when we modify some
variables. It often follow this form: “If we do X, user will do Y with will impact metric
A”.
Hypotheses should be descriptive but short and they have to generate a
consequence in the system.
Example: “If we improve our information architecture user will find easyly our
products, so our bounce rat will decrease and tire will be a bigger amount of users
starting checkout process”
30A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Correlations:
31A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
FUDs:
Fears
Uncertainties
Doubts
UX research, interviews, online surveys… will give you real feedback about the fears,
uncertainties and doubts.
32A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Designing solutions
33A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Goals Data Hypothesis Solutions
Facturación
GA / Digital
Analytics
Brainstorming UX
Leads
Datos
cualitativos
5 why rule Traffic
Incrementos % Logs
Covariation
hypothesis
Technology
Engagement CRM / LTV
Correlation /
Cause-effect
Business model
User Research FUDs Triggers
34A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Test briefing:
35A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Testing
36A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Goals Data Hypothesis Solutions Test
Facturación
GA / Digital
Analytics
Brainstorming UX A/B test
Leads
Datos
cualitativos
5 why rule Traffic MVT
Incrementos % Logs
Covariation
hypothesis
Technology Split
Engagement CRM / LTV
Correlation /
Cause-effect
Business model JS Custom
User Research FUDs Triggers Personalization
37A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Testing - scientist method:
Independent variable: The variable you´re going to study and modify, and the one
that affect results in the test. You should control that variable. Classic example:
nutrition. Digital Example: information architecture.
Dependent variable/s: DIt depends on the independent variable and its value is
affected by it. They respond to the change made to the independent variable. Classic
example: nutrition. Digital example: bounce rate.
Controlled variables: Controlled variables are quantities that should remain
constant to make the test reliable, and they has to be observed them carefully
because any change in these can modify test results. Classic example: training.
Digital example: Release of new Adwords campaing.
38A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Test brochure:
39A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
How many users do I need for a solid testing? - https://www.optimizely.com/sample-
size-calculator/:
40A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
How many users do I need for a solid testing? - https://www.optimizely.com/sample-
size-calculator/:
41A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
How many users do I need for a solid testing? - https://www.optimizely.com/sample-
size-calculator/:
42A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Conversion rate metric we´re measuring:
conversion rate, download, bounce rate, etc
Miminum relative change we want to detect
in that metric. In this case we want to detect
a range between 0,75%-2,25%
How reliable your test is. Tthe probability
that the results don´t come from chance /
luck (p-value)
How many users do I need for a solid testing? - https://www.optimizely.com/sample-
size-calculator/:
43A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Traffic volume - statistical significance:
Can I do A/B testing if I don´t have enough amount of traffic to gain statistical
significance?, Should I do it?, is A/B testing only for big projects?
44A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Testing type:
A/B Test - Split testing: The metrics tested of two versions of a page — version A
and version B — are compared to one another. Site visitors are bucketed into one
version or the other. There´s only one variable changing at the same time.
MVT Test: Compares a higher number of variables at the time times. Several items
(variables) are tested in the same main layout.
Personalization: Based on cookies and navigation paths, we offer a customized
digital experience to the users that fix into a set of rules previously defined.
45A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
A/B Test - Split testing:
A/B Test - Split testing:
46A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
A: Original Image
B: Alternative Image
C: Alternative Image
47A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
MVT Test:
MVT Test:
48A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
A: Original Image
B: Alternative Image
C: Alternative Image
A: Original CTA
B: Alternative CTA
C: Alternative CTA
49A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Split Test:
50A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Segments and distribution:
Before launching any kind of A/B testing we have to set the amount of traffic we
want to send to the test (statistical significance - reliable) and we have to define the
right segments tu run the test.
51A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Segments and distribution:
Main segments we have to consider:
Desktop / Smartphone: Una Same solution can be solved so differently depending
on the interface type, so we can have very different results from the same test.
Countries: Same experiment can give us different data depending on the country we
´re runing it.
Business units: Although we´re working in the same e-commerce, each one has
several business units or activities. Not all actions will give us same feedback
depending on the business unit.
Traffic channels: Different results depending n the origin of the traffic.
52A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Segments and distribution:
Main segments we have to consider:
Date: Different results depending on running the test in Christmas, summer, etc.
Political / Society variables
Login or not
Weather
53A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Tools to run an A/B testing
• Optimize (Google)
• Optimizely
• Visual Website Optimizer
• A/B Tasty
• Qubit
• Target
• Wise Conversion
• Maximizer
• Convert
54A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Analyze
55A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Goals Data Hypothesis Solutions Test Analyze
Facturación
GA / Digital
Analytics
Brainstorming UX A/B test Context
Leads
Datos
cualitativos
5 why rule Traffic MVT
Statistical
significance
Incrementos % Logs
Covariation
hypothesis
Technology Split Cognitive bias
Engagement CRM / LTV
Correlation /
Cause-effect
Business model JS Custom
User Research Triggers Personalization
56A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Lead generation performance - Insurance:
57A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
E-commerce add to basket:
58A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
E-commerce related products:
59A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
E-commerce price management:
60A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Insurance process:
61A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Improvement
62A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Goals Data Hypothesis Solutions Test Analyze Improvement
Facturación
GA / Digital
Analytics
Brainstorming UX A/B test Context Learning
Leads
Datos
cualitativos
5 why rule Traffic MVT
Statistical
significance
Change
Incrementos % Logs
Covariation
hypothesis
Technology Split Cognitive bias Fail
Engagement CRM / LTV
Correlation /
Cause-effect
Business model JS Custom
User Research Triggers Personalization
63A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
Stop guessing. Start
thinking. Put design
in business context.
Go testing.
Thank you Bucharest!
Ricardo Tayar
ricardo@flat101.com

A/B Testing Ultimate Guideline. How to design and analyze digital testing.

  • 1.
    1A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com A/B Testing Ultimate Guideline How to think, design, release and analyze A/B testing and learn about them 30/05/2018 GPeC Summit - Bucharest Ricardo Tayar - @rtayar
  • 2.
    2A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Hello Bucharest!
  • 4.
    4A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com ¿What is A/B testing?
  • 5.
    5A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com In web analytics, A/B testing (bucket tests or split-run testing) is a controlled experiment with two variants, A and B.[1] [2] It is a form of statistical hypothesis testing or "two-sample hypothesis testing" as used in the field of statistics. A/B testing is a way to compare two versions of a single variable typically by testing a subject's response to variable A against variable B, and determining which of the two variables is more effective. Wikipedia
  • 6.
    6A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 7.
    7A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 8.
    8A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 9.
    9A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 10.
    10A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 11.
    11A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 12.
    12A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 13.
    13A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com ¿Why A/B testing (and personalization) is so needful in a digital business?
  • 14.
    14A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com A/B testing is the main tool we have to prove that something we have built a hypothesis around (action, strategy, change) will improve our results in our digital business, starting from data. A/B testing allow us to validate with quantitative data how useful and needful a changue could be in our context in a specific moment. It helps us to avoid guessing or self-reference design. Both things are great problems in digital design processes today. It helps us in our constant learning about how can we built a stronger and more efficient digital product.
  • 15.
    15A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 16.
    16A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Data put design in business context
  • 17.
    17A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Goals
  • 18.
    18A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Goals Revenue Leads Increasing % Engagement
  • 19.
    19A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Data
  • 20.
    20A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Goals Data Facturación GA / Digital Analytics Leads Qualitative data Incrementos % Logs Engagement CRM / LTV User Research
  • 21.
    21A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 22.
    22A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 23.
    23A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 24.
    24A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 25.
    25A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 26.
    26A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com
  • 27.
    27A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Hypothesis
  • 28.
    28A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Goals Data Hypothesis Facturación GA / Digital Analytics Brainstorming Leads Datos cualitativos 5 why rule Incrementos % Logs Covariation hypothesis Engagement CRM / LTV Correlation / Cause-effect User Research FUDs
  • 29.
    29A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com A hypothesis should be a statement, a proposition that you make to describe what will happen in a system given specific circumstances when we modify some variables. It often follow this form: “If we do X, user will do Y with will impact metric A”. Hypotheses should be descriptive but short and they have to generate a consequence in the system. Example: “If we improve our information architecture user will find easyly our products, so our bounce rat will decrease and tire will be a bigger amount of users starting checkout process”
  • 30.
    30A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Correlations:
  • 31.
    31A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com FUDs: Fears Uncertainties Doubts UX research, interviews, online surveys… will give you real feedback about the fears, uncertainties and doubts.
  • 32.
    32A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Designing solutions
  • 33.
    33A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Goals Data Hypothesis Solutions Facturación GA / Digital Analytics Brainstorming UX Leads Datos cualitativos 5 why rule Traffic Incrementos % Logs Covariation hypothesis Technology Engagement CRM / LTV Correlation / Cause-effect Business model User Research FUDs Triggers
  • 34.
    34A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Test briefing:
  • 35.
    35A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Testing
  • 36.
    36A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Goals Data Hypothesis Solutions Test Facturación GA / Digital Analytics Brainstorming UX A/B test Leads Datos cualitativos 5 why rule Traffic MVT Incrementos % Logs Covariation hypothesis Technology Split Engagement CRM / LTV Correlation / Cause-effect Business model JS Custom User Research FUDs Triggers Personalization
  • 37.
    37A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Testing - scientist method: Independent variable: The variable you´re going to study and modify, and the one that affect results in the test. You should control that variable. Classic example: nutrition. Digital Example: information architecture. Dependent variable/s: DIt depends on the independent variable and its value is affected by it. They respond to the change made to the independent variable. Classic example: nutrition. Digital example: bounce rate. Controlled variables: Controlled variables are quantities that should remain constant to make the test reliable, and they has to be observed them carefully because any change in these can modify test results. Classic example: training. Digital example: Release of new Adwords campaing.
  • 38.
    38A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Test brochure:
  • 39.
    39A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com How many users do I need for a solid testing? - https://www.optimizely.com/sample- size-calculator/:
  • 40.
    40A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com How many users do I need for a solid testing? - https://www.optimizely.com/sample- size-calculator/:
  • 41.
    41A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com How many users do I need for a solid testing? - https://www.optimizely.com/sample- size-calculator/:
  • 42.
    42A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Conversion rate metric we´re measuring: conversion rate, download, bounce rate, etc Miminum relative change we want to detect in that metric. In this case we want to detect a range between 0,75%-2,25% How reliable your test is. Tthe probability that the results don´t come from chance / luck (p-value) How many users do I need for a solid testing? - https://www.optimizely.com/sample- size-calculator/:
  • 43.
    43A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Traffic volume - statistical significance: Can I do A/B testing if I don´t have enough amount of traffic to gain statistical significance?, Should I do it?, is A/B testing only for big projects?
  • 44.
    44A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Testing type: A/B Test - Split testing: The metrics tested of two versions of a page — version A and version B — are compared to one another. Site visitors are bucketed into one version or the other. There´s only one variable changing at the same time. MVT Test: Compares a higher number of variables at the time times. Several items (variables) are tested in the same main layout. Personalization: Based on cookies and navigation paths, we offer a customized digital experience to the users that fix into a set of rules previously defined.
  • 45.
    45A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com A/B Test - Split testing:
  • 46.
    A/B Test -Split testing: 46A/B Testing Ultimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com A: Original Image B: Alternative Image C: Alternative Image
  • 47.
    47A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com MVT Test:
  • 48.
    MVT Test: 48A/B TestingUltimate Guide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com A: Original Image B: Alternative Image C: Alternative Image A: Original CTA B: Alternative CTA C: Alternative CTA
  • 49.
    49A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Split Test:
  • 50.
    50A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Segments and distribution: Before launching any kind of A/B testing we have to set the amount of traffic we want to send to the test (statistical significance - reliable) and we have to define the right segments tu run the test.
  • 51.
    51A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Segments and distribution: Main segments we have to consider: Desktop / Smartphone: Una Same solution can be solved so differently depending on the interface type, so we can have very different results from the same test. Countries: Same experiment can give us different data depending on the country we ´re runing it. Business units: Although we´re working in the same e-commerce, each one has several business units or activities. Not all actions will give us same feedback depending on the business unit. Traffic channels: Different results depending n the origin of the traffic.
  • 52.
    52A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Segments and distribution: Main segments we have to consider: Date: Different results depending on running the test in Christmas, summer, etc. Political / Society variables Login or not Weather
  • 53.
    53A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Tools to run an A/B testing • Optimize (Google) • Optimizely • Visual Website Optimizer • A/B Tasty • Qubit • Target • Wise Conversion • Maximizer • Convert
  • 54.
    54A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Analyze
  • 55.
    55A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Goals Data Hypothesis Solutions Test Analyze Facturación GA / Digital Analytics Brainstorming UX A/B test Context Leads Datos cualitativos 5 why rule Traffic MVT Statistical significance Incrementos % Logs Covariation hypothesis Technology Split Cognitive bias Engagement CRM / LTV Correlation / Cause-effect Business model JS Custom User Research Triggers Personalization
  • 56.
    56A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Lead generation performance - Insurance:
  • 57.
    57A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com E-commerce add to basket:
  • 58.
    58A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com E-commerce related products:
  • 59.
    59A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com E-commerce price management:
  • 60.
    60A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Insurance process:
  • 61.
    61A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Improvement
  • 62.
    62A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Goals Data Hypothesis Solutions Test Analyze Improvement Facturación GA / Digital Analytics Brainstorming UX A/B test Context Learning Leads Datos cualitativos 5 why rule Traffic MVT Statistical significance Change Incrementos % Logs Covariation hypothesis Technology Split Cognitive bias Fail Engagement CRM / LTV Correlation / Cause-effect Business model JS Custom User Research Triggers Personalization
  • 63.
    63A/B Testing UltimateGuide / GPeC Summit Bucharest / Ricardo Tayar - @rtayar / ricardo@flat101.com Stop guessing. Start thinking. Put design in business context. Go testing.
  • 64.
    Thank you Bucharest! RicardoTayar ricardo@flat101.com