She'll be sharing use cases, lessons learned and business impacts made using customer experience data and experimentation. She'll also share how combining quantitative data with unstructured quantitative data could provide powerful insights into user behavior that can be leveraged for improving business KPIs. Her team is using AI to support thousands of employees at PayPal and fueling innovation and data driven culture.
22. Password Reset: Users Timed Out If Not Completed Reset Within 5 minutes
22
You have 5 minutes to complete the entire Password Recovery Process
You must successfully set your password
before the 5 minutes ends, or you
“timeout”.
A customer may
experience one or
more “Security
Checks”
The timer starts once you
submit your email address.
Password
Reset
Complete
0:00
< 5:00
23. 1 in 13 Consumer Timed-out in PWR
81% Time-outs Between 5-10 mins
23
Data source: Tealeaf
81%
Extended the timer by a few minutes to reset password
25. Used Site Feedback To Identify Address Verification Service Issue
25
Feedback comment:
“I can't even make a account. I try and
sign and put my name and address and
it says check your information and try
again. I checked my information and
nothing's wrong.”
Leveraging NLP to identify the common customer issues
Address verification service rules
needed to improve to match with or
without space in address
“Winter crest Dr” vs
“Wintercrest Dr”
27. 1) KPI Considerations: Critical To Identify Business Metrics
Conversion Rates Heat Maps &
Session Replay
Voice of the Customer
Revenue
31%
5% 4%
1.2%
0.7%
1.5%
< 1% visitors clicks on
each individual link
< 1% visitors clicks on
each individual link
0.7%
< 1 % visitors clicks on
each individual link
0.1%
3%
0.3%
0.14% 0.08%
27
Customer Experience
Metrics
Business Metrics
Sign-ups Support Cost
28. 2) Apply Segmentation To Identify The Opportunities Within Segments
Identify which segments the test is performing well ( or not) compared to other,
segment the test by customer attributes such as:
• Device Types: Desktop or Mobile
• New or Returning
• Region
• Language
28
29. 3) Plan Measurement Methodology Upfront And Let The Data Reach
Significance To Ensure Quality of Analysis
Greater than
0.05:
Insufficient
evidence
0.01-0.05:
Some
evidence of
statistical
significance
Less than
0.01: Strong
evidence of
statistical
significance
29
P-Value
• Decide between Fixed-sample size vs
Sequential to optimize the time to complete
the test
• An experiment designed to detect a smaller
lift will require fewer observations using a
sequential test.
• An experiment designed to detect a larger lift
will on average require fewer observations
using a fixed-sample test.
30. 4) Monitor Platform Health: Verify Bias Free Random Sampling
Create A/A test
• A/A Test serves up identical experiences for all visitors on a page or experience
Orthogonality vs Interaction Effect Testing
• Regression Analysis for concurrent experiences
Monitor
• Split %: The accuracy of desired split %
• Conversion Rate: The validity of the randomizing algorithm
30
32. Leveraging NLP, Augmented Intelligence To Support Internal Customers
32
Stretching analytics teams’ bandwidth via AI chatbot to support internal users
34. 1. Set Goal: Measurable Business Impacts
34
• Set your goals upfront
• Communicate goals and what you would
deliver at the end of the project
• Goal examples
ü I want to analyze password recovery flow
and identify opportunities to reduce friction
and thus reducing customer contacts to the
call centers.
Expected business impact: xx$ Cost saving
I want to build a report to identify top
paths users take in password recovery flow.
35. 2. Start Small and Then Scale
35
• Plan for scalability but initially focus
on a couple of use cases
• Demonstrate solid methodology and
best practices that can be expanded
36. 3. Don’t Let Perfection Get On The Way Of Progress
36
• Utilize limited data-set that you
do have even if it may not be
complete
• Make the best of what you have
STRIVE FOR
PROGRESS
NOT
PERFECTION
37. 4. Utilize The Eco-System
37
Leverage other tools to
augment insights. Create
seamless integrations with:
• Site feedback
• Client web logs
• Server logs
• Existing Dashboards
38. 5. Empowerment: Self-Service And Training
38
Invest time in empowering cross-
functional teams:
• Democratize data
• Trainings/ FAQs
• Self-service reports
• Alerts
• Utilize Robotics & AI to power
self-service
40. 7. Track The Impact And Value Of Analysis
40
• Track value you have added via
analytics and insights
• Partner with business to
implement the actions you
recommend from your analyses