Nike By You Member Data Analytics Project
Yue Zheng
05/12/2022
Introduction
Nike By You allows consumers to customize Nike shoes. The purpose of this study is to help teams explore the characteristics of the Nike By You buyers, and analyze how Nike By You influence the customer behaviors.
This study focuses on five main aspects:
- Exploratory data analysis
- New NBY abtesting (before and after 12-10)
- Modeling
Data
1. Inline Transaction Data 1
ALL transactions occurred in China. Time period is from 01/01/2021 to 12/30/2021.
2. Member Data 2
Member data includes information about the members who made inline purchases in stores, including register_date, UPMID, and so on.
3. Member Segment 3
Membership segments were created by the member science team to better understand our existing members and to drive direct engagement of members through targeting.
4. Scalper Data 4
Scalper data was based on Yunhai’s Scalper Detection Machine, including scalpers’ UPMID, and so on.
Exploratory data analysis
1. The general analysis
NBY order percentage
| Type | NBY_order_quantity | Total_order_quantity | NBY_order_. |
|---|---|---|---|
| HOI | 2887 | 210738 | 1.37% |
| Online | 265962 | 5611055 | 4.74% |
The online NBY order rate is 4.74%, which is 346% higher than the HBI
NBY order rate, 1.37%. We can tell that consumers are more likely to buy
NBY products online.
NBY member and non-NBY member quantity comparison
| Type | NBY_member | non_NBY_member | NBY_member_percentage |
|---|---|---|---|
| HOI | 2637 | 147568 | 2% |
| Online | 237185 | 1998271 | 12% |
Likewise, the online platform shows a much higher NBY member
percentage than that of HOI, which also proves that online consumers
have more interests in NBY products.
2. HOI NBY exploratory analysis
The HOI NBY category distribution
As for NBY member segments, there’re more than 50% of the NBY members
are classified as Lifestyle. So it’s speculated that lifestyle products
are most popular among NBY products.
The new member percentage
The new registered member rate for NBY member is higher than the non
NBY member, which means NBY could helpt to acquire new customers.
HOI NBY member and non-NBY member gender distribution comparison
From the bar chart above, there’s no significant discrepancy on
gender distribution between NBY members and non-NBY members for
HOI.
Inline NBY member and non-NBY member age distribution comparison
From the HOI age distribution, NBY members show the younger age trend
comparing with non-NBY members.
Inline NBY member and non-NBY member segment distribution comparison
The NBY members show the much stronger interests in Lifestyle than
the non-NBY members, which is aligned with the conclusion we got from
member segments distribution that lifestyle products are most popular to
NBY members. Besides, NBY members are less likely to be classified as
Sneaker head.
3. Online NBY exploratory analysis
The online NBY category distribution
The majority of the products consumed by online NBY members are lifestyle products.
The new registered member rate for NBY member is higher than the non
NBY member.
The online NBY member platform
All the online NBY members are from NIKE_APP and DCOM_WEB. The NBY
member rate on DCOM_WEB is greatly higher than that of NIKE_APP.
The online NBY member shipping city
Online NBY member and non-NBY member gender distribution comparison
As for the online gender distribution comparison, the percentage of
female NBY members is 62% larger than that of the non-NBY members, which
is different from the HOI gender distribution.
Online NBY member and non-NBY member age distribution comparison
Similarly, from the online age distribution, NBY members also show
the younger age trend comparing with non-NBY members.
Online NBY member and non-NBY member segment distribution comparison
For the online NBY member segments, “Family” takes the largest
proportion and is much higher than that of the non-NBY members. It is
speculated that it’s because the family buyers are more likely to shop
online and buy Nike products for their family and friends, and the
characteristic of customization of NBY is more likely to match their
interests. Besides, The proportions of Performance and Sneaker head of
NBY members are much lower than that of non-NBY members.
New NBY abtesting (before and after 12-10)
Post NBY Purchase AUR = Average Unit Revenue AFTER Their NBY
Experience
Post NBY Purchase UPT = Average Unit Per Transaction AFTER Their NBY
Experience.
Post NBY Purchase ADPT = Average Demand Per Transaction AFTER Their NBY
Experience.Percent of transactions containing a mixture of FTW/APP/EQ
AFTER NBY experiences. This helps us to see if NBY member purchase more
head-to-toe combination than regular consumer.
Post NBY Purchase Product Design – Lifestyle = Percent of Lifestyle
items purchased AFTER NBY experience
Repurchase within 30 days = Percent of NBY member that came back to HOI
and made a purchase within 30 days AFTER the NBY experience.
From the Abtesting data, we can see that NBY Member’s tendency to
purchase Lifestyle product is much higher than control group. After the
new NBY, we see that NBY member’s tendency to repurchase is much higher
than the control group. Besides, NBY member shows the lower UPT/ADPT
values comparing with the control group before new NBY, but the UPT/ADPT
value gaps are gone after the new NBY.
Modeling
1. HOI NBY modeling
From the EDA part, we have conclusions as shown below:
1. NBY members show the younger age trend comparing with non-NBY
members. NBY members have much higher percentage (8%) on age group,
‘0-17’, which is four times higher than that of non-NBY members.
2. The NBY members show much stronger interests in Lifestyle and less
interests in sneaker head than the non-NBY members.
Let’s verify these findings from the modeling.
Variables to be used in training:
• Gender
• Age Group
• Member Segment - Lifestyle
• Member Segment - SNKRHEAD
• Member Segment - Performance
• Member Segment - Family
• High Value Member - Member who purchased $375 or more in a year
• Multi color - Member who purchased products with 3 or more colors in a
year
• combo - Member who purchased products with 2 or more divisions in a
year
• Dependent Variable: Buy NBY or Not (Binary Variable) - processed from
HOI purchase history
Algorithm used: Logistic Regression Classification
Logit(P(Y=BuyNBY(1)))=log(p1−p)=β0+β1Gender+β2AgeGroup+β3Lifestyle+β4Performance+β5SNKRHEAD⋯+βnHighValue
Logistic Regression is commonly used for binary classification
problems. one advantage of this algorithm is that it gives β for each
term, and the signs can explain the effect of each term on dependent
variable.
To check the relationships among all the variables, we create the predictive power score plot.
From the plot, we can’t find any significant variables to predict
if_NBY_member. We will put all the variables in the model and check the
significance of them.
Model result
| Variable | Log_Odds.β. |
|---|---|
| age_0_17 | 3.84 |
| Lifestyle | 3.69 |
| Male | 3.04 |
| Female | 2.71 |
| if_high_value | 2.17 |
| if_combo | 2.09 |
| age_18_24 | 1.4 |
| SNKRHEAD | 0.59 |
| age_46_60 | 0.46 |
The model result confirms our understanding from the previous
chapter. The member with younger age and ‘lifestyle’ segment is more
likely to purchase NBY. Female is slightly more likely to purchase NBY
than male, but the difference is not very significant. The member who
like buying combo and with high values is more likely to purchase NBY.
The Sneaker head and the elder members don’t show much interests in
NBY.
Model performance
McFadden
The model’s McFadden r^2 value is 0.122, which is decent enough for a logit regression.
ROC
Testing set (25% of the sample data) is used for measuring model’s prediction performance, and a ROC is created.
The ROC Plot:
The model achieved an AUC of 0.7169. The model can be used for both
explanatory and predictive purposes. The model accuracy on the test set
is 0.6656.
2. Online NBY modeling
From the online NBY EDA part, we have conclusions as shown
below:
1. Female shows more interests in NBY than male. And similar to HOI, NBY
members with the younger age are more likely to buy NBY, but the trend
is not strong as HOI.
2. Most of the online NBY members are classifid as Family segment. And
similar to HOI, the NBY members show less interests in sneaker head than
the non-NBY members.
Similarly, we use Logistic Regression Classification to analyze the
variables which could affect or predict if a member buys NBY or not for
online platforms.
Variables to be used in training:
• Gender
• Age Group
• Member Segment - Lifestyle
• Member Segment - SNKRHEAD
• Member Segment - Performance
• Member Segment - Family
• buyer_type - LastYear
• buyer_type - Lapse
• buyer_type - ActiveNonBuyer
• High Value Member - Member who purchased $375 or more in a year
• Multi color - Member who purchased products with 3 or more colors in a
year
• combo - Member who purchased products with 2 or more divisions in a
year
• Dependent Variable: Buy NBY or Not (Binary Variable) - processed from
HOI purchase history
Algorithm used: Logistic Regression Classification
Logit(P(Y=BuyNBY(1)))=log(p1−p)=β0+β1Gender+β2AgeGroup+β3Lifestyle+β4Performance+β5SNKRHEAD⋯+βnHighValue
Here’s the predictive power score for if_NBY_member:
We can see that the features, if_high_value and segment have strong
predictive power score for if_NBY_member, which means they might be
quite significant in the predictive model. The features,
preferred_gender and age_group also show some effects on
if_NBY_member.
Model result
| Variable | Log_Odds… |
|---|---|
| Family | 18.66 |
| Lifestyle | 2.14 |
| buyer_type_LastYear | 1.95 |
| Female | 1.84 |
| if_multi_color | 1.74 |
| if_high_value | 1.37 |
| age_31_35 | 0.61 |
| age_36_40 | 0.39 |
| SNKRHEAD | 0.34 |
| age_41_45 | 0.34 |
| age_46_60 | 0.22 |
Similar to the results of HOI modeling, the member with ‘lifestyle’
segment is more likely to purchase NBY and the Sneaker head doesn’t show
much interests in NBY. The members who like buying multi color products
and with high values are more likely to purchase NBY online. Besides,
the members with elder ages are less likely to purchase NBY, which is
align with the conclusion we got from HOI modeling.
What’s different from HOI is, the female has more interests in NBY
through online shopping than male, while the gender difference for HOI
is not that significant. Besides, the member with ‘family’ segment is
greatly significant to the prediction of if_NBY_member.
Model performance
McFadden
The model fits pretty well, with a McFadden of 0.3537, which is good for a logit regression.
ROC
Testing set (25% of the sample data) is used for measuring model’s prediction performance, and a ROC is created.
ROC Plot:
The model achieved an AUC of 0.8235. The model can be used for both
explanatory and predictive purposes. The model accuracy on the test set
is 0.7855.
Conclusion
• The members with ‘lifestyle’ segments are more likely to purchase
NBY and the Sneaker head doesn’t show much interests in NBY
• The members with younger ages are more likely to purchase NBY. And for
online platforms, female shows stronger interests than male.
Acknowledgment
Thank Yunhai Zhang for the supervision and guidance.
Source: Nike Inline Order Line↩︎
Source: Nike Membership Data↩︎
Source: Nike member science team↩︎
Based on the scalper detection machine built by Yunhai Zhang (Yunhai.Zhang@nike.com)↩︎