Exploratory Analysis of Inactive Customer Dataset

Cancellation reason

After cancelling their subscription, customers are sometimes called by a Full Circle representative to provide a reason for cancellation (recorded in latest_cancel_reason_quit. If the customer provides an answer, the representative selects the reason that most closely matches the customer’s answer.

ggplot(inactive_filtered, aes(x=latest_cancel_reason_quit, y=weeks_subscribed, 
                     fill=latest_cancel_reason_quit)) + 
  geom_boxplot(outlier.shape=NA) + 
  guides(fill=FALSE) + 
  coord_flip(ylim=c(0,60)) + 
  labs(title="Avg. Weeks by Cancel Reason", x="avg. weeks subscribed", y="cancel reason")

• Note: outliers removed from fig. 2
• A density graph was chosen with a log scale to better visualize the shape of the distributions.
• term completed and end of prepay / gift cert customers tended to have the shortest subscription periods. FC has stated customers fall off a cliff after a customer’s prepaid subscription ends.
• customers who cancel because they are moving or the pickup site has closed are more likely to be long term customers, because they probably would’ve cancelled for another reason prior this. Many of these customers are past the initial “stay or leave” hurdle, and would likely have been strong long term customers. If pickup sites have re-opened, or if those customers moved to a FC location, reconnecting would be a valuable effort.

Signup Promotion

Full Circle tracks how customers begin their subscription by marketing channel via initial_signup_promo_category.

ggplot(inactive_filtered, aes(x=initial_signup_promo_category, y=weeks_subscribed, 
                     fill=initial_signup_promo_category)) + 
  geom_boxplot(outlier.shape=NA) + guides(fill=FALSE) + 
  coord_flip(ylim=c(0,60)) + 
  labs(title="Avg. Weeks by Signup", x="avg. weeks subscribed", y="signup category")

ggplot(inactive_filtered) + geom_bar(aes(fill=initial_signup_promo_category, x=latest_cancel_reason_quit), position="fill", binwidth=30) + ggtitle("Membership Changes over Time") + coord_flip() + scale_fill_brewer(palette="Set2")

• Note: outliers removed from fig. 4
• Our FC contact said that they have a good grasp on promotions, but regardless it’s helpful for data to reinforce their knowledge.
• What is a successful customer? FC contact mentioned that even though Daily Deal customers have a shorter membership span, still considered successful.

Distribution Center

ggplot(inactive) + geom_density(aes(x = weeks_subscribed)) + 
  facet_wrap(~ distribution_center, ncol = 2) + 
  scale_x_log10(breaks = c(1,10,100)) + 
  ggtitle("Fig 5. Density (log): Dist Center")

• Less competition in Alaska, tend to have longer memberships there.
• Lower priority data set to investigate.

Grocery Region

ggplot(inactive) + geom_density(aes(x = weeks_subscribed)) + 
  facet_wrap(~ grocery_region, ncol = 2) + 
  scale_x_log10(breaks = c(1,10,100)) + 
  ggtitle("Fig 6. Density (log): Grocery Region")

• See notes on distribution center.

Membership History

ggplot(membership) + geom_bar(aes(x=timestamp, fill=history), position="fill", 
                              binwidth=30) + 
  scale_fill_manual(values=c("#CC6666", "#9999CC", "#66CC99")) + 
  ggtitle("Membership Changes over Time")

• Just started digging into this data set.
• Questions:
  • Can we identify what the spikes in reactivations are (early 2010, 2012)? Some sort of new year marketing campaign?
  • Overall, seems like a negative picture: few new members, mostly reactivations. How can FC find new audiences?
• To do:
  • prefer area plot over bar. Change.
  • get a better sense of proportions over time.

Credits

The data sets

The following analysis “credit” related variables was performed on three data sets:

1. Inactive
2. InactiveW
3. Active

Each data set consists of rows that represent customers of Full Circle.

The Inactive data contains subcribers who cancelled. This data set was narrowed from 126,439 rows by applying the following criteria: 1. These subscribers cancelled once and only once.
2. They cancelled AND signed up between 3/1/2012 and 3/1/2014. Any customer that both subscribed and dropped within this two-year window. 3. Customers who were never invoiced – i.e. a “0” under invoice_count were removed.

The resulting “inactive” data set has 34,216 customers (“observations”) with a median subscription length of 4.71 weeks.

The InactiveW data set is a subset of Inactive. It consists of 16,914 customers who subscribed for less than 4.71 weeks – all customers below the median of the larger Inactive popultion. These represent the most dissatified customers.

Line of inquiry

To determine whether there is a relationship between credits received and length of subscription. We use invoice count as a proxy for subscription length since it more accurately reflects the value of a customer and excludes weeks when a subscription is put on hold.

credits_count

Summary statistics, plots, and correlations – per each of the three data sets.

Inactive:

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##  0.0000  0.1111  0.4000  0.5061  1.0000 13.0000

## [1] -0.6612984

InactiveW:

## Length  Class   Mode 
##      0   NULL   NULL

## [1] -0.5363024

Active:

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
## 0.00000 0.03030 0.06250 0.09675 0.11950 1.25000

## [1] -0.06023276

Exit Survey