BETA DEMO SCENARIO - ATTRITION 1
Please Note: IBM s statements regarding its plans, directions, and intent are subject to change or withdrawal without notice at IBM s sole discretion. Information regarding potential future products is intended to outline our general product direction and it should not be relied on in making a purchasing decision. The information mentioned regarding potential future products is not a commitment, promise, or legal obligation to deliver any material, code or functionality. Information about potential future products may not be incorporated into any contract. The development, release, and timing of any future features or functionality described for our products remains at our sole discretion. Performance is based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput or performance that any user will experience will vary depending upon many factors, including considerations such as the amount of multiprogramming in the user s job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve results similar to those stated here. 2
PCA: designed for the marketer and channel owner Evan is responsible for designing and executing personalized, effective interactions across his channel s offerings, but he struggles with: Quickly and clearly understanding the most important things about his customers so he can target more effectively Relying too heavily on data scientists to do his job Grouping/segmenting customers in meaningful ways Identifying the most valuable/impactful action to take Translating insights into actionable, targeted campaign lists Evan needs answers to specific questions: What is important to know about my customers? Am I picking the right customers for a campaign? Which customers are most valuable? Who is at risk of leaving and why? How do I anticipate segment behavior? 3
Evan logs on and lands on the Predictive Starting Points page He sees a single, linear graph that recommends five lists of customers that are at risk of attrition. These lists are automatically generated using predictive models that reveal top indicators that a customer is at risk. The lists are displayed as bubbles indicating population size and have various profile makeups based on a combination of key predictive drivers and other profile attributes. Evan can see the the number of customers in the list, overall attrition score, and average CLTV are displayed when he clicks on a bubble. Understand Details How might Evan use this insight? These recommended lists help Evan quickly identify groups of customers at risk for different reasons. This is a great first step in breaking down all at-risk customers into smaller groups and assigning the most effective retention offers. It also helps in prioritizing outreach: perhaps Evan wants to focus on the largest population, or the one at highest risk, or the most valuable. 4
Evan first wants to examine each of the lists to understand profile makeup Then he ll be able to decide the best course of action for each one. PCA uses predictive models to uncover predictors of attrition, so these lists may change as the models refresh and reveal new insights. This means Evan can always stay one step ahead of his customers at risk and know the best way to intervene. Today s Recommended Lists show groups of customers who: 1. Are at a very high risk of churning (more than 80% likely) 2. Churn risk has increased by more than 50% since last month 3. Have made more than two complaints in the last month 4. Have either abandoned a cart or returned items and haven t logged back in after two weeks 5. Are normally active, but whose average spend and login activity have been decreasing Understand Details How might Evan use this insight? Evan can click on a bubble and see the profile makeup of each list, the average CLTV of the customers in that list, and the number of customers. This helps him evaluate the general impact of attrition each group presents. By looking at the makeup of the lists and understanding the underlying predictors, he can understand where and how he might take action. 5
Evan spots the Code Red list as an important audience Evan can easily see that the first recommended list, the Code Red customers, are not only at a very high risk of attrition, but that they are also the second largest group of customers. He clicks on the bubble to reveal the details. The Code Red list includes 418,513 customers with an average attrition score of 87 and average CLTV of $139.02. Understand Details How might Evan use this insight? Evan can immediately see that a sizeable number of customers are at risk, and he needs to act quickly. 6
Evan wants to examine the segments within the Code Red list With the Code Red list selected, Evan is ready to apply a segment lens. Because Evan is preparing to send an email campaign, he wants to divide out groups of customers based on email behavior. He scrolls down to the Customer Composition chart and now sees segments of Code Red customers arranged by Attrition Score and CLTV. As with the Predictive Starting points, the size of the bubble indicates relative population density of each segment. Clicking on each bubble will reveal the name of the segment, number of customers, Average attrition score, and average CLTV. Evan is using these engagement segments: Enthusias ts Sleepy Phantom Ignored Newbie Opt-out Contacts opted in to email communication, open at least 2/3 of emails that they receive, and are among top 10% in likelihood-to-engage Contacts opted in to email communication, have opened at least one email they ve received, and are among the bottom 20% in likelihood-toengage Contacts opted in to email communication, have been sent at least one email but never opened any email, and have a moderate predicted likelihood to engage Contacts opted in to email communication, are active customers, but have not received any emails Contacts opted in to email communication, have made a purchase, but have not received any emails contacts who have opted out of all email Mainstreet Contacts who don t fall in any of the above groups 7 How might Evan use this insight? The segment definition helps Evan identify those customers most likely to receive/respond to an email offer that is, if email might be an effective channel. This again helps Evan narrow his target lists and prioritize where to make a marketing investment.
Evan follows one of the PCA insights for recommended actions Above the bubble chart, Evan will be offered three guiding insights that guide him toward key areas for action, based on factors like acute revenue impact or missed opportunity. Clicking on one of the three insights will highlight the segments of interest. Focus on Biggest Opportunity These customers within the Phantom segment are your biggest opportunities for growth/upsell. Focus on Most Valuable These customers within the Ignored segment will most significantly impact your revenue. Focus on #1 Driver of Attrition These customers within the Sleepy segment have had more than two complaints in the past month. 8 How might Evan use this insight? The bubble chart and guiding insights helps Evan identify where the most impactful opportunities lie so that he can prioritize campaign activities. The insight buttons will help Evan focus on the largest populations in the upper right (high churn risk + high value) and upper left (high churn risk + low value) quadrants of the bubble chart, plus the segment containing customers demonstrating the #1 attrition predictor. Within those quadrants, segment definition helps identify those customers most likely to receive/respond to an email offer based on various profile attributes. This will help Evan further narrow target campaign lists.
Evan can now identify audiences for specific campaigns By understanding each group s attrition risk, CLTV, and engagement segment in concert, Evan can now identify appropriate and effective campaigns for each and consider where he might make a marketing investment. Sleepy Phantom Ignored Contacts opted in to email communication, have opened at least one email they ve received, and are among the bottom 20% in likelihood-toengage Contacts opted in to email communication, have been sent at least one email but never opened any email, and have a moderate predicted likelihood to engage Contacts opted in to email communication, are active customers, but have not received any emails How might Evan use this insight? Evan can tailor campaign nurture activities more effectively now. The Phantom segment has opted in to email but has never opened any, but they do have a moderate predicted likelihood to engage. This suggests that email may still be an effective channel, provided the subject line is compelling. Many of these customers may be directing communications from Evan s company to an unmonitored ( junk ) address, in which case Evan needs to be sure to record campaign response and continue to monitor effectiveness of email offers. With any luck, these customers will migrate to a more actively engaged segment. Evan can see that his company has been ignoring one of his most valuable groups of customers! These customer have opted in to email, but never received any offers. Evan needs to send them something right away! The Sleepy group is quite valuable and these customers are disgruntled. While some emails have made it through to this group, it s not proven to be a very effective channel, and they re unhappy on top of it. Evan might consider a more hightouch, personal campaign nurture event for this group. And because this group is relatively small, investing in a more personal interaction may be worthwhile. 9
Evan digs deeper by examining the Key Drivers of Attrition Evan is ready to take an even closer look at the Code Red customers. He scrolls down to the Key Drivers of Attrition graph, which tells him the top five predictors of churn that the predictive models revealed behind the scenes. He can see that the number one predictor, complaints per month, is a very heavily weighted indicator that a customer will leave, followed closely by abandoned cart items. Below the insights, Evan can explore a heat map which draws his attention to the greatest concentrations of customers who have are at risk of churn due to each attrition predictor. How might Evan use this insight? This chart helps Evan quickly spot thresholds where single attrition drivers present material impact to overall churn score. In this example, all Code Red customers already have an elevated churn score (>80), but Evan can see that there are quite a few customers with excessive numbers of complaints and a large number of abandoned cart items. This may help him identify the more acute risks and further refine target lists for a retention campaign. 10
Evan creates lists for his selected target audiences Now that Evan has a deeper understanding of the attrition drivers for the most at-risk customers, he s ready to create some targeted lists and launch a retention campaign! Going back to the insight guidance from the Customer Composition chart, he s going to focus on the Ignored, Phantom, and Sleepy segments within Code Red. Instead of sending the same offer to all Code Red customers, he is going to create lists by segment. This is so that he can select more relevant content for each segment perhaps a welcome offer for the Ignored segment and a bundle discount offer for the Phantom group. From within the Identify screen, he clicks on New Customer List. Because he knows that Code Red customers have a churn score >80, his custom list simply selects the segment (Behavioral Attributes) and the churn score (Model Attributes) for Phantom and Ignored. But Evan wants to do something a little different for his Sleepy group, because they re also demonstrating the #1 driver of attrition: customer complaints. 11
Evan creates a special, more refined list for the Sleepy segment Evan thinks that the Sleepy segment should get a little more attention. He will still send an email offer, but he also wants to generate a list for a targeted phone campaign. He starts by creating the Code Red Sleepy list. Unfortunately, this list has too many people in it for a phone campaign to be practical. He s going to have to focus and be a little more selective. So, recalling that the second most important driver of attrition is abandoned items in cart, he decides to refine the list based on on customers with the highest numbers of abandoned items. The list is still a little bit too big, so he narrows again to focus on higher value customers. Voilà! The list now only has 473 customers certainly manageable for their large customer service department to handle. And with an average churn score of 91 and CLTV of $171, this group is worth the extra investment in a phone campaign. The heat map helps Evan see the concentration of customers for each of the key predictors 12
Evan exports his target audience lists for use in upcoming campaigns Now all Evan has to do is export his target lists. With the Sleepy, Grouchy Spenders list selected, he clicks on Act and selects Essential Data to generate a.csv file that can then be forwarded to the customer care department for the phone campaign. His other lists will be exported for use in targeted email campaigns. And since he tagged them all as Evan the Marketing Superhero, all of his lists appear in one, easy, organized screen. 13
To be continued 14
APPENDIX 15
PCA Facets Facet Description Value to marketer Predictive Starting Points Single, linear graph that recommends five lists of customers plotted by risk of attrition. Lists are displayed as bubbles indicating population size and have various profile makeups based on a combination of key predictive drivers and other profile attributes. Number of customers in the list, overall attrition score and average CLTV are displayed when a bubble is selected. Generates lists based on predictive analytics vs. rules Lists are dynamic will change as models refresh and predictors change Marketer could just go ahead and export these lists directly! Customer Composition Drivers of Attrition Bubble chart that plots Attrition Score (Y) by CLTV in USD (X) by Density of Customers for defined segments in the selected list. Above the bubble chart will appear three text-based insights that help marketers guide decisions around target audience selection, based on various priorities. Predictive models reveal key insights about drivers of attrition, displayed as short text-based insights. An accompanying heat map helps marketers quickly see the impact of at-risk groups, plotted as Attrition Score (Y) by CLTV in USD (X) by Density of Customers for defined segments in the selected list. Placement of bubbles helps marketers easily see where the most impactful opportunities are (e.g. upper right, need high-touch intervention; lower right, all good no investment necessary; lower left, ignore; upper left, opportunity for upsell) Breaks down the selected list into smaller segments for more targeted outreach (can help focus marketing budget) Quickly and easily reveals and explains drivers of attrition and affected groups; marketers can make targeted campaign decisions addressing single attribution drivers. 16
Predictive Starting Points List Name Description Fields CODE RED SPIKERS DISGRUNTLED GHOSTERS FADERS At very high risk of churning (any combination of predictors) Customers whose churn risk has increased by more than 50% since last month All with >2 complaints in last month (#1 attrition driver) All who have either abandoned a cart or returned items and haven t logged back in after two weeks Normally active customers whose average spend and login activity have been decreasing CHURN_SCORE >.80 CHURN_DELTA_MON > 50% COMPLAINT_COUNT_CURR_MON > 2 [ABANDONED_CART_ITEM_COUNT >0 OR RETURN_COUNT_PREV_MON >1] AND DAYS_SINCE_LAST_LOGIN > 14 ACTIVE_CUSTOMER_IND=1 AND DAYS_SINCE_LAST_LOGIN > 45 AND AVG_SALES_VALUE_DELTA_CURR_QTR >.40 not sure how to write this as a negative value! 17
Segment Definition Segment Name Suggested Description Fields What does this tell a marketer? Enthusiasts Contacts opted in to email communication, open at least 2/3 of emails that they receive, and are among top 10% in likelihood-to-engage EMAIL_RESPONSE_IND=1 LIKELIHOOD_TO_ENGAGE_PCTL =.90 [RESPONSE_BY_EMAIL_COUNT / These customers don t need a lot of attention email has been an effective way to reach them. Sleepy Phantom Ignored Contacts opted in to email communication, have opened at least one email they ve received, and are among the bottom 20% in likelihood-toengage Contacts opted in to email communication, have been sent at least one email [in the last 60 days?] but never opened any email, and have a moderate predicted likelihood to engage Contacts opted in to email communication, are active customers, but have not received any emails OFFER_BY_EMAIL_COUNT >.66] EMAIL_RESPONSE_IND=1 LIKELIHOOD_TO_ENGAGE_PCTL =.10 OFFER_BY_EMAIL_COUNT > 0 RESPONSE_BY_EMAIL_COUNT >0 EMAIL_RESPONSE_IND=1 LIKELIHOOD_TO_ENGAGE_PCTL >.65 RESPONSE_BY_EMAIL_COUNT = 0 OFFER_BY_EMAIL_COUNT >0 EMAIL_RESPONSE_IND =1 OFFER_BY_EMAIL_COUNT = 0 ACTIVE_CUSTOMER_IND=1 These customers are worth sending an offer to; they may reengage with the right offer. Email may not be effective medium. Could possibly expand this segment to look at web/sms channels. Would not be in scope for email list segments, but could be used as another way to select offers SMS_RESPONSE_IND WEB_RESPONSE_IND Send these customers an offer! Newbie Opt-out Contacts opted in to email communication, have made a purchase, but have not received any emails contacts who have opted out of all email EMAIL_RESPONSE_IND =1 OFFER_BY_EMAIL_COUNT = 0 ACTIVE_CUSTOMER_IND = 1 Or do we use PURCHASES_ONLINE_QTY >0 EMAIL_RESPONSE_IND =0 Send these customers an offer! These customers should be excluded from any list selection. Could possibly expand this segment to look at web/sms channels. Would not be in scope for email list segments, but could be used as another way to select offers SMS_RESPONSE_IND WEB_RESPONSE_IND Mainstreet 18 Contacts who don t fall in any of the above groups