12
Using Interaction Analytics To Reduce Customer Churn

Predictive Analytics And Churn

Embed Size (px)

Citation preview

Page 1: Predictive Analytics And Churn

Using Interaction Analytics

To Reduce Customer Churn

Page 2: Predictive Analytics And Churn

Why Churn? Price CX

Page 3: Predictive Analytics And Churn

Who

To T

arge

tLim

ited Resource

Page 4: Predictive Analytics And Churn

demographics

t ransactions

inte

racti

on

s

Page 5: Predictive Analytics And Churn

PROCESS

• Agent Tech Troubleshooting

• Appointments

• Offer Expiration

• Service not restored

• Equipment not received

• Service not activated

• Installation pending

• Technical Issues

• Outage

• Broken Appointments

• Service Restricted

• Number not ported

INTERACTIONS EARLY INDICATORS PRODUCT/PRICE

• Call Reason Categorisation

• Average Handle Time/Reason

• Non-Talk Time/Holds

• Long/Short Call Reason

• First Call Resolution

• Transfers

• Escalation Reason

• Agent Performance

• Channel Drivers

• Agent Tools/Downtime

• Disconnects

• Escalation Requests

• Customer Satisfaction

• Offers Made/Offers Taken

• Retention Saves

• Appropriate & Bad Credits

• Downgrades

• Regulatory/Compliance

• Competitive Discussion

• Billing Line Item Review

• Rude Behavior

• Residential Move

• Service Performance

• Competitive Pricing

• Equipment Failure

• Ease of Use/Navigation

• Features and Setup

• Usage

• Speed

• Fees

Interaction Clues From

Triple Play Customers

Page 6: Predictive Analytics And Churn

THE INTERACTION

Why CallingIssueProduct/Price/OfferSentiment

THE CUSTOMER

BillingProduct HoldingCRMPayment MethodPromotion

The Customer LifecycleData Collected

MARKET SELL BILL SERVE MAINTAIN

CUSTOMER INTERACTIONS

O v e r ti m e w e c o l l e c t a l o t o f u s e f u l i n s i g h t

Page 7: Predictive Analytics And Churn

Making Churn Modeling 25-30% More Accurate

Run established churn models

Identify the interaction characteristics of previously churned customers

Run that profile against all interactions to identify those at risk

Export into suitable format for data warehouse

Match with transactional and other metadata

The Customer LifecycleData Collected

Page 8: Predictive Analytics And Churn

Page 8

Predictive Churn Advanced AnalyticsMaking Churn Modeling 25-30% More AccurateThe Customer LifecycleData Collected

Interactions

CONTINUALLY REFINED

Queries

Defined Words and PhrasesCombined with Operators

Provides Rich Meaning

Utilises Machine Learning toIdentify Non-Intuitive Words

and Phrases to Improve Results

Neural Phonetic Speech Analytics

Metadata

Datasets are Combined,Transformed to Account Level,

and Joined with Metadata

What Was Said

What Was

MeantWhat Was Done ρX,Y = corr

Algorithm Development

PEOPLE WHO WILL STATISTICALLY CHURN Based

on their interactions and basic customer data

(LIST, INTERFACE…)

Focusing on transactional data combined with what they said in

their most recent interactions

Page 9: Predictive Analytics And Churn

Making Churn Modeling 25-30% More Accurate

So far tested with Comcast, Time Warner, Verizon and BT

Improvement to accuracy of existing churn model 25-30%

Example Comcast have 20m customers. 0.5 m churn every month

Predictive Churn Advanced AnalyticsMaking Churn Modeling 25-30% More AccurateThe Customer LifecycleData Collected

Page 10: Predictive Analytics And Churn

0%

20%

40%

60%

80%

100%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Page 10

Combined Model

Interaction Data Model

Traditional Data Model

Percentage of Non-Churn

True

Pos

itive

Rat

e

Case Study: The impact of adding Speech Analytics data to a large set of Traditional data for Churn Prediction in a mitigation program with ~20 FTEs.• The team would be able to contact ~450k accounts/year, equivalent to ~.74% of all unique accounts

that call each year. Results assume a $25 Cost per account contacted for cost of call and offers.

.74% of Population

Interaction Data Only Result

Structured Data Only Result

Combined Data Result Identified = 45%Cost = $10M/AnnuallyOpportunity = $187M/AnnuallyAt 20% Save Rate = $37M/Annually

Profit At 20% = $27M/Annually

Identified = 31%Cost = $10M/AnnuallyOpportunity = $120M/AnnuallyAt 20% Save Rate = $26M/Annually

Profit At 20% = $16M/Annually

Identified = 27%Cost = $10M/AnnuallyOpportunity = $114M/AnnuallyAt 20% Save Rate = $23M/Annually

Profit At 20% = $12M/Annually

0 50,000 100,000 150,000 200,000 250,000 300,000

0 50,000 100,000 150,000 200,000 250,000 300,000

0 50,000 100,000 150,000 200,000 250,000 300,000

Identified (113,784)

Not Identified (308,757)

Identified (130,374)

Identified (186,826)

Not Identified (291,311)

Not Identified (230,807)

$ Conservatively $11M in Straight Operating Income

(with significantly less effort / cost) =Correctly identified CHURN accounts

Identified accounts that Will Not Churn

Making Churn Modeling 25-30% More AccuratePredictive Churn Advanced AnalyticsMaking Churn Modeling 25-30% More AccurateThe Customer LifecycleData Collected

Page 11: Predictive Analytics And Churn

Add Interaction DataImprove Predictive

Modelling25-40%

Page 12: Predictive Analytics And Churn

Keep More CustomerFor Longer

-:)