Using Models & Analysis to Optimize Expense October 29, 2015
William “Brad” Bradley Senior Vice President Retail Credit Risk Management
SunTrust Auto Portfolio ~$9 billion portfolio Primarily originated though dealers Very high credit quality High FICO Low DTI Low LTV
Low charge-off rates Charge-offs increased during recession, but declined in recent years Many loans that go early stage delinquency are just “sloppy payers” who cure prior to rolling to the next bucket
Collections Team & Credit partnered to update the collections strategy 2
Previous Collections Strategy Distribution of Dials by Delinquency Bucket 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
90%
71% of dials through a live agent channel (dialer, manual, etc.) 29% of dials were through an automated channel
8% 1-29
30-59
1%
1%
0%
60-89 90-119 Delinquency Bucket
120+
Dials prior to the strategy change were focused on early stage delinquency – Goal was to catch loans early to prevent late stage later Required a large collections team to handle call volume
3
Roll Rates Roll Rates of 1-29DPD Bucket 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
95%
5% Cure/Remain
Roll to Worse Next Month Status
Most of the accounts we were dialing in the 1-29DPD bucket were curing Indicates we were spending a lot resources on loans that would cure anyway
Decided that a model was needed to predict which loans in the 129DPD bucket would roll to 30-59DPD Model could be used to guide strategy
4
Model Solution Roll Rates to 30DPD by Model Decile 25%
22% 85% of accounts that roll to 30DPD are captured in 50% of the population
20% 15%
17%
12%
10% 5% 1%
1%
2%
1
2
3
3%
4%
6%
6%
0% Lower Risk
4
5 6 Model Decile
7
8
9
10
Higher Risk
We developed a model that predicted the probability of a loan in the 1-29DPD bucket rolling to 30-59DPD Model incorporates credit variables as well as recent performance (# of times 1-29DPD in past 6 months) DPD also in the model so as the account moves closer to 30DPD, the probability of roll increases 5
Collections Levers Lever
Calling Strategy
Skill Based Routing Intensity Placement Strategy
Options • • • •
Call/No Call Live Agent/Automated # of Attempts Skiptrace Efforts
• High Risk Accounts to Above Average Collectors • On-shore, Off-shore • # of Calls Per Month • Work, Place, Sell, Warehouse
We have multiple options for implementing our calling strategy The model can tell us which loans are most likely to roll but we need to test to determine the best strategy 6
Test Design All 1-29DPD Accounts
5% All Live Agent
5% All Automated
90% Existing Strategy
Randomly selected by last 2 digits of Loan Number
10% of our accounts were randomly selected to go to test Represented accounts in all risk deciles
Test was run for 3 months, with a review of results afterwards Daily monitoring occurred to ensure Operations was executing test Previous instance where Operations Managers did not follow strategy leading to tainted results 7
Test Results – 30 Day Window Roll to 30+ Rate – Agent vs. Automated Test Groups Agent
30.0%
Automated 27.3%
25.0%
22.4%
20.0%
17.4% 16.7%
15.0% 10.0% 5.0%
9.1% 6.7%
1.1% 0.9%
1.8% 1.3%
3.4% 2.2%
4.6% 3.3%
5.7% 5.5%
11.7% 10.5%
6.4%
6.1% 5.0%
3.9%
0.0% 1
2 Lower Risk
3
4
5
6
7
8
9
10
Total
Higher Risk
Reviewing our results after 30 days show that our live agent channel performs better than the automated channel A few deciles showed flipped results – this was attributed to small sample sizes It appears as though we should be sending all accounts through our live agents… 8
Test Results – 60 Day Window Roll to 30+ Rate – Agent vs. Automated Test Groups Agent
30.0%
Automated
25.0% 20.0% 15.0%
12.3%
10.0% 5.0%
9.1%
1.0% 1.0%
2.2%
3.1%2.7%
4.4% 3.5%
5.7% 3.7%
4.3%3.6%
6.3% 4.6%
4
5
6
7
13.0% 11.9%
8.5%
6.4% 4.1%4.3%
1.1%
0.0% 1
2 Lower Risk
3
8
9
10
Total
Higher Risk
After a loan rolls to 30+DPD, it automatically receives agent treatment
This higher treatment leads to more cures and after 60 days, the number of 30+DPD loans is almost the same between channels 9
Test Results – Charge-Off Charge-Off Rate – Agent vs. Automated Test Groups Agent
4.0%
Automated 3.5% 3.5%
3.5%
3.3% 3.2%
3.0% 2.5%
2.2%
2.0% 1.5% 1.0% 0.5% 0.0%
0.0% 0.0%
1
0.3%
0.3%
2 Lower Risk
0.9% 0.6%
0.6% 0.2%
3
4
0.8%
0.9%
1.0% 0.9%
1.1%1.1% 0.8%
0.4%
0.2%
5
6
7
8
9
10
Total
Higher Risk
Charge-off levels were essentially the same between the two test groups We performed a cost benefit analysis and decided that we should implement the following: Reduce call volumes to 1-29DPD Switch call volumes in 1-29DPD to automated channels 10
New Collections Strategy Distribution of Dials by Delinquency Bucket 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
Old Strategy
90%
New Strategy
81%
8% 1-29
16%
30-59
1% 2%
1% 1%
0% 0%
60-89 90-119 Delinquency Bucket
120+
Call volumes moved away from early delinquency and towards higher delinquency
Calls that are made in the early stage delinquency buckets are focused on the high risk accounts as determined by the model 11
New Collections Strategy Distribution of Dials by Channel 1-29DPD Bucket 80% 70%
71%
Old Strategy
New Strategy
76%
60% 50% 40% 30%
29%
24%
20% 10% 0% Live Agent
Automated Call Channel
Call mix moved from mostly live agent to mostly automated Win/win for customer and bank Customer: Early delinquency now receives softer courtesy automated reminder call rather than a call from a collections agent Bank: Reduced expense as automated calls are much less expensive, plus agents have more time to focus on solutions for customers who actually need help (rather than just a reminder to pay) 12
Strategy Results
As expected, after implementing the strategy, our 30-59 DPD delinquency rate increased by 40%
The new strategy did cause an uptick in our 30-59DPD rates Lesson learned – Remind executives multiple times prior to strategy implementation and after implementation so they understand this is going to happen 13
Strategy Results
As expected, our 60-89 DPD and 90-119 DPD delinquencies remained flat other than some normal seasonal changes This translated through to no impacts to charge-offs
14
Expense Savings Product
Expense Savings
C/O Impact
DDA
80%
$0
Auto
45%
$0
Real Estate
20%
$0
Credit Card
TBD
TBD
Recoveries
TBD
TBD
We conducted this type of modeling and testing across all of the major portfolios and have generated significant savings thus far Next steps include continuing to test the various levers we can pull to optimize expense while holding C/O down 15