Predictive Models

```WORKERS COMPENSATION PREDICTIVE
MODELING: THE CRYSTAL BALL
BECOMES CLEARER
RIMS SESSION RIF 010
WEDNESDAY, APRIL 30, 2014
2:00 P.M. TO 3:00 P.M.
Recording of this session via any media type is strictly prohibited.
Page 1
TODAY’S PRESENTERS
Melissa Bowman-Miller, Staffmark
David Duden, Deloitte
Sean Martin, Travelers
Jeff Branca, Marsh
Recording of this session via any media type is strictly prohibited.
Page 2
Today’s Agenda
Introductions & Housekeeping
Defining Predictive Modeling
Risk Manager’s Perspective
Insurer’s Viewpoint
Consultant – Bridging the Gap
Questions & Discussion
Recording of this session via any media type is strictly prohibited.
Page 3
What Differentiates Claims Organizations?
“If we want to make better decisions and take the right actions, we have to use
analytics. Putting analytics to work is about improving performance in key business
domains using data and analysis.”
- Tom Davenport, author of Analytics at Work: Smarter Decisions, Better Results
5-10%
Can create new algorithms
Analytical semiprofessionals
15-20%
Can use visual and basic statistical tools, create simple
predictive models
Analytical amateurs
70-80%
Can use spreadsheets and use analytical transactions
Insight
Analytical professionals
True Claims
Predictive
Modeling
Hindsight
1%
Foresight
Analytical champions
Recording of this session via any media type is strictly prohibited.
Page 4
PM Helps Organizations Target High Exposure Claims
 When a claimant’s injury is a sprained back, there is a wide and varying distribution of claim outcomes
 The worst 20 - 30% of claims contribute to 70 - 80% of loss costs
 PM uses a variety of data sources and analytics techniques to enable organizations to predict which claims
are most likely to be the worst claims
 The graph below shows the varying distribution in total lost days for back sprain injuries
40%
Injury: Back Sprain
Pct of Total Losses
35%
30%
25%
20%
15%
10%
5%
0%
1
2
3
4
5
6
7
8
9
Recording of this session via any media type is strictly prohibited.
10
Page 5
Can Be Predictive
Insights can be revealed through both traditional and non-traditional risk characteristics. Even use of a relatively
small set of predictive variables can enhance claim segmentation.
100%
Relative claim severity
80%
Claimant Age
60%
40%
20%
0%
-20%
-40%
-60%
-80%
< 25
25-30
30-35
35-40
40-45
45-50
50-55
55-60
60-65
65+
40%
Relative claim severity
30%
Distance: Claimant Home and Employer
20%
10%
0%
-10%
-20%
-30%
-40%
Recording of this session via any media type is strictly prohibited.
<1
1 to 3
3 to 5
5 to 7
7 to 10
10 to 15
15 to 20
20 to 25
25 to 30
30+
Page 6
Used to Predict Outcomes
By combining internal data with external data from a number of sources, enhanced segmentation can be achieved.
External data can also provide an early indication of existing co-morbidities.
Claimant Data
Medical Data
External Public Databases
• Claimant Specific
Information
• Zip Code Demographic
• Medical History
• Household Demographic
• Treatment History
• Diagnosis Information
• Claimant
• Treating Physician
• Years of Employment
• Medical
• Diagnosis Information
• Type of Work
• Legal
• Treatment Patterns
• Job Level
• Prescription Usage
• Average Weekly Wage
• Co-morbidities
Claims Data
Employer Data
Policy History Data
• Losses
• Experience Data
• Financial Stress
• Timing/Patterns
• Policy Data
• Years in Business
• Settlement Data
• Public Record Filings
• Jurisdiction
• Loss Control Data
• Fraud/Lawsuit
Recording of this session via any media type is strictly prohibited.
Page 7
Clients Are Realizing Significant Benefits From Our
Claims Predictive Models
Workers’ compensation models for claim operations are designed to help injured claimants return to work sooner,
with reduce loss costs.
Claim Routing & Assignment
Fraud Detection
 Right claim, right resource
 Reduce lag time of SIU referrals
 Improve routing to auto-adjudication
 Improve mix of claims referred to SIU
 Increase triage consistency through automation
 Deterrence of “soft-fraud”
4-8% reduction in
loss and expense
5-10% improvement in
SIU managed claims
Typical Range of
Savings for Clients
WC
Spend
4% – 8%
\$100M
Medical Management
 Prompt assignment of nurses on those cases that need it
most
 Integrate behavior issues into nurse assignment
 Cost effective use of field case management
20-25% redeployment
of supervisory resources
3-7% improvement in nurse
managed claims
Savings Per \$100M
Of WC Spend
\$4M
\$8M
Top Line Growth
 Demonstrated ability to close claims faster and
 Improved client satisfaction strengthens the
relationship and brand
Recording of this session via any media type is strictly prohibited.
Page 8
RISK MANAGER’S VIEW
Recording of this session via any media type is strictly prohibited.
Page 9
Why Predictive Analytics?
Workers’ Compensation!
Will you be my miracle?
• Always been metrics focused
o
Track losses monthly by Business Unit/Branch/Customer
– Avg. cost per claim, Loss Rate, Frequency
Annual Workers’ Compensation Actuarial Reserve Analysis
o Quarterly Roll-Forwards estimating Ultimates
o Annual estimates of Pure Premiums (Loss Rates)
o
Recording of this session via any media type is strictly prohibited.
Page 10
“Early Intervention Is the Next Best Thing
to Prevention”
The power to see the future!
If we knew from the start which claims were going to become
complex and costly we would:
- Assign the claim to the appropriate level adjuster
- Increase Management review and involvement
- Involve appropriate medical cost control measures
- Retain the best legal defense
- Focus on Return-to-Work
Shout-out to the hard-working Adjuster
•
•
Not a replacement
Tool to help manage claims and reduce workload
Recording of this session via any media type is strictly prohibited.
Page 11
Data Ex Machina
How can predictive analytics be used? …Let me count the ways…
Through Scoring (i.e., High (red zone) to Low (green zone)) and
action items can provide guidance on:
- Claim prioritization
- Expedition of low exposure claims
- Proper assignment of claims to appropriate level
- Cost effective use of field case management
- Loss Reserving
- Settlements
- Future Allowable estimates (Medicare and Rx Risk)
- Subrogation potential
- Litigation management
- Fraud detection (better utilization of Investigative resources)
Recording of this session via any media type is strictly prohibited.
Page 12
A Predictive Model Enables
Predictive models prospectively identify adverse claims to enable proactive management strategies across all areas
of a claim to drive better business results.
Maintain
Value
Proposition
Optimize
Claim
Outcomes
 Deliver high quality service
 Connect to customers and agents
 Ensure quality medical care
 Pay the right amount
 Ensure appropriate return to work for
all injured workers
 Accelerate the claims life cycle
Minimize
Costs to
Handle
 Improve process and operational
efficiency
 Properly match skills with work
Maintain
Discipline
 Improve reserve accuracy and
consistency
 Enhance regulatory and corporate
compliance
Drive to
Excellence
 Leadership vision and commitment
 Organizational readiness to execute
Claim Assignment
SIU Mgmt.
Escalation
Medical Case
Mgmt.
Subrogation
Litigation
Recording of this session via any media type is strictly prohibited.
Page 13
Other Uses of Predictive Analytics
Review of Client/Location Performance
Loss Ratio/Frequency Rate by Industry Average
(WC Code/State)
Tracking locals with higher Loss Ratios
Pricing (Lost Cost prediction)
Underwriting (Risk Selection and Triggers to ask
Recording of this session via any media type is strictly prohibited.
Page 14
INSURER PERSPECTIVE
Recording of this session via any media type is strictly prohibited.
Page 15
Workers’ Compensation
1
Indemnity
Indemnity
Indemnity
53%
41%
Medicals
47%
1989
Medicals
33%
59%
2011p
Medicals
67%
2019 Estimate
1 Top five states only, normalized by state; includes medical only and indemnity claims. Accident Year evaluated at 24 months. As reported in NCCI State
of the Line Report. 2019 Data: Insurance Information Institute
Recording of this session via any media type is strictly prohibited.
Page 16
Recording of this session via any media type is strictly prohibited.
Page 17
Data & Analytics: Predictive Models
Early
Identification
& Intervention
The right resources
on the right claims
at the right time
Recording of this session via any media type is strictly prohibited.
Page 18
Predictive Models Used on All New Claims
Nurse Triage which determines the need for nurse case management
Return-to-work Target Dates model identifies expedient and safe return-towork expectations
Subrogation Triage model helps us ensure that we pursue every opportunity
for recovery
Risk Control Triage model helps determine if it would be beneficial to bring in
risk control expertise to help mitigate future, similar risks
67%
of injured workers
with our RTW focus*
*Return
To Work: National Accounts book of
Recording of this session via any media type is strictly
prohibited.
business results for accident year 2012.
Page 19
Models Used During the Life of the Claim
Early Intervention Chronic Pain model which helps us manage chronic pain
from the beginning of an injury
Recidivism model to intervene in claims where re-injury could threaten
permanent return-to-work
Pharmacy Intervention model targets specific high-risk medications and drug
interactions which can harm return-to-work efforts
Environmental Scans help us identify and alert claim professionals to state
specific variations in the claim handling process
Recording of this session via any media type is strictly prohibited.
Page 20
Evolving
Structured
Unstructured
Other Factors
Strategic Claim File Review
Selection
Recording of this session via any media type is strictly prohibited.
Page 21
Discussion
Recording of this session via any media type is strictly prohibited.
Page 22
Thank You for Participating
Recording of this session via any media type is strictly prohibited.
Page 23
```