underwriting to

1
underwriting to
Modelling
understand
now
the
future
and the

• Nigel Bradshaw

2
Review of current modelling

• Cost effectiveness of medical limits
• e.g. Swiss Res LIMITS
• Teleunderwriting process costs
• mainly call time
• and medical evidence
• Ad-hoc process modelling
• often associated with business cases
• or effectiveness on some impairment

3
Why model?
Benchmark process costs
Evaluate processing alternatives
Cost-effectiveness of potential outsourcing
Medical evidence alternatives
Distributor costs
Optimise speed of processing versus completion
rates
Optimise disclosures
Preferred underwriting
4
Lets be about this
systematic

• Identify what issue(s) needs modelling
• Isolate the key criteria to predict
• Evaluate the data needed
• Collect what data you can
• Make assumptions about the rest
• Decide upon a model
• Build it
• Test it, use it, improve it, test it, use it,

5
Issues to be modelled

• Improvement of process efficiency and cost
• time and motion
• including medical evidence
• predictive modelling for small changes through to
  radical alternatives
• Reduction in not proceeded with rates
• model benefit versus extra costs involved
• Improvement in disclosure rates
• varies with when and how evidence collected
• Results
• spread of ratings
• performance of individual underwriters
• performance on individual impairments
• effect of new underwriting standards

6
Criteria that could be modelled
M O D E L

• Time to issue
• Cost (per application / life / issued policy)
• Not proceeded with rate
• Percentage of PMARs / PMEs / etc.
• Level of disclosure / non-disclosure
• Work levels in underwriting new bus.
• Skills needed in underwriting new bus.
• Ratings
• Performance versus ratings
• Agent performance
• Agent / customer satisfaction

7
Data

• Route tree through underwriting
• Result at each branch
• Process timings
• Final decision
• History of active case (complaints, claims,
  lapses)
• Underwriter(s) involved
• GP, (para)medical examiner involved
• Details of lives (age, sex, conditions)
• Details of agent

MODEL
8
Assumptions

• You will not have all the data you need
• either you cannot collect it
• or it does not exist
• This is especially the case with new processes or
  approaches
• though if you break them down you might be
  surprised how much is available
• Where data does not exist you make assumptions
• keep them explicit and model them so you can test
  alternative values for them
• Where possible test assumptions through separate
  mini investigations
• Setting assumptions is where the old heads win
  over the upstart new graduate

9
The model

• Modelling underwriting is nothing special

Data
Model
Results
Assumptions

• Well planned and structured
• No more complicated than needed
• Well tested and documented if to be reused
• Pilot if possible

10
The short-term control cycle for a model
Data
Model
Results
Assumptions
Validate results common sense
Revise model and assumptions
Sensitivity check assumptions
11
The long-term control cycle for a model
Data
Model
Results
Assumptions
Experience
Validate results versus experience
Revise model and assumptions
12
A case study

• A model in development
• It will be generic for use by different insurers
  to answer a range of questions
• Initially it will only be available as part of a
  consulting contract
• In the future it may be available separately
• It is being developed from a pilot model used
  with two different clients

13
Model scope

• Improvement of process efficiency and cost
• time and motion
• including medical evidence
• predictive modelling for small changes through to
  radical alternatives
• Reduction in not proceeded with rates
• model benefit versus extra costs involved
• Improvement in disclosure rates
• varies with when and how evidence collected
• Results
• spread of ratings
• performance of individual underwriters
• performance on individual impairments
• effect of new underwriting standards

14
Some issues

• These have been identified through the pilot
  modelling work
• Route tree through underwriting
• Data (and lack of it)
• Complexity
• This is not exhaustive

15
Route tree through underwriting

• This is a process map

Clean case?
Clean case?
Automated 5 yes processing
Manual 1st pass processing
Yes
Yes
No
No

• This is a key base on which the model will be
  built
• I have broken it down to be very granular
• can look at incremental changes
• will have many of the components from which more
  radical models can be built up
• Hard to get right first time

16
Data

• It has proved difficult in the pilots to get hold
  of sufficient hard data
• Even combining the data from 2 companies
• Current systems are not set up to record
  underwriting process data in the level required
• If they are set up at all!
• One-off extracts have proved useful
• In particular looking at the dates and key stages
  from a tranche of business written in one month
  last year

17
Complexity

• My model is complex
• Generic, highly granulated, wide range of
  criteria
• It is a series of modules
• Modules to be altered to reflect available data
• Intermediate results can be tested

• Its still a pig
• If you want to build your own you may prefer to
  build a simple one objective model

18
To finish my thoughts

• Underwriting needs modelling more often
• There is a wide range of issues that would
  benefit from modelling
• A more systematic approach to the underwriting
  process and results will improve the status of
  underwriting
• Building a model can be fun!