Analytics in practice sunz 2012

Analytics in practice Driving your business and saving you money

Todd Nicholson

February 2012

Three examples of analytics in

practice

Domestic Tourism Segmentation

(Ministry of Tourism)

Serious Injury Model

(ACC)

Claim Duration Model

(ACC)

Example 1: Domestic Tourism

Segmentation Want to understand targetable groups within the NZ

tourism market

Want to profile each segment so we know:

- who they are

- what they want

- how to reach them

Want to ensure each segment can be targeted

Key learning: Always keep your

target audience in mind

Key learning: Always keep your

target audience in mind Ensured all segments were marketable by segmenting on:

- interests

- age

- ideal holiday type

- inhibitors of travel

This gets us away from segmenting on holiday frequency and directs us towards holiday type.

Key learning: Prettiness matters

Key learning: Prettiness matters

Key learning: Prettiness matters

Example 2: ACC serious Injury

Model ACC has just over 5000 seriously injured clients. For these

clients ACC provides:

- Carers to help with day-to-day living

- Weekly compensation and medical expenses

- Capital items

The outstanding claims liability is currently $9.3 billion.

However, increasing consistency is one of the ways growth in expenses has been successfully dampened.

Analytical models can give you

transparency and consistency

Nu

mb

er

of

care

ho

urs

per

week

Functional Independence Measure

Key Learning: Milk your results

Key Learning: Milk your results

When dealing with people there will

always be variation. So lets measure it!

Nu

mb

er

of

care

ho

urs

pe

r w

eek

Functional Independence Measure

National serious injury

service - Evidence based needs assessment

- $820 million reduction in outstanding claims liability (due to the hard work of the serious injury team – this is simply a tool they’ve commissioned to help them).

- Minimal bad publicity or media comment involving clients with a disability

- No change in the usual, small number of legal challenges to decisions

- An increase in the number of clients achieving their self directed outcomes

Example 3: Claim duration

analysis (ACC) • ACC has over 50,000 weekly compensation claims every

year

• They cost almost $1 billion a year

Aim of the analysis: to accurately predict claims duration

based on past experience

Ultimate Goal: to make better decisions about services

required and achieve shorter claim durations

International tools – Medical

Disability Advisor (MDA)

Risk Factor - Age

Key learning: Know the power of

your data!

Key learning: Know the power of

your data!

• ACC’s data is not perfect. For example, losing your job is a key risk factor but isn’t recorded in ACC data.

• But it is still the best dataset available anywhere in the world for modelling injury recovery

– Longevity

– Universal coverage

– No coming or going

Know the power of your data!

There is nothing magic in our analysis. Its just

• Survival analysis

• Logistic regression

• Generalized Linear Modelling (GLM)

• Text mining (yes, even free text data can be a

gold mine)

The real difficulty is dealing with such a large

amount of data

Milk your results (again!)

Two parts of the analysis have been

implemented for frontline day-to-day use

• They allow branches to accurately identify

which claims will require weekly

compensation

• Staff can take a more proactive approach

• Staff can ensure high risk clients receive

the correct services right from the start

Milk your results

• Predict which claims are likely to last for a

long time and WHY

• Draws case managers’ attention to

potential risks

• Allows accurate allocation of resources –

identifying clients who need no services is

as important as those who need a lot.

Milk your results

0

0.2

0.4

0.6

0.8

1

-30 20 70 120

Pro

po

rtio

n o

f c

laim

s s

till o

pe

n

Weekly compensation days

High Risk Claim

Low Risk Claim

So remember

• Always keep your audience at the front of

your mind

• Prettiness is important

• Milk your results

• Models can give you consistency and

transparency

• Don’t underestimate the power of your

data