1

Understanding the Customer Value Chain

Building Evidence-Based Models

Michael J BrockmanPricing Seminar 13th June 2008

Improving Bottom Line Performance

ProfitOptimisation

StrategicOptimisation

•Marketing•Relationship Management

•Channel Strategy•Growth Objectives•Reputation Risk•Capital Management

PriceOptimisation

•Each Product/Channel

•New Business/Renewal

TechnicalAnalysis

•Risk Premium Models•Make/Model Analysis

Geographical•Demand Models

The Customer Value ChainCan apply statistical analysis to understand customer behaviour and value at each stage in the customer value chain. Each stage has a cost

Different types of analyses are appropriate at different stages

Up-Sell

>media modelling

>DM response propensity analysis

>lead value models

>campaign response value optimisation

>propensity analysis

>value models

>trigger-based communications

>campaign response value optimisation

>channel utilisation propensity analysis

>contact centre optimisation

>Training

>price optimisation

>Training

>POS discounts

>price optimisation

>communications strategy

>relationship management and service levels

Gaining Customers Retaining Customers

Growing Relationships

Retention Management

Cross-Sell

Lead Conversion

Prospect Identification

Lead Handling

Lead Generation

>“lookalike”analysis

>geographical analysis

>value models

Maximised Customer

Value

The Leads Pipeline

Unaware

Aware

Consider

Enquire

Buy

Mass Media Advertising

Direct Marketing

PRICING gets you across this threshold, along with:quote scriptingcontact centre

response/operator

OBJECTIVE is to get (ultimately profitable) customers across this threshold, cost-effectively

Customer

Lead

Prospect

“I know that half of my advertising is wasted

I just don’t know which half”Lord Leverhulme

An Age Old Challenge

Still looking for answers in the current landscape

An explosion of communication channelsBusinesses seeking robust measures to demonstrate accountabilityFragmentation of strategic advice from numerous specialist agenciesNeed for an evidence based framework, delivered from a neutral perspective

Strong relationship between media strategy and pricing strategy

Invest in media Invest in price discounting?

In some industries media strategy and customer pricing fulfils two very different roles

Generate Leads

Media and Headline

Offers

e.g. save 35%, 15% off for

online

Customer Pricing

Question

PROFITABILITYVOLUME

Convert

Media modelling techniques recently developed now mean we can understand this relationship

Supply side led

Cheapest media deals

What was done last time

What competitors did

Gut Feel Decisioning

First Media Models

Basic Econometric

Analysis

Generalised Linear/Non Linear

Modelling Techniques

Intuitive belief in what worked

Codified this belief into a “model”

High level link between sales success and media strategy using some form of time series analysis

Detailed link found between pricing and media activity and the volume and profitability of new business

What is the Benefit?

Performance efficiencies of 10-30% gained through more efficient decision making

Overall level of investmentGeographic allocationMedia Mix including BTLProduct mix (home/motor)Which brands to supportTiming/phasingMedia costs perAdapting to competitor activities

Primary focus of Media Modelling is exposure across channels

The focus is on the marketing mix rather than channel specific performance

The process is independent of channel specific ‘source code’ analysis

Projects focus on macro analysis across channels

The micro analysis of the deployment of media within channels is sometimes a secondary consideration

National Press

Local Press Journals Outdoor Direct

Mail Online etc

Day in Week

Daypart

TV

Position in Break

Genre

Station

Identifying Cause and Effect Between Spend and Sales

Inputs Outcomes

Exte

rnal

Fac

tors

Inte

rnal

Fac

tors

Market Conditions

Competitor Activity

Brand Strength

Performance Efficiency

&

Brand Growth

EnquiriesProducts /

Offers

Creative

Media Used

Sales

Many Variables Initially Considered (Between 400 and 500)

Predictive Factors

Radio Ads Short Term

Online Search Engine Impressions

Community Press

Radio Ads Long term

Total Media Exposure

TV TARPS (25-54)

TV Exposure

Press Exposure

Magazine Ads

Magazine Exposure

Online Impressions

Radio Exposure

TV Frequency

Stats

Online Exposure

Cinema/Outdoor Exposure

Media

CreativeOffer

Competitor

Market (Related to Industry)

SeasonalityTechnology Index

GDP Growth

CPI Housing Growth

Unemployment Growth

Housing Purchases

Interest Rates

Consumer Confidence

Radio Ads

Product LaunchesPress Ads

TV TARPS

Cinema/Outdoor Spend

Total Media Spend Discounted

Rate

Added Value Offer

Free Service

Trial

Campaign A Campaign

BCampaign C

Rainfall

World Events

Press Most Expensive

TV Reach Stats

Cinema/Outdoor Spend

TV TARPS Overall

The modelling can be carried out at the postcode district level…

Weekly media exposure for each channel is allocated at a postcode level

TVPr

ess

DM

TVPr

ess

DM

TVPr

ess

DM

TVPr

ess

DM

E14 E16

SE10 SE7

This Provides the Necessary Richness of Data

Media exposure in adjacent postcodes in a given week may be quite similarHowever differences between weeks and geographies allow us to gain a true understanding of the impact of media exposure on customer behaviour

By considering multiple years of data across all geographies, we have a large number of ‘experiments’ or data points on which to conduct the modelling

Modelling at the postcode level gives us greater certainty around the impact media has on enquiries/sales

By looking at the data in a more granular way, more understanding can be drawn about the relationship between exposure and response

Identifying the slope and shape of the relationship offers significant benefits

Lead

s G

ener

ated

Exposure

120

115

110

105

100

95

900 40050 100 150 200 250 300 350

Pure Effects of Each Variable Modelled for Influence

The specific effect of each influence is isolated (Media and non media factors)

These ‘pure effects’ can be used to guide each channel’s optimal deploymentLe

ads

Gen

erat

ed

120

115

110

105

100

95

900 40050 100 150 200 250 300 350

No uplift in leads below 50 GRPs Point of diminishing

returns No additional leads after

300 GRPs

Impact of Number of GRPs (Ages 25-54) In Last 7 Days

Number of GRPs (Ages 25-54) In Last 7 Days

Demographics vary by postcode

Populated mainly by younger people

London

0

0.1

0.2

0.3

0.4

0.5

0 to 14 15 to 29 30 to 44 45 to 50 60 to 74 74+

Age Demographic

Prop

ortio

n of

Peo

ple

Honiton

0

0.1

0.2

0.3

0.4

0.5

0 to 14 15 to 29 30 to 44 45 to 50 60 to 74 74+

Age Demographic

Prop

ortio

n of

Peo

ple

Evenly populated

New Milton

0

0.1

0.2

0.3

0.4

0.5

0 to 14 15 to 29 30 to 44 45 to 50 60 to 74 74+

Age Demographic

Prop

ortio

n of

Peo

ple

Populated mainly by older people

Not surprising because demographics vary so significantly by region – example age

Modelling at the postcode level allows for many different types of media

Include above the line and below the line accurately

Because different areas have different intrinsic value

In some areas you are more competitive

Have higher conversion rates

Have higher customer lifetime values

Modelling impact of media requires a specific approach

We employ generalised linear modelling (GLM) techniques, which are best practice within insurance pricing

GLMs permit “error structures” which correctly describe the natural uncertainty in the data

GLM approach solves key challenges within “traditional” methods:

Multicollinearity of factors, appropriateness of fitting method, testing of residuals, heteroscedasticity, interaction searching and testing

GLMs are:

Able to handle large quantities of data and sparse observations

Robust to changes in exposure mix and changes in the predicted response

GLM Modelling

Channel PerformanceEach channel’s effect can be understood in isolation

Note: Illustrations only

RadioCompetitor

TV

Online Press

Number of ads per person last 7 days GRPs in last 2 weeks

Impressions per person last 4 weeksOnline impressions last 7 days

Rel

ativ

e le

ad v

olum

e

Rel

ativ

e le

ad v

olum

e

Rel

ativ

e le

ad v

olum

e

Rel

ativ

e le

ad v

olum

e

We have developed this approach so we are better at predicting future behaviour

Since we now have sufficient data to incorporate all of the factors which influence lead outcomes and the right shape and slope for each factor

0

1

2

3

4

5

6

7

8

9

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140

Week

Lead

s G

ener

ated

Typical Econometric

Prediction Range

GLM Prediction Range

Spend by Month

0

500

1,000

1,500

2,000

2,500

3,000M

ar-0

4

Apr-

04

May

-04

Jun-

04

Jul-0

4

Aug-

04

Sep-

04

Oct

-04

Nov

-04

Dec

-04

Jan-

05

Feb-

05

Mar

-05

Apr-

05

May

-05

Jun-

05

Jul-0

5

Aug-

05

Sep-

05

Oct

-05

Month

Spen

d ($

000'

s)

TV Info Pr - OTP CPr - OTP Mag - OTP Pr - Ins CPr - Ins Mag - Ins Doordrops Online

Sale

s (V

olum

e)Sa

les

Volu

me

Spen

d ($

000s

)

The benefits of analysis beyond straight lines

Consider September 2005

a high level of media spend was not met with a peak in sales

Media Spend and Sales Volume

September 2005 was Particularly Heavy in OTP Press and TV Exposure

Almost £1m was spent in the month beyond the point of zero marginal returns as identified in our modelling

Relativity by TV Impressions Per Person Last 2 Weeks

60708090

100110120130140

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5TV Impressions Per Person Last 2 Weeks

Rel

ativ

e Pr

open

sity

to

Purc

hase

00.10.20.30.40.50.60.70.80.91

Expo

sure

Exposure - Excl Sep05 Exposure - Sep05 Prediction

Allowing for Customer ValueAllowing for Propensity AND ValueWish to direct marketing spend towards the most valuable prospects

…or towards those who can be influenced to increase value.

Lead Response Propensity

High Propensity

Low Propensity

High Retention

Low Retention

High Retention

Low Retention

Retention Rate

Most Valuable

Prospects

Medium Value

Prospects

Least Valuable

Prospects

High Profitability

LowProfitability

High Profitability

LowProfitability

High Profitability

LowProfitability

High Profitability

LowProfitability

Profitability

Segment Characterisation –The Customer DNA

Can characterise segments, and identify segment-specific levers of influence, and trigger events – and so frame targeted activities.