SV-AMA Case Study: Conjoint Analysis

SV-AMA

Case Study: Conjoint Analysis

Kurt WolfDirector, Technical Marketing

Non-Volatile Memory Group, AMD(408) 749-5977

January 14, 1998

Agenda

• Conjoint Analysis Overview

• Purpose and Logistics

• Interpreting Utility Values

• Relative Purchase Likelihood

• Positioning and Competitive Response

• Resource Allocation

Conjoint Analysis• Typical question and answer market research is predictable

– Higher performance is typically preferred over lower performance– Lower price is typically preferred over higher price

• However, customers typically buy a product that has a combination of specific features/attributes– A $20,000 car, status of BMW, performance of a Ferrari– What is the optimal combination of attribute levels a particular market segment is most

likely to purchase

• Conjoint analysis measures the trade-off process customers make when purchasing products– This capability is computer driven

Product Attributes• Can include:

– Product features (speed, reliability, etc.)

– Marketing/customer service (credit policy, quotation, billing, etc.)

– Image, compatibility, endorsements

– Price (absolute or relative)

• They can be defined on two levels:– What currently exists

– What might exist

Purpose of Conjoint Analysis Project

• Define Next Generation Flash memory product attribute set (Flash memories are electrically reprogrammable and non-volatile)– First generation Flash products in the industry

did not adequately address customer needs– Actively incorporate the customer (by market

segment) into the definition process• Industrial customer base

Logistics of Conjoint Project• 3rd party consultants implemented mechanics of conjoint analysis

– Product attribute levels and demographics at AMD

– Consultants integrated these project specifics into standard software programs

• AMD Field Applications Engineers (FAEs) administered the project with target customers– Collaboratively developed market segment focus and identified appropriate

contacts/candidates at specific customers

– FAEs explained conjoint project to participants with aid of explanatory documents

• Communicate AMD’s purpose of project, value of customer participation, and overview of conjoint process

“The Commercial”

• Existing product families are based on results of conjoint analysis

• AMD is continuing to gain market share– 24% worldwide Nov-YTD based on WSTS– Unit shipments growing 50% faster than market

• Greater than 50% of all new memory sockets are AMD compatible

Attribute Levels and Utility Values

• Flash memory attribute examples:

Attribute Level– Programming Voltage * 5.0 Volt

* 12.0 Volt

– Sector Erase * Chip erase* 4K Byte sector erase* 8K Byte sector erase

• Customers have different utility values for each product attribute level

Customer Participation• Conjoint project administered by FAEs

• Customers went through conjoint process on their own time

• Estimate a realistic time for completion of project, then double it– Many parties involved; all have their own time

schedule

Interpreting Utility ValuesUtility Utility

Attribute Level Value ValueProgram 5.0V 35 +33 Voltage: 12.0V 2

Sector Chip Erase 10 Erase: 4K Byte Sector 22 +12

8K Byte Sector 20 +10

Price 1.2 x EPROM 681.6 x EPROM 35 -331.8 x EPROM 12 -56

Interpreting Utility ValuesUtility Utility

Attribute Level Value ValueProgram 5.0V 35 +33 Voltage: 12.0V 2

Price 1.2 x EPROM 681.6 x EPROM 35 -331.8 x EPROM 12 -56

• As an individual feature 5.0V programming is more valuable than 12.0V programming

• Customers are indifferent when choosing between a 5.0V device priced 1.6 x EPROM and a 12.0V device priced at 1.2 x EPROM

–The system level cost of providing 12.0V power supply isequivalent to 0.4 x EPROM according to customers

Hypothetical Example of Relative Purchase Likelihood

Product A Product BUtility Utility

Attribute Level Value Level ValueProgram Voltage: 5.0V 35 12.0V 2

Sector Erase: Bulk 7 8K Byte Sector 65

Total Utility Value: 42 67Relative Purchase Likelihood: 17 24

A customer’s Relative Purchase Likelihood (RPL) is modeled as the normalized sum of utility values per product. Larger RPLs are “better”.

Product Simulations• Product simulations are the first step in defining next

generation products

• Create a table of products by combining different attribute levels. Rank order these hypothetical products by RPLV

• The intent is to determine the product with the greatest RPL value that can realistically be produced

Relative Purchase Likelihood Values

RPLV Product Definition

5 BC @ 1.6 x EPROM

12 Base Case (BC)

(12.0 programming, bulk erase,

1.2 x EPROM)

12 BC @ 5.0V, 1.6 x EPROM

19 BC @ 5.0V, 8K Byte sectors

27 BC @ 5.0V, 8K Byte sectors,

1.3 x EPROM

Strategic Positioningand Competitive Response

• Customers’ value of competitive products can be modeled– Include competitor specific attributes and levels in the

implementation phase

– Where do you want to position your product?

• If a competitor changes their product feature set, this can be modeled also– What degrees of freedom do your competitors have?

– What degrees of freedom do you have?• What is your plan when your competitor moves?

Price Interpretations• By changing only the price attribute during a product simulation, the relative

effect on RPLV is observed– How elastic is “demand” to price

• Different application segments may absorb different price premiums for a specific attribute level– Example: Segment Characteristic

5.0 programming: I One Flash device/system

II Two Flash devices/system

III Four Flash devices/system

– The utility value for 5.0V vs. 12.0V programming is best thought of as the incremental value of a 5.0V system vs. a 12.0V system

– The price premium per device is the cost of implementing 12.0V programming amortized over the number of devices/system

Price InterpretationsUsing Utility Values

• Customers receive equivalent value between:– 5.0V devices @ 1.6 x EPROM

and– 12.0V device @ 1.2 x EPROM

• Maximum price premiums per system for a 5.0V implementation is 0.4 x EPROM ( 30%)

• Price premium per device by segmentSegment Price Premium

I 30%

II 15%

III 7.5%

Market Segments• Determine if different market segments prefer devices with

mutually exclusive product attribute levels

• 4K Byte and 8K Byte have equivalent utility values (22 and 20 respectively)– Could indicate customers are indifferent to sector size, or that

some segments prefer smaller while others prefer larger sectors

• If differences exist, these segments must be identified and researched further

Attribute Level Definition• Intermediate attribute levels can be interpolated via utility

value algorithms

• Extend attribute levels beyond (“above” and “below”) anticipated boundaries

• After original conjoint project was completed, the implementation costs for the sector erase attribute were higher than estimated

• Re-issued conjoint analysis to a smaller select group with expanded sector size attribute levels

Utility Value Categories

• Categories

Linear Plateau Increasing then decreasing (vice versa)

Attribute Level (These can be misleading)

Uti

lity

Val

ue

Utility Value/Cost Ratios• Distribute limited resources to implement the attributes that provide

the greatest utility value/cost ratio– Cost should be broadly defined to account for resources, incremental

investment, as well as time to market. Use a metric for cost like relative die size

• Each attribute level with the highest value/cost ratio per attribute type should be considered in descending order

• Market and/or application issues may lead to the choice of a particular attribute level that does not have the highest ratio– Look for attribute levels that satisfy the broadest or most attractive market

segments

3rd Party Support

• Sawtooth Software

(360) 681-2300

www.sawtoothsoftware.com

• Analytical Services Group

Market Research Data Processing

Mark Olsen, President

(510) 769-6417

Final Thoughts• Conjoint analysis is a flexible and insightful technique to

have and use in your Marketing Tool Kit

• It is cost effective and efficient to work with 3rd parties to implement the specific design of your conjoint project

• Conjoint can be performed periodically to monitor customer shifts in preferences and/or to model value of different attribute levels