4. Market share forecasts
• Market simulators
– What-if scenarios to evaluate marketing strategies
– Select a set of products to represent the market
• Often start with the current market as base case
– Each product is represented by its levels on each
feature
– Use each respondent’s utility function to calculate
his/her utility for each product in the choice scenario
– use a decision rule to predict choice for each consumer
– Aggregate the predicted choice (probability) across
respondents to calculate predicted market shares
Us
Us+them
5. Computation steps
step 1: conjoint analysis
Partworths for attribute levels
step 2: multi-attribute utility model
Utilities for competing products
step 3: choice model
Probabilities of choice
individual level
step 4: straight addition
Market share forecasts
aggregate level
Us
Us+them
6. Building blocks: Profile space
•
Profile space
Brand
Apple
Blackberry
Touch screen
Samsung
•
User
interface
Price
Keyboard
Any price
$99 to $399
Suppose we are Apple, and our product is:
Apple
Touch screen
$249
7. Building blocks: Competitors
•
Who are the important competitors?
–
–
–
•
Customer view: look for substitutes for your product
Perceptual maps helpful
Better to include too many rather than too few: conjoint will deal with lack of actual competition, but it
cannot magically account for an excluded competitor
Simplistic example: 2 competitors
Blackberry
Touch screen
Samsung
Keyboard
$199
$149
8. Building blocks: Market
• Market = your product + competitors’ products
you
Apple
Touch screen
$249
competitor 1
Blackberry
Touch screen
$199
competitor 2
Samsung
Keyboard
$149
9. Building blocks: Customers
•
Customer = partworths (each customer is a row of numbers)
Customer
Brand:
Apple
Brand:
Blackberry
Brand:
Samsung
User
interface:
keyboard
User
interface:
Touch
screen
Price:
Utility
$99 vs.
$399
Alex
20
10
0
0
10
30
Bonnie
10
10
0
0
10
30
Colin
0
10
10
0
20
15
Danielle
0
0
0
20
0
15
Ella
0
20
0
0
0
15
Utility function = sum of product’s partworths
10. Exercise: Would Alex buy your product?
•
Alex: partworths (each customer is a row of numbers)
Customer
Brand:
Blackberry
Brand:
Samsung
User
interface:
keyboard
User
interface:
Touch
screen
Price:
Utility
$99 vs.
$399
Alex
•
Brand:
Apple
20
10
0
0
10
30
Market: choice-sets
Brand
User interface
Price
You
Apple
Touch screen
$249
Competitor 1
Blackberry
Touch screen
$199
Competitor 2
Samsung
Keyboard
$149
Utility
11. Help: Would Alex buy your product?
Customer
Brand:
Blackberry
Brand:
Samsung
User
interface:
keyboard
User
interface:
Touch
screen
Price:
Utility
$99 vs.
$399
Alex
•
Brand:
Apple
20
10
0
0
10
30
Reminder on how to interpret the price partworths
–
–
–
•
For other price points: use interpolation
–
–
•
Partworth for $99 is 30 utils
Partworth for $399 is 0 utils
Partworth Gap = 30 uitls
Partworth for $L is (MaxPrice – L) * (partworth gap)/(MaxPrice – MinPrice)
In this case this interpolation formula becomes: Partworth for $L = (399 – L) * 0.1
Now calculate utility for each product
12. Solution - Utility Calculation for Alex
Brand
Price
You
Apple
Touch screen
$249
Competitor 1
Blackberry
Touch screen
$199
Competitor 2
•
User interface
Samsung
Keyboard
$149
Alex is most likely to buy (
)
– Key idea: utility maximization
(We can predict what each customer will buy)
Utility
13. Exercise #2 – Choice Prediction
Highlight the product
Table 1
Custom
er
Brand
:
Apple
Brand:
Blackber
ry
Brand:
Samsu
ng
User
interfac
e:
keyboar
d
User
interface:
Touch
screen
Price:
Utility $99
vs. $399
Utility
of
your
product
Utility
of
Comp 1
Utility
of
Comp 2
Alex
20
10
0
0
10
30
45
40
25
Bonnie
10
10
0
0
10
30
35
40
25
Colin
0
10
10
0
20
15
27.5
40
22.5
Danielle
0
0
0
20
0
15
7.5
10
32.5
Ella
0
20
0
0
0
15
7.5
30
12.5
you
•
Question: Highlight the product chosen by each customer
in Table 1. Assume that customers choose the product
which gives the maximum utility with the probability of 1
(deterministic choice rule.)
Comp 1
Comp 2
Apple
Black
berry
Sam
sung
Touch
screen
Touch
screen
Keyboa
rd
$249
$199
$149
14. Exercise #3 – Market Share Forecast
Question: Add the number of customers purchasing each product and
compute market shares in Table 2.
Custom
er
Brand
:
Apple
Brand:
Blackber
ry
Brand:
Samsu
ng
User
interfac
e:
keyboar
d
User
interface:
Touch
screen
Price:
Utility $99
vs. $399
Utility
of
your
product
Utility
of
Comp 1
Utility
of
Comp 2
Alex
20
10
0
0
10
30
45
40
25
Bonnie
10
10
0
0
10
30
35
40
25
Colin
0
10
10
0
20
15
27.5
40
22.5
Danielle
0
0
0
20
0
15
7.5
10
32.5
Ella
0
20
0
0
0
15
7.5
30
12.5
Table 2
•
Forecast:
you
Comp 1
Comp 2
Product
# persons
buying
%
share
Apple
Black
berry
Sam
sung
Your product
( )
( )%
Competitor 1
( )
( )%
Touch
screen
Touch
screen
Keyboa
rd
Competitor 2
( )
( )%
$249
$199
$149
15. What was the choice model used here?
Utility
of
your
product
Utility
of
Comp 1
Utility
of
Comp 2
•
How sure are you Colin will buy
Comp 2?
Alex
45
40
25
•
Bonnie
35
40
25
How sure are you Alex will buy
your product?
Colin
27.5
40
22.5
Danielle
7.5
10
32.5
•
Ella
7.5
30
12.5
you
Comp 1
Comp 2
Apple
Black
berry
Sam
sung
Alex gets 0.1 utils per dollar
saved (5 utils / $50). What if
Blackberry (comp 1) discounts
by $50? What will Alex buy?
Touch
screen
Touch
screen
Key
board
$249
$199
$149
16. Choice rules
•
Maximum utility rule (deterministic): predict that an individual will always buy the
option with the highest estimated utility
– Simple to apply
– Puts too much confidence in our utility measurement, not empirically valid
– Unstable: the entire prediction can tip with a miniscule discount
Improvement idea: assign probability of choice instead of 0/1!
•
Logit model (probabilistic): predict that an individual will most likely buy the option
with the highest fitted utility, but there is some uncertainty.
17. Logit Model Rule
• Robust, industry standard
• Theoretically sound: related to maximizing utility, Nobel price (2000) to Daniel
McFadden for developing this model
• c = confidence parameter ~ how confident are you in your utility estimates?
18. Logit Model Rule Example: Alex
•
Suppose we take c = 0.1
Utility (U)
c*U
Exp(c*U)
Choice probability
You
45
4.5
90.02
90.02/[90.02+54.6+1
2.18]=0.57
Competitor 1
40
4
54.60
54.60/[90.02+54.6+1
2.18]=0.35
Competitor 2
25
2.5
12.18
12.18/[90.02+54.6+1
2.18]=0.08
19. The Role of Confidence Parameter
Utility
of
your
product
40
25
35
40
25
Colin
27.5
40
22.5
Danielle
7.5
10
32.5
Ella
•
45
Bonnie
Low confidence (c=0.01)
Utility
of
Comp 2
Alex
•
Utility
of
Comp 1
7.5
30
12.5
Medium confidence (c=0.1)
•
High confidence (c=1)
Custom
er
You
Comp
1
Comp
2
Custom
er
You
Comp
1
Comp
2
Custom
er
You
Comp
1
Comp
2
Alex
36%
34%
30%
Alex
57%
35%
8%
Alex
99%
1%
0%
Bonnie
34%
36%
31%
Bonnie
33%
55%
12%
Bonnie
1%
99%
0%
Colin
32%
37%
31%
Colin
20%
68%
12%
Colin
0%
100%
0%
Danielle
30%
31%
39%
Danielle
7%
9%
84%
Danielle
0%
0%
100%
Ella
30%
38%
32%
Ella
8%
78%
14%
Ella
0%
100%
0%
Share
33%
35%
32%
Share
25%
49%
26%
Share
20%
60%
20%
20. How can we determine c?
• It is possible to use a choice-task as part of your ratings-based conjoint
• In practice, we often use
– c = 100/ [12 * Max of Rating Scale]
• With 100 point rating scales, this gives
– c = 100/1200 = 0.083 (a reasonable value based on dozens of past studies)
21. Logit model choice rule: Summary
• Works for arbitrary number of products:
• Interpretation: exp(c*Uia) ~ attractiveness of product A to person I, and logit
is just ratio of attractiveness to total attractiveness of market offerings
23. Market shares
• Prediction of market share is the average of
the individual level probabilities of choice
ˆ
SA
1
N
i
exp( c U iA )
exp( c U ij )
j
Us
Us+them
24. Profit Forecast
• We need marginal cost function and size of the
market in addition to market share forecast
• 1. Compute predicted market share s(P,p)
• 2. Compute predicted marginal costs c(P)
• 3. Compute predicted profit
= {# of customers x s(P,p)} x {p – C}
Us
Us+them
25. Exercise #4 – Profit Forecast
•
•
A medical equipment manufacturer is looking into a new testing device. It has identified a number
of key product characteristics among which price and accuracy are deemed the most important.
The company issued a conjoint analysis which you carried out. It turned out only two types of
customers exist in this market. Segment 1 is 60% of the market, segment 2 is 40% of the market.
The following table of partworths at the segment level was obtained.
Attribute
Level
Price
Accuracy
$13,000
Segment 1
0
20
Segment 2
•
$15,000
0
15
$11,000 99.9%
99%
95%
accuracy accuracy accuracy
40
55
25
0
30
15
10
0
The total size of the market is 100 units. The competition consists of only one firm. It offers a midpriced (i.e. price = $13,000) testing device that delivers 99% accuracy. Your costs to manufacture
and develop the various levels of accuracy are as follows
Costs
Variable Fixed
99.9% accuracy
11000 200000
99% accuracy
10000 150000
95% accuracy
9500
50000
•
•
If you decide to launch the me-too option the best you will be able to do is to get half of the market
and you can maximally charge the price of your competitor.
Consider two product launch options: (A) 99.9% accuracy at price of $15,000, (B) 95% accuracy at
price of $11,000. Which product would be more profitable for you to launch in this market? Show
your work by calculating expected profit for each option. (Note: Assume that the utility differences
are large enough to use the deterministic maximum utility rule.)