MMM

The managerial process of creating value by identifying

& profitably satisfying customer needs

1. The process of generating information about

customers & markets to identify

marketing opportunitys & problems & to aid decision

making by evaluating marketing actions

2. The set of techniques priciples for systematically collecting, recording data and interpreting to turn into valuable inform. that can aid decision making.

1. situation analysis
→ customers, Markets


2. Market )ing strategy
→ segmentation, Targeting, Positioning

3. Marketing mix
→ Pricing, Product, services, Communication, Distribution

4. control & Diagnosis
→ Financial Outcomes, Process Metrics, Customer value, Brand value

1. Problem formulation
2. Research Question
3. Data collection
4. Analysis and Interpretation
5. Reporting

What is the managerial problem
What decisions need to be made
Who is responsible for solving
Who is paying

What are the research questions
Can they be addressed through data analysis
What type of study
How much is already known

What kind of data is needed
Can existing data be used
Sources
How to collect
Is it feasible
How different sources merged and cleaned
How store- available for later

Full collection - whole population
+ maximum compilation of information
+ fewer statistical uncertainties
- expensive to implement
- not always feasible

Vs

Partial collection
+ cost efficient to implement
- satisfying statistical requirements on the sample selection
- statistical uncertainties in the statements (sampling error)

Primary data
- does not exist set
- ad hoc survey
Adv; optimal adaptation of the data to project goals, greater control over the data quality

Secondary data
- data already exists or is collected by third parties at specific intervals and standardized form
- population statistics, retail panels
Adv; lower cost, lower time requirement

What method will be used
What kind of visualization
What are the links between the empirical inferences and actions

Who will read and use the report
What is the technical level of audience
Which initial questions can now be answered
What kind of managerial recommendation can be given

A confidence Intervall says how reliable a measurement is if it would be repeated

1. Define the population
2. Determine the sampling frame (internet site subscriptions, telephone book)
3. Select sampling procedure
4. Determine sampling size
5. Execute

Cluster sampling is a probability sampling design where observations are grouped onto clusters and a probability sample of clusters is selected. Either all the observations in the selected sample are used or samples from the selected clusters are obtained in a second stage.
Clusters should have a representative composition and all clusters should be similar

Reasons to use
1. It could be infeasible or inefficient to sample the whole population because of cost
2.there are already natural groups with an adequate composition

Is a probability sampling where the entire population is classified into mutually exclusive groups and a random sample is drawn from each group

Rationale
If the strata are likely to provide different responses the stratification ensures that each subgroup is appropriately represented in the sample

- in a proportional stratified random sampling the sample size for each subgroup is proportional to the size of the subgroup
- stratified random sampling produces more precise estimators on the total population for a given sample
- stratified random sampling provides information about the subgroups

Internal

External; standardized data collection, published data, internet

-Enough information to solve problem / gain insight on a market situation
-source for new ideas which can be used for primary data
- first step for further studies
- benchmark for checking validity of primary data
- helps define the population to sample

Benefits
- low cost
-low waiting time
-some information can only be obtained from 2. Data

Limitations
- incomplete
-Units of measure and level of detail may not correspond to the requirements
- no control over process of collection
- data could be too old
- requires extra assumptions

- diary method and apps
- individual behavior allows us to understand switching, penetration, heavy purchases
- issues with selection bias
- testing effect

Category penetration
Number of customers who have purchased from the category at least once in a given time period
/ total target customers

Brand penetration
Number of customers who have purchases the brand at least once in a given timer period
/ total target customer

Penetration share
Brand penetration / category penetration

Share of requirements
Number of brand purchases / total category purchase by these brand buyers

Heavy usage index
Average number of category purchase by brand buyers
/ average number of category purchases by category buyers

average frequency
Number of impressions per individual reached

Share of voice
Advertising spend for a brand over the total spend in category

Impressions
Reach (number) * average frequency

Gross rating points GRP
Reach (%) * average frequency
Or
Impressions / defined population

Unstructured one to one interview in which the

interviewer attempts to elicit openresponses from the respondents, who are allowed to talk freely & express their true feelings

uses
→ Generate new product ideas & spot gaps in the market
→ Evaluate new products
→ Diagnose reasons for underperformance
→ learn usage of products

Implementation
→ Interviewer provides raw material
→ Interviewer guids discussion
→ Taped
→ 30 minuts
→ compensation

A market research technique where a small
number of carefully selected participants ( 6-8) discuss' a topic of interest, guided by trained moderator
uses:
- Genera te new product ideas
- predict new product success
- Determine reasons for decreasing sales
- Spot product gaps
- Determine product usage
- Evaluate competing products, -packaging, - pricing points
implementations
- comfortable room, Two way mirror, relaxed atmosphere, ice breaking activities
- participants compensation
-90 minutes, Taped
- Repeated (4 groups)
Max 12 part., min. 4
mix of personalities & demographics -

word association

- technique where consumers react to a stimulus (typically a brand, product or product category) by responding with the first word that comes to mind

- Results are then coded in terms of word frequency, response time, proportion of respondents who do not respond

sentenced completion

story telling
- Participants tell a story about product usage through words & pictures

customers are asked about the ultimate motivation

to purchas a product / with Why?
→ Means end theory

Ordering of the questions and groups
- from general, simple to specific
- main section: core questions on the object of research
End; sensitive questions and social demographics
- if possible, randomly rotate the order of questions/. Items and subjects of interest between the respondents
-order the questions so that the responses to subsequent questions cannot be influenced

Use of open ended questions
- used for exploratory purposes and for variety
-problems:
- variation in the willigstes to give answers among the respondents
- variation in the ability of the respondents to articulate themselves
- summarizing responses can lead to inconsistency
- ambiguity in the interpretation of the questions

Ambitious words
Leading or loaded questions that clearly suggest the answer on how the respondent should answer
Implicit assumptions about what would happen
Difficult estimates ; previous month
double barreled questions

Identification of:
Researcher
Sponsoring institution
Purpose of study
Benefits for participant
Level and type of participant involvement

Notation of risk to the participant
Guarantee of confidentiality to the participant
Assurance that the participant can withdraw any time
Provision of names of persons to contact if questions arise

Contrast effect
(Order)

Scale provides cues

1. Question
2. Null hypothesis
3. Alternative hypothesis
4. Set up data process
5. Significance level
6. Data collection
7. Comparison of solutions
8. P value

Reason ; Imprecise questioning, influence of different interviewers, change in Motivation/attention/concentration
Test retest reliability- verification of temporal stability of measurement
-problems: memory effects may lead to overestimation, opinion changes
-trade off; the larger the time Intervall between 2 tests, the more likely are opinion changes, time consuming+costly
Split half reliability-respondents make one test, questions are then divided into 2 halves and the correlation between the 2 halves is determined
-problem; different i sections lead to different reliability coefficients --> internal consistency
- average reliability coefficient of all possible bisections (cronnbachs alpha), can only be applied to a reflective measurement of constructs

- does the scale measure what it is intended to measure? reliability is necessary,
Reason; lacking validity is often the influence of not considered variables

Face validity -to which extent is the measurement representing the underlying construct semantically and with regard to contents?

Predictive validity-to which extent can the measurement predict a future phenomen?

Converge validity-extent, to which different measures of the same construct are correlated

Discriminant validity-extent to which a measure does not correlate with other constructs from which it is supposed to differ

Asks the respondents to agree of disagree with regard to a variety of statements

Characteristics
- the stated values can be summed up to a total value
The prerequisite for summing up is that different dimensions of the object are measured

Indicated the probability that the observed result would occur even though the nulll hypothesis is true

Size of the effect (-)
The larger the measured effect, the lower the probability of error

Dispersion of the measurement values (+)
The greater the dispersion of the measurement values, the greater the probability of error


Sample size (-)
The larger the sample, the lower the probability of error

Beta error indicates the probability of the alternative hypothesis being rejected even tho it its true

Power 1-beta
Indicates the probability that a significance test will decide in favor of a specific alternative hypothesis

The lower the alpha error the greater the beta error and therefore the lower the power of the test

Correlation
Relationship between 2 variables

Causation
One variable producing an effect in another variable

1. Concomitant variation
2. Temporal antecedence
3. No third factor driving both

Ideal: If he could observe everything

Reality: But we don't observe everything
→ If only either sales with Prom. or without is observed we cannot calculate the Effect

Hawthorne effect

-The presence of experimenters

changes the behavior or attitudes of participants & changes the outcome

Mere measurement effect.
- Measuring / surveying participants changes the outcome

If the Hawthorne and were measurement effects affect both treatment & control groups in the same way I we can use solomon four-group design

the lowest sum is the best line

c=linear ratio between 2 variables

R= (linear) ratio between 1 dependent

& multiple influencing variables

Event to be explained or predicted is binary 0 or 1
- response to an offer in direct marketing
- inclusion of a product in the assortment of a retailer
- first use of anew product by buyer

Option1
Logistic regression to model the probability of observing a positive outcome

Option 2
Linear model. Easy to interpret but can lead to unwanted predictions

Option 3
A regression tree

Odds ratio
Odds of satisfaction 4 / odds of satisfaction 3

Relative risk
1+odd

Profit= price*margin*(probability of response at that price and voucher level)-voucher level

Bagging
Select a few variables, compute the tree, repeat , then average all the trees

Random forest
Fancy version of bagging

Boosting
Weight some observations more, compute the tree, reweigh and repeat, then average all the trees

Extreme gradient boosting
The tool of choice of high performing forecasters

1. Benefit range of each attribute
2. Sum of benefit range
3. Benefit range of each attribute / sum = %

1. You have a table, with different attributes and 2 attribute levels .
Give each level a 1 or 0
2. Make a second table with only 1 and 0
3. Look at estimate
4. 1= attribute is given = level of estimate
0= no given = o estimate level
- put it into a graph for each attribute
3. Scaling 0-1
Estimate of each attitude/ highest estimate
-put it into a graph

CONsidered JOINtly
Key concepts
- a survey that forces respondents to make trade-off decisions that reveal their true preferences
-product and services are a bundle of attributes and are viewed holistically
-the result is a mathematical model that describes the rating system and the utility function of a respondent
-it avoids the inflation of expectations and allows the discovery of the true structure of a buyer
- it can be used for
1. Stimulate market shares
2. Predict conversion rates
3. Anticipate cannibalization rate
4. Predict profitability of product changes

Decisions
How many
Which attributes
How many levels
Which min and max level
Which evens in between
How to label attributes and levels

Tips
Only select truly relevant attributes
Max 8 attributes average 5-6
Attributes should not influence each other, if possible
Ideally the same number of attribute levels with a max of 4
Realistic range of attribute levels
Label attributes and levels clearly and uniquely

Nominal scales
Tanimoto coefficient (jaccard)
M coefficient (single matching)

Metric scales
CIty block distance
Euclidean distance

Purchase
Similarity
1. Common purchase categories / number of ALL categories they purchase in (even if only one purchases it counts)

Dissimilarity
1-similarity

Behave
Similarity
1. Categories in which the behave the same way
(Blank spaces also count)
/
Total of all different categories

Dissimilarity
1-similarity

Single linkage
Shortest distance

Complete linkage
Longest distance

Average linkage
Average distance

Ward methods
Minimizing the variation within the cluster

Centroid method
Centroid distance

Single linkage
- tends to form chains with few large clusters and many small clusters
-sensitive to outliers

Complete linkage
- tends to form many small compact clusters

Average linkage
X

Ward technique
- forms roughly equal sized, stable clusters

--> all hierarchical techniques are susceptible to outliers, incorrect specifications on the variables or other error influences

A representation on how the market sees a set of product in a memory-like or cognitive sense

Aim to answer 2 specific questions
1. Along what relevant dimensions is competition taking place
2. How are competition positioned along those dimensions?

1. Segmentation
1.1 identify relevant base variables
1.2 develop the profiles for the identifiable segments based of descriptor variables

2. Targeting
2.1 evaluate each segment according to its attractiveness
2.2 select the target segment

3. Positioning
3.1 determine the ideal product positioning for each target segment
3.2 develop and communicate the appropriate concept

Segmentations variables
- key variables that drive choice
- explain why the segments are different; psychological/behavioral variables

B2C/B2B-- needs, utilities, preferences, loyalty rate, responsiveness to marketing measures, consumption rate

Descriptor variables
- help to identify and reach segments

B2C; age, income, education level, profession
B2B; industry, company size, location

1. Variables; identify key relevant variables that explain customers choice. Some of these variables are hard to observe but can be measured through survey
2. Segment formation; generate groups such that the variation in preferences and behaviors across groups is larger than the variation within a group. Can be done visually or by hand, often requires statistical techniques like clustering
3. Profiling; find observable variables that identify and discriminate distinct segments . These are typical demographic variables that correlate with the main segmentation variables
4. Strategic targeting; evaluate how attractive each of these segments is and pick those woth the highest profit/ value. You serve some segments, ignore others
Or
Tactical targeting; optimize marketing mix for each targeted segment

Divide a sample of observational units into groups (clusters) such that members of the same cluster are similar and members of different clusters are dissimilarity, according to a given similarity measure the quality of clustering demands on the choice variables and the similarity measure

1. Selection of segmenting variables and determination of the similarities
2. Selection of the clustering algorithm
- decision among various hierarchical and partitioned algorithms
-combinations of algorithms
3. Definition of the number of clusters
-selection of a decision criterion
-interpretation of solutions

Partitioning techniques
Start from a given grouping of objects
Exchange the memebership of each object between the groups by using an exchange algorithm as long as a certain objective function reaches its optimum.
Advantage: during the merging process objects can be exchanged between the groups

Hierarchical techniques
Agglomerative/divisive algorithms
Agglomerative (prevail in practical applications),
starts from the finest partition
Combining different clusters
Divisive;
start from grossest partition
Dividing clusters

One stimulus consist of all attributes and the specific attribute levels

Adv
Ranking as well as rating of profiles is possible
Fewer judgments required compared to trade of matrices
Less effort needed to collect the data
More realistic than the trade off method

Disadvantages
Risk of position effects, since the order of attributes within one stimulus may influence their relative importance
The comparison is more challenging for the respondents

The stimulus consists of 2 attributes and all of its attribute levels

Adv
No position effect
Less challenging for the respondents
Guaranteed that respondents consider all attributes consciously

Disadv
More effort needed to collect data
Task can tire out respondents of too many pair comparisons must be made
Less realistic since not all attributes are presented simultaneously

95-> 1.96
90-> 1.645