Customer Attributes

UCSD MGT 100 Week 02

Kenneth C. Wilbur and Dan Yavorsky

Let’s reflect

• What is Market Segmentation by Qualtrics

• Qualtrics exemplifies a lot of industry literature. Straightforward, hands-on, sometimes assumes you already know the basics
• We’ll focus in today’s talk more on the why of segmentation

Segmentation

• What is it
• Measurable, Accessible, Substantial, Actionable

Heterogeneity

• A fancy way to say that consumers differ, e.g.

• Product needs–usage intensity, frequency, context; loyalty

• Demographics–often overrated as predictors of behavior

• Psychographics–Orientation to Art, Status, Religion, Family, …

• Location

• Experience

• Information

• Attitudes

• Differences often predict purchases, wtp, usage, satisfaction, retention, …

• Different people have different needs and preferences.
• Nobody wants lukewarm tea.
• Why did Coke introduce Coke Zero when it already had Diet Coke? (taste, gender, Pepsi Max, shelfspace; original advertising strategy)
• We have to be careful– sometimes when you try to appeal to too many, you end up appealing to nobody

Market Segmentation

• Segments: distinct customer groups with similar attributes within a segment, different attributes between segments

      - Fundamental since the 1960s
      - Numerous segmentation techniques exist; major recent improvements
      - Customer Response Profiles embody segments
      - B2B segments: customer needs, size, profitability, internal structure

• Decision drivers: product attributes, extensions, bundling, packaging; advertising targeting, content, media; price discrimination, discounts; social media, loyalty programs, …

• Segments should be Measurable in relation to firm objectives, Accessible, Substantial, Actionable

• B2B products all offer one of: increased revenue, cost saving, time savings, risk reduction
• You read about customer response profiles, so no need to dive much deeper on that

Segments in this class, ranked by size

1. Business Econ majors who won’t like Marketing

2. Business Econ majors who will like Marketing

3. Non-technical majors (ISIB, Biz Psych, Comm, etc)

4. Data Science, Computer Science, other technical majors

Segments inform our course content for 2 objectives:

    - Broad survey of analytics, customers and marketing topics

    - Deep dive into demand estimation with pointers to further learning

Measurable

• Why do you think those high-FICO accounts are unprofitable?

Substantial

• Small segments can sustain profitable niche-focused services
• But if you go too small, you can’t sustain operations
• 1 of these is out of business, and the other 3 appear dead

Is segmenting by gender sexist?

• Literally, yes
• When is that bad?

• We have some high-quality evidence that some products targeted toward women do not cost more than identical products targeted toward men
• Gender-based price differences are sometimes conflated with preferences or costs that correlate with gender
• Yet there still may be an equity-based argument for gender-based pricing regulation, given that women tend to earn less and spend more on children

• Gender-Based Pricing in Consumer Packaged Goods: A Pink Tax?

Customer demographics

• In some markets– makeup, diapers, sports, shoes– demographics correlate strongly with behaviors

• In most markets– smartphones, universities, software, cars– demographics correlate weakly with behaviors

• Demographics don’t typically cause purchases, except when they predict real differences in customer needs

• Why do we so often overrate demographics as predictors of behavior?

• Answer: Because they are easily measurable for most consumers; because most people best understand what they can perceive visually; because many people identify themselves in these terms; but not because they describe the most important behavioral differences in marketing, so we have to set our priors appropriately
• Big caveat: There are certainly categories where these variables correlate strongly with other relevant factors (makeup, diapers, sports, trucks)

Segmentation in Action

• Who does it
• Browser users
• Why we keep it quiet

• Nearly every large business segments its markets

-Walmart CMO: Divide customers into low/mid/high income vs low/mid/high price sensitivity (9 segments). Targeted 4 segments: 3 low income and also the high-income/high-price-sensitive segment. This is why you see BMWs and Mercedes in walmart parking lots. If Walmart only targets 4 out of 9 segments, your business will not go after the whole market

Firefox User Types

-https://blog.mozilla.org/ux/2013/08/firefox-user-types-in-north-america/https://blog.mozilla.org/ux/2013/08/firefox-user-types-in-north-america/

Firefox User Types

-https://www.nngroup.com/articles/computer-skill-levels/https://www.nngroup.com/articles/computer-skill-levels/ - in combination w mozilla exercise, you can deliver the point validated by this paper: https://journals.sagepub.com/doi/10.1509/jmr.13.0296 - Why did Mozilla publicize its segmentation scheme? To convince its volunteer developers that their typical users are much, much less technical and interested in different (simpler) features than what those engineers are most interested in developing - releasing segmentation information is atypical, these are closely guarded secrets in most organizations. Let’s look at another rare example

Firms don’t publicize segments

• UO website: “We stock our stores with what we love, calling on our — and our customer’s — interest in contemporary art, music and fashion. …
• “We offer a lifestyle-specific shopping experience for the educated, urban-minded individual in the 18 to 30 year-old range…’’

Firms don’t publicize segments

• Earnings call: “Our customer is from traditional homes and advantage, but this offers them the benefit of rebellion…
• Our customer is exposed to new ideas and philosophies. This can be a real involvement and work, or it could be just talk.
• Irreverence and concern can live together. Often products sell well that represent the concerns they have but also can speak to their irreverence.
• Our customer leads a pretty cloistered existence although they deem themselves worldly…they believe that they’re right and they believe that everything that’s happening to them is what’s happening everywhere.
• Our customer is highly involved in mating and dating behavior…one of the primary drives for their spending behavior…they work hard to postpone adulthood… ’’

Firms don’t publicize segments

• A. website: “a lifestyle brand that catered to creative, educated and affluent 30-45 year-old women…
• “Our customer is a creative-minded woman, who wants to look like herself, not the masses. She has a sense of adventure about what she wears, and although fashion is important to her, she is too busy enjoying life to be governed by the latest trends.’’

Firms don’t publicize segments

• Earnings call: “We don’t think of her in terms of age or affluence or even location. We try to think of her in her life stage and her sensibilities.
• “She’s recently wed. She’s settling down. She’s very interested less in the mating rituals and actually has been trying and building and creating an environment she wants to live in for herself and family.
• “She loves art and culture… And clothing and her living environment to her are canvases in which she’s able to express and control her life, whereas workplace and those things around her, she may not control.
• “We believe in many ways that’s what’s touched her and connect her to Anthropologie and why she is more loyal to us than to most retailers.’’

The Nuts and the Bolts

• Customer Data Platform (CDP)
• Data Marketplaces

Customer Data Platform (CDP) - 4 jobs

1. Data collection

     Intake data from numerous disparate sources:
     In-house, direct partners, data brokers, public data

2. Data unification or harmonization

     Authenticate and de-duplicate rows and columns

3. Data comprehension

     Generate inferences, test hypotheses, make predictions, estimate models
     Covers descriptive, diagnostic & predictive analytics

4. Data activation

     Prescriptive analytics: Use data to inform and automate marketing actions

• Previously called “CRM systems,” “data warehouses,” “data lakes”
  

Data Marketplaces

• Relatively new phenomenon:
  Automated platforms for transacting & transmitting data

• E.g., Snowflake Marketplace

• Upsides

    More data types and sources
    Easy subscriptions, automatic updates
    Competitive marketplace may lower prices
    Less data wrangling 

• Some caveats

    Low barriers to entry
    Most datasets are not audited or externally validated (for now)
    Many buyers don't really know how to check data quality
    These conditions can lead to a lemons/peaches market
    Buyer beware: Always try before you buy

Recent evidence

• Comparing measurability: Demographics vs. Behavior
• Comparing performance of 8 demand models

Research question

• Suppose we

  1. Train demand model \(M\) to predict mayonnaise sales …

  2. … using information set \(X\) …

  3. … & choose targeted discounts for each consumer to maximize firm profits

      - Essentially 3rd-degree price discrimination

• Separately, using different data, we nonparametrically estimate how each individual responds to price discounts

        - This gives us ground-truth to assess each household's response to price discount
        - But, the nonparametric estimate can't give counterfactual predictions; we need M for that

• How do targeted coupon profits depend on \(M\) and \(X\)?

        - We use model $M$ and $X$ to predict profits of offering a price discount to each individual household
        - We use ground-truth to calculate household response, then calculate profits across all households
        - We'll also compare to no-discount and always-discount strategies

• Optimal Price Targeting (MkSc 2022)

A little bit of theory

• For any price discount < contribution margin, giving a coupon to…

  • … our own brand-loyal customer directly reduces profit

  • … a marginal customer may increase profit

  • … another brand’s loyal customer does not change profit

• So the demand model’s challenge is to correctly identify the marginal customers without accidentally identifying our own brand-loyal customer

      - This research disregards the `post-promotion dip' for simplicity

Information sets \(X\)

1. Base Demographics:
   Income, HHsize, Retired, Unemployed, SingleMom

2. Extra Demographics: Age, HighSchool, College, WhiteCollar, #Kids, Married, #Dogs, #Cats, Renter, #TVs

3. Purchase History: BrandPurchaseShares, BrandPurchaseCounts, DiscountShare, FeatureShare, DisplayShare, #BrandsPurchased, TotalSpending

• The base demographic covariate set was motivated by prior literature
• The purchase history data indicate how often each individual responded to past price discounts, and how brand loyal they were in the past to each brand, so those should be powerful predictors in this context

Demand Models \(M\)

1. Bayesian Logit models (3)

        - Based on utility maximization in which consumers compare utility and price of each available product
        - Includes Hierarchical and Pooled versions

2. Multinomial Logistic Regressions (2)

        - Estimated via Lasso and Elastic Net

3. Neural Network (2)

        - Including single-layer and deep NN

4. KNN (Nearest-Neighbor Algorithm) (1)

5. Random Forests (2)

        - Including standard RF for bagging and XGBoost for boosting

• All models trained using 5-fold cross validation. See paper and estimation code for further details

• Notice first that giving everyone a coupon gives higher profits than giving nobody a coupon, suggesting maybe the firm was pricing too high previously. We can’t typically assume optimal pricing in the marketplace
• Bayesian Hierarchical Logit performs well even without purchase history data, because it constructs a likelihood for every consumer based on observed purchases
• Among base demographics only, none of the ML models outperform the blanket coupon
• With expanded demographics, ML models come closer to blanket-coupon profits
• ML models perform much better when they have better purchase history data to learn from
• Bayesian hierarchical logit is not defined for purchase history column because the model’s likelihood essentially builds in-sample purchase history into the posterior through the unobserved heterogeneity distribution; that’s why it does well even when it only has base demos
• pooled logit does not allow for unobserved heterogeneity

• Profit dfce’s across models are not because of major gaps in average prediction performance; they come from picking the right customers
• Ranking models by average prediction error differs greatly from ranking them by expected profits!
• Can skip over this if short on time

• Past purchases on discount are, by far, the strongest predictor of models’ discount policies
• Total spending is 2nd strongest

Takeaways

• To predict behavior, use past behavior

• Economic theory can help demand models to perform well with limited behavioral data

• ML performance depends critically on data quality. Predictions do not always outperform economic models

• Statistical performance \(\ne\) economic performance

• We’ll start estimating logit models soon

• This paper is useful overall but it has a couple of weaknesses
• 1. They could have done more to show how price targeting relates to brand loyalty or nonloyalty
• 2. I wish they had shown correlations between the 3 covariate sets

• Meet your study group. Create a group chat. Arrange a regular weekly time to discuss homework, preferably in person. Enter your schedule into Canvas Intermission 2.

How we segment

• Data
• Methods

• Suppose we segment the smartphone market according to each customer’s desired brand.
• Is this a good approach?

How to pick attributes?

• We want to segment based on attributes that drive sales, profit, retention. But how?

1. Theory
2. Market research
3. Customer database
4. Consult customer experts (salespeople)
5. Find out what other firms are doing
6. Let sales data pick for us (het. logit)

• NO. We’re using our desired outcome as an input. It’s circular. We won’t get anywhere.
• Smartphone brand choices can change, so segments would be transitory.
• Customer differences interact with products to drive choices. We don’t use customer choices to define the customers who made them.

How GBK segments

• How GBK approaches a segmentation project

Cluster analysis

• “Unsupervised learning” - techniques to segment/partition data

• Whenever you hear ‘unsupervised learning,’ think segmentation
• UL can be used for other things too, but letting a computer decide which objects are similar to which other objects is one of the canonical motivations for UL. Maybe the most popular use
• BTW “Supervised learning” <–> regression

K-Means

• Simple, elegant approach to define \(k=1,\ldots,K\) segments
• Main idea: Choose \(K\) centroids \(\{C_1, ..., C_K\}\) to minimize total within-segment variation:

\[ \min \sum_{k=1}^K W(C_k)\]

• where \(W(C_k)\) measures variation among customers assigned to segment \(k\)

K-Means

• Most common \(W(C_k)\) function is Euclidean distance
• Given a set of \(i \in I_k\) customers in segment \(k\), each with \(p=1,...,P\) measured attributes \(x_{ip}\),

\[ W(C_k)=\sqrt {\sum_{i \in I_k} \sum_{p} (x_{ip}-\bar{x}_{kp})^2}\]

where \(\bar{x}_{kp}\) is the average of \(\bar{x}_{ip}\) for all \(i \in I_k\), and the centroid is \(C_k=(\bar{x}_{k1}, ..., \bar{x}_{kP})\)

• the exponents are somewhat arbitrary, could just as easily be power 1, 3, or whatever. Depends on how much we want to penalize outliers
• we’ve taken the customer assignments for granted so far; that’s like assuming we know the answer before we write the program

K-Means Algorithm

• How do we assign customers to segments?
• There are nearly \(K^n\) ways to partition \(n\) obs into \(K\) clusters
• Happily, a simple algorithm finds a local optimum:

1. Randomly choose \(K\) centroids
2. Assign every customer to nearest centroid
3. Compute new centroids based on customer assignments
4. Iterate 2-3 until convergence
5. (Optional) Repeat 1-4 for many random centroids

    - $W$ is not globally concave, so we can't guarantee a global minimum
    - Thus, we pick many starting points, and see which offers the lowest W
    - Note: Some algos promise to find global minimum, but this is usually impossible without stringent criteria. Can be a 'tell'
    

Script prelude: Data structures

• Matrix or Table:
  2-dimensional data storage structure. Often inefficient
• R data.frame: Compact, flexible way to store data
• Tibble: Tidy’s version of data.frame. Similar
• List: Set of disparate structures

Class script

• Standardizing variables
• Running canned kmeans
• Selecting from a list
• Iris example
• Coding & graphing kmeans

Wrapping up

Homework

• Let’s take a look

Recap

• Customer attributes are similar within segments and differ between segments
• Good segments are Measurable, Accessible, Substantial, Actionable
• Customer behavior usually predicts behavior better than demographics

Going further

• How the Pentagon Learned to Use Targeted Ads to Find Its Targets-and Vladimir Putin
• Firefox User Types in North America
• K-Means Clustering: An Explorable Explainer
• Optimal Price Targeting by SSA 2022
• Learning the k in k-means (Hammerly and Elkan 2003)