Consumer behavior
Class 1: Introduction to consumer behavior
What is marketing
• Critics: fluff, manipulative, wasteful
• Prof: not rocket (deterministic) but behavioral science (probabilistic)
• Marketing = offerings that have value. Not only quality of product but also marketing.
Article: Consumer ratings
• Low correlation (consumer ratings, objective ratings) = 0.15 but user ratings are highly
accessible, and people trust them, despite small sample size that is often neglected. We don’t
always trust companies because of suspected persuasion motive but we do trust online
consumer ratings.
o If product A vs B difference <0.4 stars → 50% chance that higher rating product will
be objectively better
o Increases as difference becomes larger but never >70%
• No relationship between resale value and product quality
• Relationship with reputation of brand and price → user ratings influenced by marketing
• Issues w/ user ratings:
o Statistical issues: low sample size, high variability (noise)
o Sampling issues: brag-and-moan bias, ppl w/ extreme opinions more likely to post
o Evaluation issues: evaluation should be done using a scientific approach, objective
comparison
Factors that determine product value
• Price: Study: wine liked more if higher price. ➔ value is created by marketing
• Promotion: Celebrity endorsement
• Product: if smaller packaging, looks like more product, ppl more likely to pay more
• Place: Social market: music released. Success depends on social, interdependent interactions.
independent market
➔ Value = f(Product, Promotion, Price, Place)
Also value OF customers and value TO customers
Class 2: Managerial thinking traps
What is the secret to success in marketing and what not to do?
• Egocentrism undermines effective marketing. Success is about perspective, get perspective
of consumers. Egocentrism overestimates own contribution and underestimates those of
others.
But perspective taking does not help (study). Perspective-taking vs perspective-getting.
• Overconfidence: gap between knowledge and meta-knowledge. If someone says they’re sure,
they are likely incorrect. If you are sure, you will not look for more information.
o Negative feedback and searching for contradictory evidence reduces overconfidence
(study contra evidence vs feedback condition)
▪ Accuracy higher, not less overconfidence. So increases quality of answers. If
you think you’re bad, you may think deeper and be more accurate.
, • Precision gives illusion of knowledge but is impossible to give. We think we can approximate
quantities with greater precision than in actuality, so we set smaller confidence intervals. We
crave precision.
Precise numbers create an illusion of knowledge. Precision persuades.
Precise price offers create illusion of knowledge ➔ higher bid
o Learn how to world works → accept uncertainty
o Persuade → create illusion of certainty
How effective are online ads?
• Advertising metrics
o CPM = cost per thousand impressions = price paid for 1000 exposures
o CTR = click-through rate = proportion of consumers that clicks through website
when ad exposure
o CPC = cost per click = price paid for each click
o CR = Conversion rate = proportion of customers that makes purchase when exposed
to ad
o CAC = customer acquisition cost = price paid to acquire new customer
o ROAS = return on advertising spend = sales dollars from customers who were
exposed to ad per dollar invested in marketing
▪ Non-causal metric. If you advertise at end of funnel, you get a higher
purchase ratio than when top of funnel. Does not mean one way is more
effective than other.
o ACOS = advertising cost of sales = dollars invested in advertising per dollar in sales
from customers who were exposed to ad
Article: offline impact of online ads
o Online ads → ppl also buy significantly more in store. Therefore, online stats of
online purchases may not be an accurate measurement of how much money
companies make from customer x.
o Search ads worth more than display ads. Search ad and online display ads together
sign higher increase in sales. Explanation: search ads more bc you tell google “I want
to go to Bahamas”. Search ads mean that the consumer is already planning to make a
purchase.
o Causal language: not necessarily true based on data
o Lift = percentage revenue or conversion rate increased exposed to ad vs not → not
inferring causal effect, probably correlation. Think deeper.
o Mental model of how to world works. You believe that ads increase sales so you
accept the article without thinking more deeply.
o ➔ Search ads vs display ads: correlation not causation. If you search for something,
you are more likely to buy it anyway.
• Do customers visit ebay less when ebay stops with brand keyword search ads on google? No
o ppl who clicked on search ad by ebay. If ebay would stop paying google for ads,
google paid drops to 0 and google organic goes up. Ebay is wasting money on google
search ads.
o Eg of how high ROAS is not causal metric bc sales would persist even without ads
• Twitter ads: “we have an approach where we get to know more info on customers, tell
whether customer saw promoted tweet, clicked on it…” company adds how much customer
spent on product. BUT reason why ppl see the ads is bc they’re likely to buy the product.
, • Personalized ads: FB against apple bc apple started asking ppl if they can track data for
personalized ads. FB afraid that they’ll have to do it on FB bc they get money from selling
data to companies.
o Optimize link clicks (more likely to click on link): no personalized ads
o Optimize purchases (more likely to spend): personalized ads
o Correlation does not equal causation. FB predicts that x is going to spend 5, y 10.
They spend personalized ads to ppl who are likely to buy. Non-personalized to all
ppl. No random selection. “Look at the large difference”.
Article: Facebook’s misleading campaign against apple’s privacy policy
Apple: privacy policy that requires permission for cookies. FB criticizes it, claiming that small
businesses will lose revenue without personalized ads. But uses misleading arguments.
Metric: ROAS = Return on ad spend → revenues associated with, not caused by, advertising
But methodology is misleading. Two groups: They share personalized ads (require cookies) w/ people
who they know are more likely to purchase and the non-personalized ads for people who click links
without buying. In reality only small difference.
Other argument: 44% of small businesses started or increased personalized ads during pandemic.
- Majority never did
- Doesn’t say about how much businesses spend on digital advertising
- Number of non-personalized ad use similar to personalized ads. So nothing special about
personalized ads.
Article: Does personalized advertising work as well as tech companies
claim?
“grow business w/ twitter ads”. Companies try to attract customers to advertise on their platform.
FB says lots of companies find users through personalized ads. AB test on FB for business. Normally
the sample is random but ad A gets shown to diff people than ad B. OR ad A shown to group 1 vs not
to group 2.
FB’s machine learning algorithm used to refine selection strategy. Eg young ppl more likely to click
on red ad, so red ad will be shown to young ppl.
Machine learning algorithm invalidates AB test design. No random assignment. Group 1
differs from group 2 by more than just ad, different inherent personality characteristics.
Associative, not causal metrics
Google lacks point of comparison. Don’t know revenue if ad hadn’t been shown.
Twitter for business: ppl who get shown ad are more likely to buy cosmetics. So ppl who see
promoted tweets are different from those who don’t. ppl don’t buy cosmetics bc they have been shown
ad but bc ppl who get shown ad are more likely to buy.
Different questions companies can ask. Important to know difference.
- Prediction questions: will customer buy
- Causal inference questions : will this ad make customer buy → difficult to answer bc of the
algorithms
Small companies should know this about ads.
Are vaccines effective? – the importance of an experiment, RCT, and use of
A/B tests
o RCT (randomized control trial) in medicine, how it should be done in marketing also.
o If you don’t have RCT, you have to ask “what other variable is missing and how are
they related to the variables they claim to have a causal relation?”
o Eg Moderna 94% effective vs pfize >90. Difference between 2 is not statistically
significant. So although moderna seems more effective statistics say no
o Simpson’s paradox: rel x and y. 2 groups. If you compute r ignoring the groups →
negative r. Within blue ppl and red ppl: positive r.
