Key Concepts
- 1Personalization is powerful but bounded — by consent, by data freshness, by fairness, and by the creepiness threshold where 'helpful' becomes 'surveilling'
- 2Consent scope: data a user gave for order fulfillment isn't automatically fair game for behavioral targeting; PMs must map which data can power which personalization
- 3Staleness is a silent failure: personalizing on a preference the user has since changed (a past diet, an old size) actively harms the experience — freshness and decay matter
- 4The creepiness line is context-dependent: referencing a purchase the user made is fine; referencing browsing they didn't realize was tracked feels invasive
- 5Fairness and bias: personalization that systematically shows worse prices or options to certain groups is a legal and ethical liability, not just a UX issue
- 6Cold-start and sparse data force graceful degradation — a good agent personalizes lightly with thin data rather than over-fitting on one signal
- 7Give users control: visible, editable preferences ('you told me you're vegan — still true?') turn personalization from something done to the user into something done with them
Why It Matters for PMs
Trust is fragile and asymmetric: One creepy or wrong personalization can undo months of goodwill. PMs must treat the creepiness and staleness thresholds as first-class design constraints.
Legal exposure is real: Consent misuse and discriminatory personalization carry regulatory penalties (GDPR, CCPA, fair-pricing rules). Scoping data use is compliance, not polish.
Control drives adoption: Letting users see and edit what the agent 'knows' increases willingness to share, which improves personalization quality — a virtuous loop when done transparently.
Test Your Understanding
Question 1 (Multiple Choice)
Why is data staleness a personalization risk, not just a quality nitpick?
Question 2 (Multiple Choice)
Which use of data is most likely to cross the 'creepiness' line?
Question 3 (Multiple Choice)
Personalization that systematically shows worse prices to certain demographic groups is primarily...
Question 4 (Short Answer)
How would you decide which pieces of user data are fair to use for which kinds of personalization?
Think about your answer before revealing the explanation.
Question 5 (Short Answer)
What's the tradeoff between richer personalization and the 'creepiness' threshold, and how would you manage it?
Think about your answer before revealing the explanation.