# The Personalization Guide

> A plain, vendor-neutral reference on personalization: what it is, how it works, the data and identity it needs, how it is measured, how it compares to segmentation and recommendation, where it came from, and the ethics.

## Guides

- [What is personalization?](https://personalization.guide/): Personalization is a decision loop: a system chooses what to show each person from data about them, and only controlled measurement shows whether it helped.
- [The data and identity layer](https://personalization.guide/data): Why personalization most often fails at the data and identity layer: freshness, identity resolution, and the gap between the profile and the person.
- [How personalization is measured](https://personalization.guide/measurement): Measuring personalization honestly: holdouts, randomized experiments, lift, and why most published lift figures are vendor-reported estimates.
- [The personalization landscape](https://personalization.guide/ecosystem): Personalization tools mapped by function, not rank: the data layer, decisioning, experimentation, delivery surfaces, and the shift onto the data warehouse.
- [Where personalization comes from](https://personalization.guide/history): A short history of personalization: collaborative filtering, one-to-one marketing, the CDP era, the privacy turn, and real-time and generative systems.
- [Personalization and its critics](https://personalization.guide/ethics): The case against personalization, weighed fairly: filter bubbles, creepiness, fairness and price steering, dark patterns, homogenization, and privacy as ethics.

## Library

- [Techniques](https://personalization.guide/library/techniques): How personalization decisions get made, from rules and collaborative filtering to bandits, uplift, and LLM and agentic decisioning.
- [What is rule-based personalization?](https://personalization.guide/library/techniques/rules-and-segmentation): Rule-based personalization uses hand-written if-then conditions to change what a visitor sees. How it works, what it is good at, and where it breaks down.
- [What is collaborative filtering?](https://personalization.guide/library/techniques/collaborative-filtering): Collaborative filtering predicts what a person will like from the behavior of many users, through item similarity or by factorizing a user-by-item matrix.
- [What is content-based filtering?](https://personalization.guide/library/techniques/content-based-filtering): How content-based filtering works: recommending items by attribute or embedding similarity, its cold-start strengths, and its drift toward sameness.
- [What are hybrid recommenders?](https://personalization.guide/library/techniques/hybrid-recommenders): A hybrid recommender combines collaborative filtering, content, and rules in stages: retrieval, ranking, re-ranking. Almost every production recommender is one.
- [What are multi-armed and contextual bandits?](https://personalization.guide/library/techniques/contextual-bandits): A multi-armed bandit chooses among options while learning which works; a contextual bandit also reads the user and situation to choose per context.
- [What is propensity scoring?](https://personalization.guide/library/techniques/propensity-scoring): Propensity scoring predicts how likely a person is to act, such as buy or churn. It is useful, but a high score does not mean a treatment caused the outcome.
- [What is uplift modeling?](https://personalization.guide/library/techniques/uplift-modeling): Uplift modeling estimates the causal effect of a treatment on each person, so a campaign targets the people it actually moves, not those who would act anyway.
- [What is next-best-action?](https://personalization.guide/library/techniques/next-best-action): Next-best-action (NBA) chooses the most valuable action for a person at a given moment from the eligible options, including doing nothing.
- [What is lookalike modeling?](https://personalization.guide/library/techniques/lookalike-modeling): Lookalike modeling finds new prospects who resemble a seed audience. It is an acquisition technique, not a way to tailor experiences for known users.
- [Reinforcement learning in personalization](https://personalization.guide/library/techniques/reinforcement-learning): How reinforcement learning applies to personalization: sequential decisions, why full RL is rare in marketing, and how to test a vendor's RL claim.
- [What is generative personalization?](https://personalization.guide/library/techniques/generative-personalization): Generative personalization produces the tailored content, recommendation, or interface with a model, instead of selecting from options prepared in advance.
- [What is agentic personalization?](https://personalization.guide/library/techniques/agentic-personalization): Agentic personalization: an LLM agent given a goal, tools, and guardrails that plans and runs the steps itself, not a human-built flow.
- [Comparisons](https://personalization.guide/library/comparisons): The terms the industry conflates, each with a distinguishing test: personalization vs segmentation, targeting, recommendation, and more.
- [Personalization vs customization](https://personalization.guide/library/comparisons/personalization-vs-customization): The difference between personalization and customization: who makes the change, the system from inferred data or the user through explicit settings.
- [Personalization vs targeting](https://personalization.guide/library/comparisons/personalization-vs-targeting): How personalization differs from targeting: targeting fixes the message and varies the audience, while personalization fixes the person and varies the content.
- [Recommendation vs personalization](https://personalization.guide/library/comparisons/recommendation-vs-personalization): How recommendation differs from personalization: recommendation is one mechanism, a ranked list of items; personalization is the broader goal.
- [Segmentation vs individualization](https://personalization.guide/library/comparisons/segmentation-vs-individualization): How segmentation differs from individualization: grouping people into buckets versus tailoring to a single person, with a test to tell them apart.
- [Propensity vs uplift modeling](https://personalization.guide/library/comparisons/propensity-vs-uplift): How propensity and uplift models differ: propensity predicts how likely an action is, uplift predicts the change a treatment causes for a person.
- [A/B testing vs personalization vs bandits](https://personalization.guide/library/comparisons/ab-testing-vs-personalization-vs-bandits): A/B testing, personalization, and bandits all choose among variants but answer different questions. How they differ, with a clear distinguishing test.
- [Real-time serving vs real-time decisioning](https://personalization.guide/library/comparisons/real-time-serving-vs-decisioning): Real-time serving delivers a decision fast; real-time decisioning computes it from current-session signals. Why serving stale data fast is not real-time.
- [CDP vs personalization engine vs marketing cloud](https://personalization.guide/library/comparisons/cdp-vs-personalization-engine-vs-marketing-cloud): How a CDP, a personalization engine, and a marketing cloud differ: the data foundation, the decision layer, and the suite that bundles them.

## Elsewhere

- [Snowplow Signals](https://signals.snowplow.io): Real-time customer context for in-session personalization. Maintains this guide.
- [Generative UI Guide](https://generativeui.guide): The sibling reference site on generative UI.

## For agents

- Every guide and Library article is served as markdown at its path with a .md suffix, for example /library/techniques/collaborative-filtering.md.