The personalization landscape

This page maps the personalization ecosystem by function rather than by rank. The goal is an evergreen map a reader can orient by, not a leaderboard that is wrong within a quarter. Vendors move between categories, product names change often, and most larger platforms now span several of the layers below. Read the map for the shape of the market and the direction it is moving, and treat any single-quarter fact as a snapshot.

How to read this map

Two things make the landscape confusing, and both are structural rather than accidental.

The first is that one company usually runs several personalization systems at once. That is because different teams grew their own tools. Marketing owns the clouds, email, and on-site engines; product owns experimentation and feature flags; data science owns the warehouse, feature stores, and custom models; and customer service owns its own routing and assistance. Each added personalization features on its own schedule, so the overlap in the market mirrors the overlap inside the buyer.

The second is that the categories below are blurring. A tool sold as a customer data platform now also decides and orchestrates; a tool sold for experimentation now also targets and personalizes. The functional layers are still the right way to understand the space, but few products stay inside one of them.

The examples named below are illustrative as of 2026, listed in alphabetical order. They are neither endorsements nor a complete list, and product names in this space are renamed often.

The functional layers

Layer	What it does	Examples (2026)
Packaged CDPs	Collect events, resolve identity, build profiles and audiences, and keep their own copy of the data	Adobe Real-Time CDP, Salesforce Data Cloud, Segment, Tealium, Treasure Data
Composable, warehouse-native CDPs	Build and activate audiences from the company’s own warehouse through reverse ETL or zero-copy, without a second copy	Census, Hightouch
Personalization engines	Decide and render web and app content, recommendations, and offers, often with testing built in	Adobe Target, Bloomreach, Dynamic Yield, Optimizely
Recommendations and search	Retrieve, rank, and merchandise catalog items	Algolia, Amazon Personalize, Constructor, Coveo
Email, CRM, and marketing clouds	Multichannel messaging, journeys, segmentation, and decisioning	Adobe Journey Optimizer, Braze, Klaviyo, Salesforce
Ad platforms	Targeting, lookalikes, and auction optimization	Amazon Ads, Google Ads, Meta, The Trade Desk
Clean rooms	Privacy-controlled joins and aggregates across parties	AWS Clean Rooms, Google Ads Data Hub, InfoSum, LiveRamp, Snowflake
Orchestration and decisioning	Sequence interactions and arbitrate the next best action	Adobe Journey Optimizer, Hightouch, Pega, Salesforce
Experimentation and feature flags	Run experiments, flags, holdouts, and sometimes bandits	AB Tasty, GrowthBook, LaunchDarkly, Optimizely, Statsig, VWO
Feature stores and ML platforms	Keep training and serving features consistent and fresh for custom models	Databricks, Feast, Google Vertex AI, SageMaker
Measurement	Marketing mix modeling and incrementality	Google Meridian, Meta Robyn
LLM and agent personalization	Agents for audiences, content, and actions, and assistant-mediated commerce	Agentic commerce protocols, warehouse and cloud agent products

Data and identity

This is the foundation, and it is covered in depth on the data and identity layer. In the market it splits into two shapes. Packaged customer data platforms ingest events, resolve identity, and store their own copy of the profile. Composable, or warehouse-native, platforms leave the data in the company’s warehouse or lakehouse and build audiences on top of it. The second shape is growing faster, which is the clearest single signal of where the market is heading.

Deciding what to show

Three categories make the actual choice. Personalization engines decide and render content, offers, and recommendations on the page, and usually bundle testing. Recommendation and search services specialize in retrieving and ranking items from a catalog. Orchestration and decisioning tools sit above individual channels and arbitrate which action to take next, including the option to do nothing. The vocabulary here moves fast, with “AI decisioning,” “next best action,” and “agentic marketing” used loosely, so treat each as a claim to test against what the system actually does. The methods behind these tools are on how personalization works.

Measuring it

Experimentation platforms and feature-flag tools run the randomized tests and holdouts that tell whether a change caused anything, and some now include bandits. Marketing-mix and incrementality tools answer the aggregate question of what each channel contributed. The methods, and the ways lift figures get inflated, are on how personalization is measured.

The infrastructure beneath it

Feature stores and ML platforms are personalization infrastructure owned by data science rather than marketing. They exist to keep the features used to train a model consistent with the features served at decision time, and to serve those features fresh. Their rise, and the acquisition of feature-store specialists by the large data platforms, is part of the same move toward deciding on data where it lives.

Privacy infrastructure

Consent and clean rooms are now part of the stack, because they determine what data a decision system is allowed to read. Clean rooms let parties join and aggregate data under privacy controls, often with differential privacy. Consent signals and governance increasingly sit inside the data platform itself, as access policies, purpose tags, and consent-aware audiences. The ethics and regulatory history behind this are on personalization and its critics.

Delivery surfaces

Finally, the decision has to land somewhere: a web or app slot, a search result, an email, a push, a paid placement, a price, or a support interaction. Marketing clouds, email and CRM platforms, and ad platforms are as much delivery surfaces as they are deciders, which is one more reason the categories overlap.

The convergence story

The strongest pattern in the market is convergence. The same capabilities are arriving in the market from several directions at once.

• Customer data platforms are adding predictive traits, orchestration, and decisioning, moving from holding the data to acting on it.
• Experimentation and personalization are merging: testing tools are adding targeting and bandits, and personalization engines are adding testing.
• Marketing clouds are rebuilding their personalization onto a shared customer-data foundation rather than running it as a separate product.
• Data platforms are absorbing the CDP, blurring the line between a data platform and a customer data platform.
• Analytics and observability vendors have been acquiring experimentation vendors, folding causal measurement into larger suites.

The practical effect is that the category a tool is sold under tells you less than it used to. The functional questions on what is personalization (who decides, on what data, how fresh, where it acts, how it is measured) are a more reliable way to compare two products than the label on the box.

The shift onto the data warehouse

Underneath the convergence is a single architectural move: from “copy the data into a separate tool” toward “decide on the data where it already lives.” Composable CDPs, zero-copy federation, and lakehouse-native decisioning all follow this pattern. The appeal is that it shortens the distance between the data and the decision, reduces the number of copies to govern, and lets the same governed data serve analytics and activation. The limit is latency: warehouse-native designs are strongest at batch decisions that draw on history, and have had to add streaming and caching to serve in-session decisions within a sub-second budget. The batch versus in-session divide, which is where this architecture is tested hardest, is covered on the data and identity layer.

New surfaces that move control elsewhere

The most consequential near-term change is that some personalization is moving outside the merchant’s own stack entirely.

Agentic commerce protocols let an AI assistant handle discovery and checkout on a person’s behalf. When that happens, the assistant’s model of the user decides what is shown, not the merchant’s personalization engine. The merchant’s levers shift toward the quality of its product data and the clarity of its structured offers, because those are what the assistant reads.

Generative UI is a related shift on the interface side: instead of filling a fixed template, a model assembles the interface for each request. Where personalization composes the interface itself, it overlaps with generative UI. This guide covers what is selected or generated for whom and why; the sibling Generative UI Guide covers how interfaces are composed at runtime. Both raise the same measurement problem, which is how to evaluate an experience when every person sees a different one.

These surfaces are early, and some of the specific launches will look dated quickly. The durable point is that control over the decision is moving from the merchant toward the assistant and the model, which is a continuation of the long contest over who holds the model of the person described in where personalization comes from.