Once upon a time, in Part 1 of Adobe Target series, we covered the difference between testing and personalization and mapped each concept to Adobe Target's activity types. In Part 2, we walked through the organizational model and the iterative process that turns ideas into measured outcomes. Now let's open the hood and look at how each activity type is actually structured.

What do all Target activity types have in common?

Every Adobe Target activity follows the same three-step guided workflow:

Compose experiences (or offers)

Using the Visual Experience Composer (VEC) or the Form-Based Experience Composer (FEC), you create one or more experiences. For A/B tests, Experience A is the control and additional experiences are variants. For Recommendations, you specify locations and insert recommendation modules.

Targeting

Define the audience and traffic allocation. You assign specific percentages of visitors to each experience, choose a traffic allocation method (Manual, Auto-Allocate, or Auto-Target), and create or combine audiences.

Goals and Settings

Configure objectives, priority, duration, reporting source, success metrics, and advanced settings. You set when the activity starts and ends, select conversion or revenue metrics, and decide if additional metrics or dependency rules are needed.

It's easy and just by understanding these three steps helps you navigate any activity type. The differences come down to how experiences are defined, how traffic is allocated, and how content is selected or personalized.

A/B Test (Manual)

A/B tests compare two or more versions of web or mobile content to identify which improves conversions, sales, or other metrics. They are ideal for testing large changes or discrete experiences when you have a clear hypothesis.

After entering an activity URL, the VEC or FEC shows tabs for Experience A (control) and Experience B (variant). You can add more experiences via the Add icon. In the Targeting step, you divide traffic evenly or choose custom percentages across experiences. Totals must equal 100%.

The Goals and Settings step lets you set an optional objective, priority, start and end dates, and reporting source (Target, Adobe Analytics, or Customer Journey Analytics). You choose a primary goal metric (conversion or revenue) and optionally specify additional metrics or dependencies.

What makes manual A/B testing straightforward is the full control over traffic distribution. You decide exactly how much traffic goes where.

How do the "Auto" variants differ?

The A/B test workflow offers three traffic allocation methods. Manual is the default, but Auto-Allocate and Auto-Target change how traffic flows after the activity goes live.

Auto-Allocate

Auto-Allocate uses a multi-armed bandit algorithm to identify a winning experience quickly and dynamically reallocate traffic. It monitors conversion performance and sends more new entrants to high-performing experiences while reserving some traffic (typically 20%) for exploration. The algorithm performs comparisons across all experiences, not just against the control, and continues until the confidence intervals of the best experience no longer overlap with any other.

Use Auto-Allocate when you want to find a winner fast and are comfortable letting the algorithm shift traffic away from underperforming variants.

Auto-Target

Auto-Target uses a Random Forest machine learning model to predict which experience will perform best for each individual visitor, based on their profile and behavior. Unlike Auto-Allocate, Auto-Target does not look for a single winner. Instead, it seeks to maximize conversions at the individual level.

A portion of traffic is reserved as a control group to measure lift. You can choose different control/personalized splits (50/50, 90/10, or custom). Auto-Target can run indefinitely because the model continually retrains and adapts.

Use Auto-Target when you have enough traffic and want machine learning to personalize the experience for each visitor rather than picking one winner for everyone.

Automated Personalization (AP)

Automated Personalization is a Target Premium feature that combines offers or messages and uses machine learning (Random Forest) to match different variations to each visitor's profile. It is designed for ongoing, always-on personalization.

AP automatically assembles offers into combinations and decides which combination to show each visitor. The personalization model rebuilds every 24 hours and reserves a portion of traffic for exploration using a multi-armed bandit approach. It uses automatic data collection from Adobe Experience Cloud Audiences and allows offline data (CRM or propensity scores) to be uploaded via profile parameters or APIs.

The key difference from Auto-Target is that AP works at the offer level (mixing and matching individual content pieces), while Auto-Target works at the experience level (choosing between predefined full-page experiences).

Multivariate Test (MVT)

Multivariate tests compare combinations of offers in multiple page elements (locations) to identify the best combination and determine which elements most influence conversion. They are useful when you need to test multiple elements simultaneously rather than single changes.

Each element on the page (hero image, call-to-action, banner) can have multiple content options. Target uses full-factorial testing, so all possible combinations are tested with equal probability. For example, three elements with three, three, and two offers produce 18 combinations.

Because the number of combinations grows quickly, MVTs require sufficient traffic to achieve statistical significance. Target provides a Traffic Estimator to help you plan. The analysis identifies which element most influences conversion (main effect) and whether combinations interact to affect performance (interaction effects).

Use MVT when you want to understand how multiple page elements work together, not just which single change wins.

Experience Targeting (XT)

We covered XT in Part 1, but here is how it fits structurally. Experience Targeting delivers specific content to defined audiences based on rules. There is no algorithmic traffic reallocation. Experiences are served solely based on rule matching.

Each experience is associated with an audience definition (location, device type, user behavior). When a visitor meets the criteria, they see the corresponding experience. If a visitor qualifies for multiple rules, experiences are served in priority order (highest first). Marketers can reorder experiences to control delivery.

Use XT when you already know who should see what and want deterministic control over the experience.

Recommendations

Recommendations activities automatically display products, services, or content based on user behavior and preferences. This activity type has the most distinct anatomy of all Target activities.

The key components:

• Entities and feeds - The items being recommended (products, articles, videos) are imported via feeds (CSV, Google Product Search, or Analytics classifications) into a catalog.
• Collections - Group similar items from the catalog (e.g., a product category).
• Criteria - Define the recommendation logic ("People who bought this also bought that," "Viewed this, viewed that," or "Similar attributes"). Multiple criteria can be selected, splitting traffic evenly among them.
• Designs - Determine how recommended items appear (list, grid, table) using Velocity templates or default designs.
• Locations - The page area (mbox) where recommendations display. In the VEC, you can replace or insert recommendations at a selected element.
• Promotions (optional) - Front or back promotions to highlight specific items.

The Goals and Settings page defines metrics and duration similar to other activities.

Conclusion or when should you use which?

Join me in this Adobe Target series, where I discuss testing and personalization, integration approaches, and the implementation of use cases.

• Link to Part 1: Adobe Target 101 - Part 1: Testing or Personalization?
• Link to Part 2: Adobe Target 101 - Part 2: From Vision to Variant