DSP AI Managed Group: Operating Guide

1. Overview

DSP AI Managed Groups allow you to manage multiple Amazon DSP campaigns (Orders) as a single unit. Under a unified set of objectives, the system coordinates optimization across the group to improve overall performance. You can use Action Space to explicitly authorize what the AI is allowed to change, so you gain automation benefits while maintaining control over risk boundaries.

Managed Groups are typically used in the following scenarios:

You run multiple strategies or product lines in parallel and want to manage overall performance with one consistent objective framework.
You need ongoing bid, budget, or targeting optimization, but manual maintenance is time-consuming.
You want clear authorization boundaries for AI actions and the ability to trace and audit changes through logs.

A Managed Group includes the following core modules:

Objectives: Configure the Promotion Objective, Primary Objective (including KPI), and Secondary Objective definition.
Action Space: Configure which dimensions the AI can adjust (e.g., bids, budgets, targeting, creatives) and the allowed boundaries.

2. Before You Start

Before creating a Managed Group, please review the following rules and limitations (the in-product UI is the source of truth):

Site and currency: The UI displays amounts in the local currency for the selected site. Dates and times follow the local time zone under DSP rules.
Eligibility: Only Orders that are not already in another Managed Group can be selected (ineligible Orders are automatically filtered out).
Limits: Each Managed Group must include at least one Order. There is a maximum number of Orders per group (as shown in the UI).

If you cannot find an Order in the “Select Orders” dialog, the most common reasons are that the Order is already assigned to another Managed Group or it does not meet the eligibility conditions at the moment.

3. Key Concepts and Objective Setup

Optimization objectives in a Managed Group consist of three parts: the Promotion Objective, the Primary Objective (including KPI), and the Secondary Objective. To avoid misunderstanding, review this section before configuring your group.

3.1 Promotion Objective

The Promotion Objective defines the overall direction for optimization. Common options include:

Drive Growth: Focused on conversion and sales growth. Optimization tends to emphasize purchase outcomes and improving return on ad spend.
Drive New Traffic Increase: Focused on reach expansion and traffic generation. Optimization tends to emphasize impressions, clicks, and detail page views to build an audience for future conversion.

Selection guidance: If you primarily care about conversion and sales outcomes, choose “Drive Growth”. If you primarily care about expanding reach and bringing in traffic, choose “Drive New Traffic Increase”.

3.2 Primary Objective and KPI

The Primary Objective is the core performance metric for the Managed Group (for example, ROAS, CPC). The KPI expresses the performance level you want to reach and is an important input to the system’s optimization decisions.

KPI recommendations:

If you are unsure of the KPI value, start with the system “Recommended” KPI and then adjust slightly based on your business target.
Start with an achievable KPI and tighten (or raise) it gradually after performance stabilizes, to avoid volatility caused by overly aggressive targets.

3.3 Secondary Objective

The Secondary Objective is used for further optimization once the Primary Objective and pacing are relatively stable. Two common definitions are available (as shown in the UI):

Set individual order objectives: Continue using each Order’s own objective/definition. Recommended when Orders in the group have materially different goals.
Set unified AI group objective: Use one unified definition for the entire group. Recommended when you want consistent optimization criteria across all Orders.

3.4 Optimization Priority

Managed Group optimization follows a clear priority order: the system first safeguards the Promotion Objective direction (for example, pacing direction), then optimizes toward the Primary Objective, and finally considers the Secondary Objective when conditions allow. This helps avoid the overall delivery drifting while pursuing a single metric.

4. Create and Enable a Managed Group

Follow the steps below to create and enable a Managed Group:

On the Managed Group list page, click “Add Managed Group”.

In the “Select Orders” dialog, search or filter the Orders you want to include and confirm the selected list.

Fill in the Managed Group name and identifier. Use a clear naming convention (for example: Quarter_ProductLine_ObjectiveType).
Set the Promotion Objective, Primary Objective (including KPI), and the Secondary Objective definition.
In Action Space, configure which dimensions the AI can adjust and the boundaries (see Section 5).
After confirming all settings, turn on the AI toggle, confirm, and start running.

Note: The AI toggle is typically OFF by default at creation. Enable it only after you have confirmed your objectives and Action Space configuration.

5. Action Space Configuration

Action Space is where you explicitly authorize which optimization actions the AI can execute. Each action usually includes an on/off toggle and optional parameters to define operational boundaries and risk limits. We recommend enabling actions in phases: start with the 1-2 most critical dimensions, and expand gradually after performance stabilizes.

Recommendation: Prioritize enabling Budget Optimization. Budget is the key lever for pacing and efficient allocation of spend across multiple Orders. In multi-Order scenarios, budget optimization often addresses practical issues more directly (for example, underspend, uneven spend distribution, or budget allocation lagging behind performance changes). After Budget Optimization is stable, consider enabling Bid, Targeting, and Creative actions.

5.1 Bid Optimization

Purpose: Allows the AI to adjust bids at the Line Item level (for example, Base Bid and Max Average CPM). You can use parameters such as Base Bid Range to constrain how far bids can move.

How it works: The system considers performance signals and the selected bid strategy to coordinate adjustments between Base Bid and Max Average CPM. If Dayparting is enabled, the system can apply hour-level bid values across a 24-hour schedule and optionally use eCPM as a reference to avoid bids that are too high or too low.

Key settings:

Base Bid Range: Constrains bid movement. For first-time use, set a reasonable range to control volatility.
Dayparting: If performance varies significantly by time of day, select the hours to apply hour-level bidding. If no hours are selected, the UI typically defaults to all hours.
eCPM reference: Uses the current Line Item’s eCPM as a reference to avoid bidding too aggressively or too conservatively.

Bid strategy and what the system tends to adjust:

Prioritize spending full budget: The system tends to place more emphasis on Max Average CPM adjustments.
Prioritize KPI target: The system tends to place more emphasis on Base Bid adjustments.

Note: The system will also validate the relationship between Base Bid and Max Average CPM to avoid conflicts (for example, Base Bid should be lower than Max Average CPM). If you set a Base Bid Range, ensure it aligns with your bid strategy expectations.

Recommendation: If delivery stability is a priority, enable Bid Optimization first and apply a conservative range. After 3-7 days of stable performance, you can gradually relax boundaries if needed.

5.2 Budget Optimization

Purpose: Allows the AI to dynamically allocate budget within the Managed Group (at the Order level and/or Line Item level), and optionally adjust the budget ceiling (Budget Cap) based on your authorization.

How it works:

Allocate across levels: If “Allow AI to allocate Order budgets” is enabled, the system can redistribute budget across Orders in the group. If “Allow AI to allocate Line Item (Ad Group) budgets” is enabled, the system can rebalance budget at the Line Item level and prioritize spend toward units that better support current objectives and pacing.
Rebalance based on pacing: The system uses the current Flight’s execution progress (for example, actual spend vs. expected spend through yesterday) and remaining time to identify underpacing/overpacing risk and adjust budgets to bring delivery closer to plan.
Operate within your boundaries: If you configure the budget adjustment unit/range (percentage or fixed amount), the system will execute strictly within that boundary. If not configured, the system follows the default limits in the UI.
Coordinate with Budget Cap: If “Allow AI to adjust Budget Cap” is enabled, the system may increase or adjust the Order’s budget ceiling when scaling is needed or to address underpacing. If the authorization is not enabled, the system will not modify Budget Cap and will only optimize within the existing ceiling.

Applicable scenarios:

You run many Orders in the group and budget allocation requires frequent adjustments.
You want to maintain pacing while shifting budget toward better-performing Orders/Line Items.

Key settings:

Conflict with Amazon automatic budget optimization: If Amazon automatic budget optimization is enabled, the system may restrict Line Item budget adjustments. Disable Amazon automatic optimization to allow AI Budget Optimization to operate as intended.
Multi-level budget allocation: You can allocate at the Order level and/or the Line Item (Ad Group) level, depending on the options shown in the UI.
Flight-aware optimization: Budget allocation is dynamically adjusted based on the current Flight’s execution progress and remaining time to better match pacing.
Budget adjustment limits: Set upper/lower bounds (for example, max/min budget ratio) to control the magnitude of budget movement and reduce volatility.

Budget Cap controls:

What is Budget Cap: A ceiling that limits the maximum spend for an Order within its configured time period.
When to enable: If you want the AI to be able to adjust the budget ceiling (including increasing it) to expand delivery, enable this authorization. The AI will only change Budget Cap when this is enabled. If you require a strict cap mode (hard cap), make sure this option is OFF.

Advanced setting: limit the magnitude of budget adjustments

If the UI provides a unit/range setting after enabling “Allow AI to allocate budgets”, configure it to control risk:

Percentage: Limit the adjustment magnitude (for example, no more than 10%-20%). Recommended if you want steadier performance.
Fixed amount: Limit adjustments by absolute currency amount (for example, no more than $50-$200). Recommended for smaller budgets or tighter control.
If you want the AI to fully optimize pacing and efficiency, choose the most permissive option available in the UI.

Recommendation: If your primary goal is to deliver steadily and allocate budget toward the most effective delivery units, Budget Optimization is usually the highest-priority Action Space. Start with conservative boundaries, then gradually relax or add Bid/Targeting actions after stabilization.

5.3 Targeting Optimization

This Action Space includes two independent capabilities: Targeting Optimization (optimize existing Line Item targeting conditions) and Audience Harvesting (test recommended new audiences and scale them). You can enable them separately and phase them in based on maturity.

5.3.1 Targeting Optimization

Purpose: Allows the system to dynamically adjust targeting conditions (within the boundaries you authorize) that impact delivery efficiency, including Inventory / Supply Source, Frequency, and Viewability.

How it works: Based on the current Flight’s pacing and Primary Objective performance, the system can add/remove or tune these settings to expand eligible inventory and improve unit efficiency. All changes are recorded in the AI Log and reflected in the corresponding AI Summary modules.

Applicable scenarios:

You experience limited impressions or spend and need to expand eligible traffic supply.
Your targeting setup is complex (multiple inventory/frequency/viewability combinations) and is difficult to maintain manually.
You need ongoing balancing between brand safety/viewability requirements and cost efficiency.

Configuration and usage guidance:

Use together with Budget Optimization and Bid Optimization. If budget/bid constraints are too tight, Targeting Optimization alone may not materially improve pacing.
If you have hard requirements for inventory, frequency, or viewability, configure those fixed constraints in the advanced settings first, then enable Targeting Optimization to avoid policy or governance conflicts.

Advanced setting: Targeting rules (by Order)

After enabling Targeting Optimization, it provides “Set order targeting rules”, you can further constrain the optimization boundary:

Inventory (Supply Source): Select the inventory types the AI is allowed to optimize within (for example, Amazon only, or Amazon + APD + Third-party exchange).
Viewability / Frequency: Define viewability and frequency targets or ranges (as provided in the UI).
You can filter by Order goal and apply bulk replacement, which is useful when the group includes many Orders and you want consistent rules.

5.3.2 Audience Harvesting

Purpose: Allows the system to identify and test high-potential audience segments and bring them into delivery for scaling reach or improving audience quality. AI Summary provides a dedicated Audience Harvesting view, including comparative reporting (AI-harvested audiences vs. manually targeted audiences).

How it works:

Audience recommendation and testing: The system selects audiences from available pools that best fit your objectives for testing. In supported versions, it may create new audiences or Line Items and attach them to your existing delivery structure (as shown in the UI).
Performance comparison and scaling: The system continuously compares AI-harvested audiences with manually targeted audiences and gradually expands coverage on higher-performing audiences while reducing investment in lower-performing audiences, creating an “explore -> select -> scale” loop.

Key mechanisms:

Budget sufficiency check: New audience testing typically starts only when sufficient budget capacity is available. If budget is tight or underpacing is severe, the system generally prioritizes pacing and the Primary Objective over exploration.
Line Item rotation: To free up budget for testing, the system may reduce or pause long-term underperforming Line Items/audience combinations. This requires authorization in “Structural actions (pause/remove underperforming Line Items)”.

Applicable scenarios:

You want to expand reach and discover new audiences, or current audiences are fatigued and performance is declining.
You have sufficient budget and want to improve audience quality through ongoing rotation and testing.

Usage recommendations:

Use together with Budget Optimization: harvesting requires stable test budget. Typically enable Budget Optimization first (with reasonable boundaries), then enable Audience Harvesting.
Monitoring and evaluation: In AI Summary, review tested audience count, average daily spend, and AI vs. manual performance comparison. In AI Log, track new audiences/Line Items and the stated reasons for changes for internal reviews and customer-facing explanations.

Advanced setting: Select harvesting types (by Order)

It provides “Set target harvesting type for each order”, choose what the AI should primarily test:

Audience targeting types (e.g., In-Market, Interest): Suitable for scaling and audience rotation.
Contextual targeting types (e.g., Product, Category, Similar product targeting): Suitable for intent-aligned expansion around products/categories.
You can filter by Order goal and apply bulk replacement.

5.3.3 Structural Actions (Pause/Remove Underperforming Line Items)

Purpose: When authorized, the system can pause or remove persistently underperforming Line Items to free budget and traffic capacity. This is commonly used together with Audience Harvesting for audience rotation.

Recommendation: Structural actions change your delivery structure. For first-time use, start with a small Managed Group and closely monitor the rationale, the number of paused/removed units, and subsequent performance in AI Log and AI Summary.

5.4 Creative Optimization

Purpose: Allows the AI to adjust creative weights and match creative types (for example, Component/Display/Video) based on performance.

How it works: Creative Optimization assigns weights at the Line Item level and adjusts delivery share based on performance signals such as CTR and conversions. The system considers creative type and size and only optimizes where it can meaningfully affect delivery.

Applicable scenarios:

A Line Item contains many creatives across sizes or types, and creative weight allocation requires continuous tuning.
You want to reduce manual effort for creative iteration and allocation.

Key settings:

Type-specific optimization: Different creative types (for example, REC vs. Video) typically have separate weighting to avoid ineffective cross-type competition.
Size matching: Creatives of the same type but different sizes typically participate in weighting separately to reflect actual placement and inventory availability.

Recommendation: Creative Optimization yields clearer benefits when you have a large creative set that is hard to manage manually. If your strategy relies on strict manual scheduling or you only have a few creatives, consider keeping it off initially.

6. FAQ and Troubleshooting

6.1 Cannot select or search for an Order

The Order is already assigned to another Managed Group (it will be filtered out automatically).
The Order does not currently meet eligibility conditions (for example, it is expired).

6.2 AI is enabled but there are no visible actions

The system evaluates conditions daily and may not take actions in every Action Space every day. If there are no actions for an extended period, check the following in order:

Confirm Orders are in a deliverable state (for example, Delivering).
Confirm the Managed Group AI toggle is enabled and the configuration is saved successfully.
Confirm the required Action Spaces (e.g., Bid/Budget/Targeting) are enabled and constraints are not overly restrictive (for example, a very narrow Base Bid Range).
If you want Budget Optimization at the Line Item level, ensure “Allow AI to allocate Line Item budgets” is enabled and Amazon automatic optimization is disabled.

6.3 Goal achievement is unstable or highly volatile

Check whether the Primary Objective KPI is overly aggressive. Consider loosening to a more achievable range.
Check whether you enabled too many Action Spaces at once. Consider phasing in actions and applying boundaries (for example, enable Bid Optimization with a range first).

7. Best Practices

Group by business dimension: Create Managed Groups by product line, campaign stage, or objective type to align goals and simplify review.
Enable in phases: Prioritize Budget Optimization (with boundaries and Budget Cap strategy), then layer in Bid Optimization after stabilization. Once pacing and core performance are stable, expand to Targeting and Creative actions.
Set KPIs reasonably: Start from the Recommended value with small adjustments. Avoid setting KPIs too tight at once, which can reduce stability.

DSP Ad Management Overview

Data Management

Last modified: 2026-01-30Powered by

Outline