Experience Platform · sandbox tooling
From development to production
The abandoned cart use case works in the development sandbox. Now it has to reach production through a sandbox tooling package. Choose what goes in and see what arrives ready, what arrives as a draft or disabled, what is reused without updates and what has to be rebuilt by hand.
Example assumptions
Example project, objects and volumes. This models normal object import or a full sandbox into an empty target, not the update workflow. Rules checked against Adobe documentation on 17 September 2026.
Development sandbox
How each object arrives
The thirteen objects in the use case, one per row. The column shows how each one reaches production, and the red one is what has to be rebuilt by hand.
Production sandbox
What needs preparation and completion
Some tasks precede the import; others follow it. The checklist distinguishes what is copied, reused and needs preparation.
The recommendation
Separate what is copied, what is reused and what needs preparation
A package is not a full deployment: it moves definitions, not data, credentials or activations. Whatever is missing should be written down before you start.
- 01
Dependencies before import
A journey needs an existing target merge policy. Decisioning can be transported, with conditions: add selection-strategy items to the package and copy them first. Ranking formulas using AI models cannot be copied.
- 02
Reuse is not an update
Normal import reuses objects with the same name. Update existing objects handles supported changes to schemas, field groups and data types. Review the differences; not every modification is supported.
- 03
Activate in order
Enable datasets for Profile, reconnect credentials, activate dataflows and review merge policies and audiences. Only at the end, publish journeys and activate destinations.
- 04
An intermediate sandbox
Test in staging first. Compare permissions, credentials and dependencies with production: the same package can behave differently when the environment changes.
Explore this decision
- Can the AI read, or can it also change things?
Coworker, agents and MCP in Adobe: separate reading, proposing and executing. Context, permissions and review before changing a platform.