Overview
Growth Engine uses campaign or resource scope, action permissions and guardrails to control supported work. This reference explains their different roles. Available controls depend on the connected platform and account; a prompt instruction does not establish a software-enforced limit.
Operating modes
Mode | Meaning | Boundary |
Control | Keeps selected scope outside managed work. | Can prevent recommendations for that scope. It is not a review-only Copilot preview and does not disable unrelated workflows. |
Copilot | Prepares supported actions for required human review. | Approval does not prove the connected platform accepted the change. |
Autopilot | Allows supported actions to run within saved scope and controls. | Does not make every action autonomous or remove platform requirements and approval exceptions. |
Campaign scope and action permissions
Campaign settings answer where can Growth Engine act? Action permissions answer which changes may run automatically? Automatic execution requires the relevant scope and action permissions to allow it, together with applicable limits and platform requirements.
Illustrative setting | Intended effect |
Established campaign: Autopilot | The campaign is eligible for permitted automatic changes. |
Toggle Status: Autopilot | Eligible status changes can run automatically within the configured controls. |
Create Ad: Copilot | New-ad proposals still require review, including in the Autopilot campaign. |
This is an example, not a default. Action settings can be shared across jobs. A setting shown inside a job is not necessarily private to that job, and a separate action-policy combination may not be available for every individual campaign.
Warning: Store promotions always require approval. An Autopilot label in an action catalog does not authorize automatic promotion creation.
Guardrail controls
Control | What it governs | Important distinction |
Budget and bid limits | Maximum cumulative changes over a selected period, using applicable amount or percentage limits. | The comparison is with the starting value for the period, not just the previous change. A starting budget of $100 and a proposed $125 is a cumulative increase of $25, or 25%, even if an intermediate change set it to $110. |
Cooldown periods | Windows for recently modified entities, separately covering relevant budget, bid/target or status changes. | While a cooldown applies, the agent does not create optimization tickets for entities modified within that window. It does not disable unrelated tools or changes. |
Store promotion controls | Whether discount-code proposals are allowed and the monthly discount-code budget in USD. | An empty budget keeps discount codes off. Budget use is based on potential discount exposure: usage limit multiplied by the largest discount per order. Twenty uses at a maximum $10 discount counts as $200, not actual redeemed value. Approval is still required. |
Creative generation | Separate permissions for image generation, video generation and landing-page drafts. | When generation is disabled, agents use existing assets. Generation or drafting does not authorize publication, sending or ad activation. |
Where settings apply
Setup → Guardrails brings together scope, action permissions and controls. Relevant settings also appear in jobs. Connected-account permissions, the target account and publishing settings still apply.
Strategic instructions explain priorities and exceptions; configured limits control supported changes. Scheduled checks do not guarantee an immediate stop or a hard spend ceiling. Before enabling a job, review the saved settings as described in Set up your first Growth Engine job.
If a required boundary cannot be expressed by the available controls, do not enable automatic work for that scope. Ask support to clarify the supported configuration, including how a change affects already-pending actions.