Every careful buyer asks the same question within the first 10 minutes: "What is your AI allowed to do without human approval?"
The right answer is not "a lot" and not "nothing." It is a level of autonomy you chose on purpose, that you can point to and explain.
We use a five-level autonomy ladder for this conversation. One thing to be clear about up front: Marketeer runs at L1 today. Everything below L1 on this page is how we think about the levels above it, not a list of switches you can flip in the product right now.
L1: Draft, route for review
The system drafts. A human reviews everything. Every asset, every claim, every channel object. This is where most teams want to start, and it is how Marketeer works today: nothing publishes or spends without your yes.
L2: Draft, route in parallel
Same as L1, but approvers run in parallel rather than serial. Brand, compliance, and spend approvers all see the request at once; the slowest reviewer is the bottleneck, but the queue is half as long. Surface area for governance is the same.
L3: Publish pre-approved variants
The system can publish variants that match a pre-approved template. New claims still need approval; rewordings of approved claims that pass the brand-voice and reading-age checks can go live. Typical use: refreshing creative within an existing campaign without re-routing claims.
L4: Reallocate budget within guardrails
The system can shift budget between cells within a single campaign, up to defined caps, when a winning variant reaches 95% confidence. Spend approvers still gate the original budget release.
L5: Reserved
L5 stays deliberately undefined. Reserving the level signals there is a ceiling: Marketeer will never silently spend money or make claims without a defined human path, whatever we ship at L2 to L4.
Why this framework matters
It turns a vague conversation about "AI autonomy" into a concrete configuration choice. Buyers stop asking abstract questions about safety and start asking specific questions about which level you want for which workflow.
It also sets the bar for the audit trail. As we ship each level, the log has to record the actor type (user, agent, system) and the autonomy level that authorised the action, so anyone asking "who decided this?" gets a precise answer. If a level can't meet that bar, it doesn't ship.
Where to start
Start at L1 and stay there until the review loop feels boring. Only consider more autonomy for a workflow once you have baseline cycle-time data, a clean library of approved wording, and a documented stop-loss rule for anything that touches spend.
If you are evaluating an AI marketing platform that won’t commit to a defined autonomy ladder, that’s a signal.