AI Growth Marketer

Write Permissions

Everywhere else, your control depends on the AI following instructions. With Maxma, the system will not let the AI past the permissions you set.

You choose what the AI is allowed to do by itself and what it has to ask you about first. Budgets, new campaigns, creative going live, keywords, tracking: each one is a setting you control.

Those settings live in Maxma, not in instructions written to the AI. It cannot work around them, forget them halfway through a long job, or be talked out of them.

It is the difference between telling a new hire not to touch the budget and not giving them the login to the budget in the first place.

How that compares

Most tools that connect an AI to your ad accounts simply hand it the keys. It can spend your money from the moment it is connected, and the only thing stopping it is a line of text telling it to check with you.

AI with direct account accessMaxma
What stops a change you did not wantA line of text asking it not toYour permissions, which it cannot bypass
Still holds after a long, messy sessionDepends on the AIYes
Still holds if it reads a page trying to trick itDepends on the AIYes
Who sets the rulesWhoever wrote the AI's promptYou, in your settings
What you get to seeWhat the AI says it didThe exact change, before it happens

The control you get

Actions are grouped into classes, and each class carries a setting: review or execute.

ClassExamples
Budgets and bidsRaising a daily budget, changing a target cost per acquisition, switching bidding strategy
Campaign structureCreating campaigns, ad groups or ad sets, turning things on or off
Creative going livePublishing new images or video into an account
TargetingAudiences, interests, placements, geography
KeywordsAdding keywords, adding negatives, pausing terms
TrackingTags, pixels, conversions and key events
Store and catalogProducts, collections, product groups, pages

Set them all to review and nothing reaches an account without a person. Set them all to execute and the AI runs the account, reporting rather than asking. Most teams sit in between, and the useful middle is usually shaped by risk rather than by volume: spend and creative wait for a person, while the housekeeping that is tedious and low-risk, like adding negative keywords, runs on its own.

Three examples of how teams set it:

  • Enterprise: every class set to review. Marketing proposes, a named person approves, and the record shows what was applied and when.
  • A busy owner or founder: budgets and creative wait for a person; keyword clean-up, pacing adjustments and tracking fixes execute.
  • Autopilot within limits: everything executes inside the budgets you set on the platform, with reports rather than requests.

You can change the setting at any time, and tighten it for one account without tightening it for another.

What stays true at every setting

  • Changes are checked before they reach your account. An invalid change is caught rather than applied.
  • A change either lands or is undone. Your account is never left half-changed.
  • You can accept part of what is proposed and leave the rest.
  • Everything is recorded, whether a person approved it or your settings allowed it.

Every change leaves a record

Whatever the setting, nothing happens quietly. Each change the AI makes to a live account is written down, and the record answers the questions you would ask afterwards.

The record showsExample
What changedThe daily budget on one campaign, from $80 to $110
WhenThe time it was proposed, and the time it was applied
Why it happenedThe request, workflow or co-pilot it came from
Who allowed itApproved by a person, or executed automatically under the permissions you set
What came backThe platform accepted it, or the error if it did not

That makes two things possible. You can answer "who changed this, and when?" about your own account without reconstructing it from platform change history. And when a month looks unusual, you can read what was done in order, rather than inferring it from the spend.

Rejected proposals are kept too. What the AI wanted to do and was not allowed to do is part of the picture.

Why this matters more as the AI gets better

The safety of a capable agent cannot rest on it being well behaved. As these systems do more of the work, the only durable control is structural: the AI proposes, the platform enforces, and the boundary is yours to set.

That is also why the control gets more useful over time rather than less. When the work is trustworthy, you move classes from review to execute and keep the audit trail. When something new is at stake, you move one back. The setting is yours, and it is enforced whether or not the AI agrees with it.

Availability

Today every change to a live account waits for a person, and each one is recorded with what changed, when, and what the platform returned. Per-class settings, so you can let some classes execute directly, are rolling out, together with the record showing whether a change was approved by a person or executed automatically under those settings. See approvals and limits for what applies today.

Previous
Co-Pilots