AI ads with human control

Partly. Learning phases are real and constant meddling does reset them. But that argument is also used to defer scrutiny indefinitely. The test is whether the account has a measurable hypothesis running; if nobody can say what is being learned, time is not the missing ingredient.

Yes, and it often does. We can sit in as the specialist layer, hand a roadmap to your own team, or own the workstream outright. What matters is agreeing who decides what before we start — most friction between an agency and an in-house team is unclear ownership rather than disagreement about the work.

We agree what we are trying to move, and how it will be measured, before anything starts. Reporting is in plain language: what ran, what the evidence shows, what changes next. Where an outcome depends on platform, market or customer behaviour outside anyone's control, we say so rather than promising a number.