Level 4 taught you to govern AI across a marketing agency. Level 5 is about building the part that acts. An assistant suggests; an automation does. It queues a post, uploads a list, sends a text, changes a bid or answers a customer, and when it is wrong the error has already reached the public, a client's account or a person who opted out by the time anyone looks. This level is for marketing technologists, marketing operations and CRM owners, ad-operations and programmatic leads, social publishing owners and the agency staff who run automation projects. It teaches automation whose authority is bounded in writing, whose every action is recorded, whose consent and opt-out state travels with the data, and whose failures are caught by design. No coding is assumed or required.
The first three modules set authority and orchestrate the work. Module one places every proposed agent at one of three levels, suggest, act inside the agency, or act outward only with a named person's approval, reviews it for excessive functionality, permissions and autonomy, writes the agent authority register before it touches a client account, and names the acts that never go to an agent. Module two automates onboarding, brief and claims intake while people decide client acceptance, scope and whether each claim is supported, checks that customer lists were collected first-party, and recognises when an audience service could make the agency a data broker. Module three binds approval to the exact version queued in a publishing calendar, keeps disclosures in the post rather than in a toggle, routes election-period content out of automated flows under the Michigan, Washington and platform rules, and pauses a calendar before one error cascades.
Modules four to six are data and consent. Module four draws the integration map, writes a data contract at each boundary, treats every pixel, tag and hashed-identifier upload as a disclosure, and quarantines missing, malformed, poisoned and drifting records before they drive targeting. Module five carries consent and opt-out state through every system, engineers opt-out preference signals under California, Connecticut, Colorado and Texas law, each state named, keeps opting out as easy as opting in, and tests consent tools against what the Sephora, Todd Snyder and Tractor Supply matters found. Module six builds consent gates for automated calls and texts: prior express written consent, the FCC's 2024 ruling that AI-generated voices are artificial voices, do-not-call and calling-hour checks, revocation by any reasonable method, and CAN-SPAM opt-outs in automated email. It teaches no federal AI-call disclosure duty, because none is in force.
Modules seven to nine are human control. Module seven designs approval gates that show a reviewer evidence rather than a verdict, explains what the research on automation bias and complacency found, sizes gate volume to the people who carry it, and measures catch rates with seeded known errors. Module eight gives every automation its own identity and never a shared staff login, applies least privilege in ad accounts, CRM, email and social platforms, segregates clients, classifies output at the level of its input, and protects automation administrators with strong sign-in and access reviews that remove. Module nine names who may stop and restart, halts scheduled sends, bids and posts cleanly, keeps one stop from breaking the next automation, honours opt-outs on time while everything is stopped, and tests the manual route before it is needed.
Modules ten to twelve are defence and observation. Module ten threat-models competitor pages, reviews, comments, creator submissions and inbound mail from the defender's side only, limits privileges by data trust, constrains actions outside the model, and treats prompt injection as a residual risk that controls reduce but no product removes. Module eleven specifies the agent action record, what never goes into a log in clear, retention set from purpose with the EU AI Act, a Colorado law and an FTC order read as examples rather than rules for ordinary campaigns, and a trail a client, a reviewer or a regulator could follow. Module twelve finds failures that raise no error, retests after model, data and platform changes, bounds consumption and spend, and watches channel metrics as AI search summaries spread.
Module thirteen responds when an automation has already acted: containing a false claim, a send to people who opted out, a call without consent, a health-data pixel under the FTC's Health Breach Notification Rule where it applies, a chatbot that misstates terms and an impersonation of the client, then telling the client and running a review that changes the gates. Module fourteen counts time, quality, rework, incidents, oversight and spend without double-counting, explains why no published study can replace the agency's own baseline and holdout, and takes a pilot through stop criteria to a standard practice decision. The capstone is the Automation Implementation Project for a fictional marketing agency. The level ships with a printable workbook and ten templates, and the examination draws forty scenario questions from a reviewed bank.
Everything here is professional education. It is not legal, privacy, telemarketing, advertising-compliance, election-law or security advice, and it does not replace counsel, a platform's own policies or the law that applies to a campaign. Completing the level earns an independent educational certificate issued by AI Coalition Network with a public verification page. It is not a licence, a platform certification, a security credential, a professional designation or a regulatory approval, it carries no professional education hours, and it satisfies no licensing or compliance requirement.