The commercial engine, from the first signal to the signed partner, across four lenses: the Pipeline, the Voice, the Scoreboard and the Brand. For each stage the line marks what a bounded agent may do and where a named human must sign. Every outward word is drafted by an agent and signed by a person; every touch, from a click to a contract, lands on one event spine; and the scoreboard is judged on qualified pipeline, never on reach.
AI takes the sourcing, enrichment, scoring, drafting, chasing, follow-up and reporting: high-volume, rule-bound, reversible work, run from licensed data and a signed voice. The human keeps the audience, the plan, the budget, every published word, every send to a named account and every number that reaches the board: anything that commits the brand, makes a claim, or cannot be unsaid. Oversight attaches to the reasoning: the source, the consent status, the template, the threshold, not just the final signature.
Open-ended generative AI has no place in front of a sponsor, a partner or a fan. Every autonomous stage below runs inside a two-tier architectural lock, so the agent cannot drift, cannot send, and cannot act where it is uncertain.
System prompts are bounded by the signed voice card of each seller, the approved templates, the segment definitions and the house rules on claims. The agent is instructed: if a fact, a figure or a name is not in the sources it was given, do not infer it; mark it for the person. No open-ended drafting to an audience.
The agent's output is checked by structured code against the consent status on the event spine, the scoring thresholds and the send rules before anything is drafted to an individual or routed to a seller. The language model proposes; the rules dispose. Nothing is sent on the model's say-so, and nothing is sent at all without a named person.
# On any send, any claim, any record without consent, any score outside its band: IF consent_status != "opted_in" OR score_outside_band OR recipient_is_named_target OR claim_unsourced: # do not infer, do not send, do not publish: draft it and hand it to a named person OUTPUT STATUS: DRAFT_FOR_NAMED_SIGNER EMIT { lead_id, segment, reason, evidence_pointer } ELSE: PASS to rule_engine → validate → await named human sign-off; the agent never sends
Eight stages across the four lenses of the CMO. Open any stage for what the bounded agent does, where the human signs, the governed prompt that keeps it from drifting, the agent from the CMO catalogue that holds it, and a maturity self-check that reads your drift exposure.
Without these, every autonomy level above is borrowed against trust you do not have.
From 2 August 2026, deferred to 2 December 2027 by Regulation (EU) 2026/1744, the Digital Omnibus on AI, in force since 27 July 2026, EU AI Act Article 14 requires high-risk AI to be designed for effective human oversight. This applies to high-risk (Annex III) systems, while the Article 4 AI-literacy duty already applies to all deployers (since 2 February 2025). This line operationalises it in marketing and commercial work: a bounded prompt and a deterministic check on every stage, a named human on every send, every published word and every claim, and a decision receipt behind every autonomous action. Where a system talks to people, Article 50 asks that they are told; where a message is marketing to an individual, UK PECR asks for consent first. You are not selling an AI deployment — you are giving the CMO a blueprint to scale qualified pipeline while keeping a named human accountable for every decision that binds the firm.
How the AI Augmentation Line™ methodology maps to Regulation (EU) 2024/1689, and what this tool is and is not.