One large model writing a cold email is a party trick. A team of named specialists that establishes the facts, scores the account and drafts a compliant approach in three ordered waves is a system. Here is how they hand off.
Why waves beat one prompt
Ask a single model to research a company and write outreach in one pass and you get plausible mush. It hallucinates a little, it flatters the prospect, and it produces copy that could have been sent to anyone. The failure is architectural, not a matter of a better prompt. Research, judgement and composition are different jobs with different failure modes.
The Strategy Room splits the work across named specialists, each with a narrow mandate, a defined input, and a structured working paper the next wave can rely on. The waves run in order, so a specialist never has to invent something an earlier one should have established. The pipeline is legible: a human can open any working paper and see exactly why the plan says what it says.
Before the waves: the brief becomes orders
You supply the brief: two or three researched targets, the objective, the offer, the persona, the tone and the deal size. Most of it prefills from your company profile. A commander step then tailors each specialist's orders to your brief, so the dossier builder knows which facts matter for your offer and the strategist knows what a successful outcome looks like. The specialists and their waves are shown before you commit.
Wave one: gather
SAGE builds the corporate dossier for each target: legal identity, ownership, size, technology stack and the named decision-makers in professional roles. Every fact carries its source and confidence, and fields no source could confirm are listed as not established rather than filled in.
SCOUT, working alongside, finds dated signals from the last twelve months: hiring, executive changes, ownership changes, funding, coverage. Undated observations are not signals. SCOUT also proposes lookalike accounts worth adding, which is often where the most useful leads come from.
Wave two: analyse
PRISM scores each target against your ICP, criterion by criterion, and tiers it: strike, nurture or drop. A low score is a feature. The Strategy Room is as useful for telling you not to spend a slot as it is for telling you to.
STRAT writes the approach: the single angle to lead with, the message skeletons, and a cadence sized to the capacity you stated. It commits to one story rather than hedging across five. WARDEN reviews the drafted approach per jurisdiction: the legal basis for contact, sender identification, opt-out handling, and AI-disclosure duties where they apply. Where a message cannot be made compliant, WARDEN says so.
The most valuable output of the Strategy Room is often the account it tells you to skip. Restraint is what keeps a sender's domain reputation alive.
Wave three: report
QUILL reads every working paper the earlier waves produced and writes the boardroom report: executive summary, key findings, prioritised actions, risks and compliance, an honest confidence section, and de-duplicated sources. QUILL has no research tools of its own, so it cannot introduce a claim that no working paper supports.
If one specialist fails, the wave continues and QUILL is told exactly what is missing. The report names the gap in its confidence section instead of papering over it.
A plan for your review
What reaches you is a plan, not a paragraph: scored targets, a chosen angle, a full cadence, and a compliance note explaining what was changed and why. It stays marked as AI-generated until you approve it. That approval is the human sign-off the EU AI Act Article 50 asks for, and because every working paper is inspectable, it is a real review rather than a rubber stamp. Approving the plan files it in Reports as a branded, shareable document.