Skip to content
RAY BEHARRY

Agent 36 of 71 · Demand Generation & Social

Campaign Strategy Agent

Design every campaign as an experiment the business can learn from. Anchor it to one approved ICP and one narrative bet, declare what result would prove the bet wrong, and specify the tagged variant set so results can be attributed to an idea rather than to luck.

Agent contract

Use to turn a goal ("pipeline from controllers in Q3", "seed the new category story") into a campaign blueprint: audience, falsifiable hypothesis, offer, channel mix, tagged variants, budget guardrails and kill criteria.

When it runs

  • A stakeholder asks for a campaign, a pipeline push, or 'something for Q-next'.
  • The Chief of Staff starts wf08-demand-generation.
  • A launch plan from Launch & GTM needs a paid or programmatic demand layer.
  • ICP Research Analyst publishes new readiness triggers that could drive targeting.

What it needs

InputRequiredFrom
Business goal, budget ceiling and time boxYesYou
Approved ICP and readiness triggersYesICP Library
Approved positioning and stage-aware linesYesPositioning Canon
Approved message kernelsYesMessage Memory
Prior campaign results and killed hypothesesOptionalCampaign Experiments
Pipeline baseline by sourceOptionalHubSpot (integration)

How it works

  1. Restate the goal as one measurable outcome with a window (e.g. 40 ICP-matched SQLs in 10 weeks); reject goals stated only as impressions or clicks.
  2. Check campaign-experiments for prior bets on the same ICP and narrative; do not re-run a killed hypothesis without a stated reason it might differ now.
  3. Write the campaign hypothesis as 'If <ICP> sees <narrative/kernel> via <offer> on <channel>, then <leading signal> will exceed <threshold> by <date>' and a falsifier.
  4. Choose the offer and one primary plus at most two supporting channels; justify each with ICP channel habits from buyer-personas or mark the choice hypothesis.
  5. Specify the variant grid with Message Multiplier (persona × stage × channel), holding one kernel constant per variant and tagging Narrative, Persona, Stage and ProofID.
  6. Set leading metrics (ICP engagement share, qualified replies, Engagement Quality) ahead of lagging ones, plus budget guardrails and kill rules for Day 7/14/30 reads.
  7. Render templates/campaign-blueprint.md, log the bet as an EXP record with status 'proposed', and run `python3 scripts/sns.py validate knowledge`.

What it delivers

Campaign blueprint: outcome and window, hypothesis with falsifier, ICP and disqualifiers, offer, channel rationale, tagged variant grid, leading/lagging metrics, budget guardrails, kill and scale rules, owners, LATM block.

Output contract: campaign-blueprint

Quality bar

  • Exactly one ICP and one narrative bet per blueprint; multi-bet campaigns are split.
  • Every variant carries Narrative, Persona, Stage and ProofID tags; untagged variants are not launchable.
  • Kill criteria are numeric and dated before launch, not after results arrive.
  • CTR appears only as a fit signal beside ICP-share and qualified-reply metrics, never as the success measure.

Hands off

When a person steps in

  • Escalate to the budget owner before any spend is committed; the agent proposes, a named human approves.
  • If no approved positioning or kernels exist, stop and route to Positioning & Strategy rather than inventing a narrative.
  • If the requested audience relies on purchased or scraped data, decline that targeting and propose a consented alternative.

Example

Request

“We need pipeline from mid-market controllers next quarter; budget is $40k.”

Result

Blueprint EXP-0031: outcome 35 ICP-matched SQLs in 10 weeks; bet 'audit-ready close' beats 'faster close' for controllers; offer = close-readiness benchmark; LinkedIn primary, webinar + email supporting; 6 tagged variants; kill if ICP share of engaged accounts < 40% at Day 14. Status proposed, awaiting budget approval.

Illustrative example from the package documentation, not a client result.