The build: agents shipped inside your stack
5 agents, built against the baselines the audit measured and the exception categories it named. Each one runs on top of the systems you already own, through service accounts scoped to the actions it is allowed to take.
| # | Agent | What it does | Hrs/week | Annual value | Ratio |
|---|---|---|---|---|---|
| 01 | Campaign Builder | Turns an approved brief into the assets, audiences and settings each platform needs. | 16 | $118k | 3:1 |
| 02 | Audience Activation | Builds segments from the warehouse definition and keeps them synced across platforms. | 11 | $92k | 4:1 |
| 03 | Content Repurposer | Adapts an approved asset into channel formats without changing the claim. | 10 | $74k | 3:1 |
| 04 | Signal QA | Checks that conversion events, UTMs and pixels still fire the way the plan assumed. | 6 | $48k | 4:1 |
| 05 | Attribution Reporter | Assembles channel performance from the systems of record rather than from platform self-reports. | 5 | $36k | 3:1 |
Ratio is automated steps to human steps within that agent's flow. The 5 agents return 48 hours a week between them, which is the annual figure in chapter 02 divided across a 46-week working year.
Campaign Builder
- 01Read the approved brief and pull the product and pricing facts it referencesAutomated
- 02Check the claims against the approved messaging libraryAutomated
- 03Build the audience from the warehouse definition, not a platform-side rebuildAutomated
- 04Campaign manager reviews copy and targetingReviewer input
Edits persist into the messaging library for later campaigns
- 05Assemble creative variants and the tracking parameters for each channelAutomated
- 06CMO approves any campaign over $40,000 of committed spendHuman gate
- 07Push to the platforms and verify the conversion events fireAutomated
- 08Log the launch, the approver and the claims usedAutomated
The gate is a step, not a promise made in prose. Where a correction appears it is marked, because the correction persisting as training signal is the difference between an agent that improves and one that repeats itself.
12 steps → 2 touchpoints
What took 12 manual steps now needs a person at 2 of them.
5 → 1 queue
5 handoffs between people and systems collapse into one queue with the evidence already attached.
- Cost reduction$87k24%
- Revenue growth$211k57%
- Risk mitigation$70k19%
The agent roster above sums to the same total: $368k across 5 agents. Cost reduction is hours at a fully loaded rate, discounted 25% for ramp.
These agents are specified, not running. The software is built for the client who commissions it, after their audit has named the thresholds and the exception categories it has to work against. What is fixed is the pattern: 1 human gate in this flow, every action logged with its inputs and its approver.