budget-optimizer Checked

This skill also allocates **paid-ads** spend — the tier/platform tables map to channels/campaigns; use the ROAS profile (direct-response|prospecting|incremental-profit) as the scenario axis and read its declared CPA/payback/contribution constraint instead of substituting CPM/CPE. Scope: this computes the spend-reallocation **plan** only. It does **not** read in-flight pacing or issue scale-up/down moves — the live pacing read (pacing vs plan, learning-phase respect) belongs to budget-pacing-monitor (../../../ad/scale/budget-pacing-monitor/SKILL.md), and bid-strategy choice belongs to bid-strategy-planner (../../../ad/orchestrate/bid-strategy-planner/SKILL.md). paid-measurement-loop (../../../ad/scale/paid-measurement-loop/SKILL.md) reads one shipped change back against a control, and premature scaling is an **S guardrail flag** in ad-account-auditor (../../../ad/activate/ad-account-auditor/SKILL.md), not a separate skill or a veto. Save paid runs under memory/ad/budget-optimizer/.

Data Sources

This family has no required live integration, but numerical rate, multiplier, reach, engagement, revenue, and return projections require a compatible anchor: a user-provided quote or assumption, source-dated market evidence, or comparable first-party history. With only a total budget, platforms, and goal, return an allocation skeleton plus the exact rate-card/forecast queries needed and status NEEDS_INPUT; do not invent a repository rate table or multiplier.

Optional connectors that sharpen the estimates when present:

• ~~influencer database — real rate cards instead of benchmark ranges.
• ~~social platform analytics — actual reach, CPM, and engagement to replace estimated projections.
• ~~CRM — past campaign spend and conversion data for ROI calibration.

Classify evidence **field by field**, not by connector provenance. Mark a connector value **Measured** only when it is an observed result for that exact metric, entity, and campaign window (record the source and as-of/window); mark numbers the user states **User-provided**; and mark benchmarks, forecasts, inferred values, and modeled downstream projections **Estimated**. A connector-backed input never upgrades an entire row or a projection derived from assumptions to Measured. See CONNECTORS.md (../../../CONNECTORS.md) for the keyless data recipes.

Instructions

When a user requests budget optimization, work these steps. Each step's fill-in template and scenario block lives in references/templates.md (references/templates.md) — copy the matching section and populate it.

1. **Gather budget parameters** — campaign goal, audience, timeline, total budget, fixed vs influencer-available split, platform priorities, and constraints (max per influencer, min count). Intake template: §Step 1 (references/templates.md#step-1--budget-parameters-intake-template).

2. **Qualify cost evidence** — inventory source-dated, market/platform/tier/deliverable/rights-compatible quotes or first-party history. A rate or multiplier without a compatible source and date stays NEEDS_INPUT; never supply a built-in default. Template: §Step 2 (references/templates.md#step-2--cost-evidence).

3. **Create the allocation** — split across tier, platform, content type, and other items (gifting, amplification, tools, contingency); sum to 100%. Use a contingency only when the user or a dated planning source supplied the rule. Template: §Step 3 (references/templates.md#step-3--budget-allocation-recommendation).

4. **Project returns** — record each formula and input ref. Keep EMV as a separate media-equivalency scenario; never add it to attributed revenue or count it as cash return. Report ROAS = attributed revenue / spend as x:1; report arithmetic ROI as (attributed revenue - spend) / spend × 100%; do not claim profitability or incrementality without the required evidence. Template: §Step 4 (references/templates.md#step-4--return-projections).

5. **Model scenarios** — compare only scenarios supported by declared inputs; otherwise return named scenario shells and the missing-input plan. Template: §Step 5 (references/templates.md#step-5--budget-scenarios).

6. **Optimization strategies** — label savings, concentration, and amplification thresholds as user rules or source-dated assumptions. Do not present unsourced percentages as standard. Detail: §Step 6 (references/templates.md#step-6--optimization-strategies).

7. **Mid-campaign reallocation** — require comparable observed results plus a preregistered readback window and decision rule. Recommend KEEP_TESTING or NEEDS_INPUT when those are absent; a recommendation requires separate action authorization before spend changes. Persist and hand off creators only as creator_ref. Template: §Step 7 (references/templates.md#step-7--mid-campaign-reallocation).

Return the run inline. Offer memory/influencer/budget-optimizer/YYYY-MM-DD-<topic>.md (or memory/ad/budget-optimizer/ for paid-ads runs) for exact WARM-save authorization, and request a separate authorization before promoting the approved total, chosen scenario, or locked tier mix to memory/hot-cache.md.

Example

**User**: "Optimize a $30,000 budget for a skincare product launch on Instagram and TikTok targeting Gen Z"

**Output**:

## Budget Allocation: $30,000 Skincare Launch

**Status**: NEEDS_INPUT

The budget, platforms, audience, and goal are User-provided. No compatible creator quotes, deliverable/rights scope, contingency rule, reach/engagement history, conversion rate, AOV, or attribution basis was supplied, so no creator count, platform split, CPM, revenue, ROAS, ROI, or EMV is fabricated.

### Evidence request

| Needed input | Exact request |
|--------------|---------------|
| Cost basis | Dated creator quotes or first-party rates for the tar