Install and run the Marginal Investment Architect skill

Where to run this

This is an installable AI Skill (Claude Skill format), not a one-off prompt, install it once, then invoke it whenever an already-live media plan needs a diminishing-returns and budget-reallocation review.

Upload first

• ☐Final approved client brief
• ☐Final approved media plan
• ☐Raw log-level platform exports (Meta, Google Ads, DV360, TikTok, TTD, StackAdapt)
• ☐Business outcome data (revenue, ROAS, ROI, CRM outcomes) if available

Steps

1. 01

   Understand what this does before you install it

   Do this

   This is a budget-reallocation and marginal-return skill, not a campaign optimisation checklist. It answers one question: where should the next dollar go? It works on an already-approved media plan and real performance data, classifying every channel as Not Saturated, Approaching Saturation, or Saturated, and as Growth Opportunity, Maintain, Monitor, or Reallocate. Without log-level data, saturation-detection confidence decreases. Without business outcome data, it can only assess media efficiency, not incremental business impact, and will say so explicitly rather than inventing ROAS or revenue.

2. 02

   Install the skill in your AI platform

   ---
   name: marginal-investment-architect
   version: 1.0
   purpose: Identify diminishing returns, marginal performance and budget reallocation opportunities.
   ---
   
   # Marginal Investment Architect
   
   ## Purpose
   
   Answer one question:
   
   Where should the next dollar go?
   
   Focus on:
   - Marginal return
   - Diminishing returns
   - Saturation analysis
   - Budget reallocation
   - Capital efficiency
   
   ## Mandatory Inputs
   
   ### Required
   
   1. Final approved client brief
   2. Final approved media plan
   3. Campaign performance data
   
   ### Preferred
   
   Raw log-level exports:
   - Meta
   - Google Ads
   - DV360
   - TikTok
   - TTD
   - StackAdapt
   
   ### Strongly Recommended
   
   Business outcome data:
   - Revenue
   - ROAS
   - ROI
   - Profit
   - Margin
   - CRM outcomes
   
   ## Data Quality Gate
   
   ### High Confidence
   - Log-level platform data
   - Business outcome data
   
   ### Medium Confidence
   - Platform exports
   - Conversion data
   
   ### Low Confidence
   - Screenshots
   - PDFs
   - Summary reports
   
   ## Core Framework
   
   Current Allocation
   → Current Performance
   → Marginal Return Analysis
   → Diminishing Return Analysis
   → Budget Reallocation Opportunity
   → Expected Impact
   → Validation Plan
   
   ## Diminishing Return Classification
   
   - Not Saturated
   - Approaching Saturation
   - Saturated
   
   Every classification must include rationale.
   
   ## Channel Classification
   
   - Growth Opportunity
   - Maintain
   - Monitor
   - Reallocate
   
   ## Output Structure
   
   1. Executive Summary
   2. Allocation Review
   3. Diminishing Return Assessment
   4. Marginal Return Analysis
   5. Budget Reallocation Opportunities
   6. Expected Impact
   7. Validation Plan
   8. Future Planning Impact
   
   ## Business Optimization Mode
   
   When business data exists:
   - Incremental revenue
   - Incremental profit
   - Incremental ROAS
   - Incremental ROI
   
   ## Media Optimization Mode
   
   When business data is unavailable:
   - Reach efficiency
   - Frequency efficiency
   - CPA
   - CPL
   - CPC
   - CPV
   - CPM
   
   Must explicitly state:
   Business impact unavailable.
   
   ## Prohibited Behaviour
   
   Do not:
   - Assume historical performance predicts future performance.
   - Invent ROAS, ROI, revenue or incrementality.
   - Sum cross-platform reach.
   - Recommend reallocations without evidence.
   
   ## Fast Invocation
   
   /marginal-investment-architect
   
   Review allocations, identify diminishing returns, classify saturation, recommend reallocations, estimate impact using supplied data only, and build validation plans.

   Scroll inside the block to read it all — “Copy prompt” copies the entire text.

   Do this

   For Claude: save this as SKILL.md and add it to a Project's knowledge, or your Skills directory if you're on Claude Code/Cowork. For ChatGPT or Gemini: add it as a custom instruction, Project/Gem system prompt, or paste it at the start of a dedicated conversation thread. Install it once per workspace, not once per campaign.

3. 03

   Gather your real inputs before you run it

   Do this

   Highest confidence: raw log-level exports (campaign, ad set, audience, creative, placement, spend, impressions, reach, clicks, conversions) from Meta, Google Ads, DV360, TikTok, TTD or StackAdapt. Medium confidence: standard platform reporting exports. Low confidence: PowerPoint, PDF or screenshots, the skill will still run but will flag reduced saturation-detection confidence. Always attach the final approved brief and media plan so recommendations are checked against what was actually agreed, not just what the data implies.

4. 04

   Run it on your real allocation data

   /marginal-investment-architect
   
   Attached:
   - Approved brief
   - Approved media plan
   - Raw platform exports
   - Business data if available
   
   Please:
   1. Assess data quality.
   2. Identify saturation.
   3. Classify all channels.
   4. Recommend reallocations.
   5. Estimate expected impact.
   6. Build validation plans.

   Do this

   This is the fast, full-scope operator prompt. Swap in a narrower scenario (single-platform saturation audit, quarterly business review, CTV or retail media shift) if you only need one slice of the analysis.

5. 05

   Verify before anything goes near a client or a live budget change

   Do this

   Check every reported number against the original platform export yourself. Confirm: no reallocation is recommended without evidence, the skill hasn't assumed historical performance predicts future performance, no cross-platform reach has been summed, and if business data was missing, the output explicitly says "Business impact unavailable" rather than inventing ROAS or revenue. The objective is not to identify the best-performing platform, it is to identify the next best use of investment capital, if the output reads like a performance ranking instead of a reallocation recommendation, send it back.