Account Structure Health Check

Variables

Replace each placeholder before you run the prompt.

• {{mcc_summary}}Number of accounts, total spend, vertical mix, agency size.

  47 accounts under one MCC, $1.8M/mo total, mix of B2B SaaS + industrial + B2B services, 6-person agency team.

• {{structure_dump}}A representative campaign-level export across 3–5 accounts (campaign name, type, status, daily budget).

  See attached CSV; sample includes 3 industrial accounts and 2 SaaS accounts.

You are reviewing the structural health of an MCC portfolio for an experienced agency lead. Your audience is sophisticated; do not explain basics.

PORTFOLIO SUMMARY:
{{mcc_summary}}

STRUCTURE DUMP:
{{structure_dump}}

Identify:
1. Structural inconsistencies between accounts of similar type (where one account does X, another doesn't, both should)
2. Naming convention drift
3. Campaign-type misuse (e.g. PMax used where Search would be cleaner; Display where in-market would be cleaner)
4. Cross-account learning that isn't happening but could (one account's negative list could help others)
5. Risk concentrations (any one account or campaign too dependent on a single signal)

Output a prioritized punch list with concrete agency-level work items, not per-account changes. Each item: 2–3 sentences max, ending with the specific action.

Do NOT recommend "consolidate accounts" or "deconsolidate accounts" unless the data clearly supports it. Stay structural.

Expected output shape

A prioritized agency-level punch list with concrete work items. Typically 10–20 items.

Why we wrote it

Most agency leadership reviews focus on per-account performance. The structural health of the portfolio itself is the bigger lever, and almost nobody systematizes it.

How to use

1. Open Claude or ChatGPT. The recommended model for this prompt is claude-opus-4-7 — opus when the prompt requires deep reasoning, sonnet for the rest.
2. Replace every {{variable}} with content specific to your account. The examples above are starting points, not templates to ship as-is.
3. Paste the prompt and run.
4. Read the output against the expected shape above. If the model produced a structurally different response, re-prompt rather than accept the drift.