Ads budgeting

Workflow: workflows/budget-review.md.

Every budget change shows this block (generated by adsgm plan budget)

CURRENT BUDGET / PROPOSED BUDGET / PERCENT CHANGE / DOLLAR CHANGE /
REASON / DATA SUPPORTING IT / EXPECTED EFFECT / RISKS / CONFIDENCE

adsgm plan budget --platform meta --account act_... --entity <adset or campaign id> --entity-label "<name>" \
  --current 50 --proposed 60 --currency USD --period daily \
  --reason "CPL $61 vs $90 target for 14 days, limited by budget" \
  --evidence "14d: $700 spend, 11 leads, CPL $63.6; 7d CPL $58; not in learning" \
  --expected "+2-3 leads/week at CPL up to ~$75" \
  --risks "marginal CPL rises with spend; may re-enter learning if >20%" \
  --confidence MODERATE --spend 700 --conversions 11

Approval is required by default. Only guarded_auto clients may skip the prompt, and only inside their limits (max increase/decrease %, max $ change/day, minimum spend and conversions in the window, minimum confidence, cooldown, high-budget threshold).

Planning a budget (setup and "what should I spend?")

Run adsgm budget-plan --client <slug> (or pass --monthly / --daily, --target-cpl or --job-value --close-rate --margin, --platforms meta,google, --google-cpc, --hard-cap, --flight-days). It returns the math in plain words:

• Budget type.
  • Always-on lead flow → daily budget.
  • "Never go over $X a month" → Meta lifetime budget per calendar month, plus an account spending limit the owner sets in Meta Billing; Google's daily budget already caps the month at 30.4× daily.
  • Fixed promotion → lifetime or flight budget.
• What it buys: an expected leads range at the target cost per lead, and whether Meta can learn (about 50 results a week per ad set) or should stay as one ad set.
• Split: with a small budget, start Google Search only and add Meta once roughly 5 leads a week of Meta money exists. Otherwise start about 60/40 Google/Meta and shift toward the cheaper qualified leads.
• Squeeze more out: if the business can afford a much higher cost per lead than the target, suggest +20–25% with the extra leads, jobs and revenue math. If the target is above what the economics support, say so and don't add budget.

Discuss, then --save the agreed plan into the profile. This changes no ad account. Live budgets still go through adsgm plan budget and approval.

How to decide

Increase when: CPA/CPL at or under target over ≥7 days with enough conversions, not in learning, and the entity is constrained (Meta: spending its full budget; Google: high lost IS (budget)). Expect marginal CPA to rise as spend rises - promise ranges, not points.

Decrease / reallocate when: CPA well above target with sufficient data, tracking verified, after fixing obvious causes (creative fatigue, search-term waste).

Do not touch budgets when: in learning, tracking broken, conversion lag not yet closed, inside a seasonal anomaly, or within the cooldown after the last change.

Platform mechanics

• Meta: budgets in minor units; ~20%+ jumps commonly reset learning (HEURISTIC); max 4 budget changes/hour/ad set (PLATFORM FACT). Advantage+ campaign budget: change at campaign level.
• Google: budgets in micros on the campaign budget resource (possibly shared across campaigns - check campaign_budget.explicitly_shared). Google may spend up to 2x the daily budget on a day while keeping the month near 30.4 x daily (PLATFORM FACT) - account for this in pacing checks.

Pacing

Month-to-date spend vs (monthly budget x days elapsed / days in month). Flag ±15% (HEURISTIC). Overspend spikes -> workflows/emergency-stop.md if extreme.