Ad Scale Scatter: plot every Meta ad on spend against CPA

The formula

For each ad, expected conversions = spend divided by your CPA target. Conversion counts are treated as Poisson-distributed, so the standard error is the square root of that expected count. The z-score is actual conversions minus expected, divided by the standard error.

An ad is KILL when its conversions fall below expected minus z times the standard error, SCALE when they rise above expected plus the same margin, and WAIT otherwise. The confidence setting picks the z threshold: 1.28 (80%, loose), 1.645 (90%, the default), or 1.96 (95%, strict). CPA is spend divided by conversions; ROAS is revenue divided by spend when a revenue column is mapped. The point of the significance test: a $200 ad with zero conversions is noise, while a $20,000 ad 20% over target is a real decision.

Expected input

Upload an ads export CSV with a header row (up to 500,000 rows). Columns are auto-detected case-insensitively and can be remapped by hand. Three are required:

• –Ad name: accepts ad_name, ad name, creative, or ad.
• –Spend: accepts amount_spent, amount spent, spend, cost, or budget.
• –Conversions: accepts purchases, website_purchases, omni_purchases, conversions, results, customers, orders, subscriptions, signups, registrations, or leads.
• –Optional: campaign_name and adset_name (keyword filters and placement drill-down), date as YYYY-MM-DD (timeline scrubbing and daily CPA trends), and revenue such as purchases_conversion_value (enables ROAS).

Assumptions and limits

• –Everything runs in the browser. The CSV is never uploaded to a server.
• –Ads are aggregated by ad name, so identical names across campaigns merge into one dot.
• –Ads with zero conversions are counted but not plotted, and ads below your minimum-spend filter are excluded from classification.
• –On first load the CPA target auto-sets to the account's spend-weighted CPA; adjust it to your actual allowable CPA before trusting the verdicts.