Lead drivers and duds

Split every ad in scope into 4 buckets (drivers, promising, non-converters, too early) and say what to do with each. The pause list with dollar amounts is usually the most valuable output.

How to reach Adwarp#

Prefer the Adwarp MCP tools if they are loaded (get_performance, list_ads, list_campaigns, list_conversions). Otherwise the CLI: adwarp report perf creative, adwarp ad list <campaignId>, etc., or npx -y adwarp. Read-only, except the explicitly-confirmed recreate flow at the end.

What to pull#

1. get_performance at creative level, default last_30_days. If total conversions across the account are under ~10, widen to last_90_days and say so.
2. list_campaigns and list_ads to join creative ids to ad names, statuses, and parent campaigns. Well-named ads (persona, competitor, offer in the name) let you analyze angle patterns; if names are opaque ids, say the analysis is per-ad only.
3. list_conversions to name which conversion is being counted. State it in the report; "leads" means nothing without the definition.

How to bucket#

Compute the account average cost per conversion first (total spend / total conversions on converting entities).

• Drivers: 3+ conversions and cost per conversion at or below account average. Rank by conversions, then cost per conversion.
• Promising: 1-2 conversions at good efficiency, or strong CTR with spend still below the judgment floor. Watch, do not scale yet.
• Non-converters: zero conversions with spend at or above ~2x the account average cost per conversion. This is the pause list. Sum the wasted spend into 1 figure.
• Too early: under ~1,000 impressions or trivial spend. No verdict.

Caveats to state#

• Conversion counting lags clicks; treat the last 2-3 days as incomplete.
• LinkedIn mixes post-click and view-through conversions depending on setup; a retargeting ad can look like a driver while merely tailgating other touches. Flag suspiciously high conversion rates on tiny click counts.
• 1 conversion is an anecdote. Never crown a winner on fewer than 3.

Angle patterns#

When ad names or copy encode the angle (persona, competitor, pain, offer), aggregate buckets by angle and report which angles convert and which never do. This is often more actionable than the per-ad list, because it tells the user what to make next, not just what to kill.

Report format#

• Headline: conversion total, account average cost per conversion, and the wasted spend figure from the pause list.
• Drivers table: name, campaign, conversions, cost per conversion, CTR.
• Pause list: name, campaign, spend burned, impressions. 1 line each on why.
• Angle read: converting angles vs dead angles, if names permit.
• Actions: scale/shift suggestions plus refresh candidates. On LinkedIn, changing copy means recreate, not edit (delete_ad + create_image_ad, drafts only). Only run deletes if the user explicitly confirms the exact ads; pausing and activating happen only on the user's explicit request, through Adwarp (update_ad with status: PAUSED pauses 1 ad and leaves its ad group running) or in Campaign Manager.
• Afterwards, suggest journaling any change made (log_ad_change) so compute_lift can measure it later.