fv-cut-wasted-spend

fv-cut-wasted-spend

The job a marketer already runs every Monday. The difference here: zero payers, not zero conversions. A term with 40 form fills and no Stripe charge is the most expensive kind of waste, and it is invisible to the ad platform's own reporting.

Read shared/reading-fullvision-data.md, shared/safety-rails.md and shared/sparse-data.md before calling anything. All three are binding. For the platform whose spend you are judging, also read shared/platforms/google.md, shared/platforms/meta.md or shared/platforms/linkedin.md — the attribution windows are not comparable and the write capabilities differ.

Steps

1. Precondition: run fv-data-health. On 🚩, abort. On ⚠️, raise every minimum-n threshold below by 50% and say so in the report.
2. Establish the measurable window. fullvision:query_view on view:ads-measurement-start. Everything before that date is unmeasurable, not wasteful. All spend figures below are within the measurable window only.
3. Pull term-level performance. fullvision:query_view on view:keyword-performance for the trailing 90 days, and view:ads-leaderboard for account context (use the clipped_* columns). Aggregate in SQL, never client-side.
4. Apply the thresholds below to produce a candidate list.
5. Corroborate. Per shared/sparse-data.md §4, zero payers alone is not enough at low volume. Require a second signal per candidate: no assisted conversions in any attribution model (view:ltv-by-campaign), or engaged-session rate far below the account baseline (view:ad-landing-pages). Drop candidates that fail corroboration into "Not proposed".
6. Emit the change-list, then STOP. Do not write in this turn. Ever. Two turns, always (shared/safety-rails.md §1).
7. On confirmation: write the change log entry first, then add the negative keywords.

Thresholds — fixed, never runtime-adjusted

• Term spend ≥ €150 over the trailing 90 days (within the measurable window)
• Term clicks ≥ 60 — below this, zero payers is expected even for a good term
• Payers attributed = 0
• Window = trailing 90 days, matching Google's click-age window
• Confidence stated at 85%

Blast radius

• Max 25 negative keywords per run (overrides the default 10-entity cap — a negative keyword is the most reversible action available).
• Max 15% of trailing-30d spend affected.
• Never propose a campaign pause here. That is fv-kill-losing-campaigns (v2).

Write path — Google Ads is read-only

The official Google Ads MCP server exposes GAQL reads only. Every run of this skill is read-only mode per shared/safety-rails.md §9: emit the change-list as an artifact with copy-pasteable negative keyword lists grouped by ad group, plus the exact Ads UI path to apply them. Report this as the normal outcome. Do not present it as a failure, and do not suggest installing an unvetted community write server.

Output

shared/report-format.md. Each proposed negative keyword line carries: term, spend, clicks, payers, attributed revenue, ad group, and the corroborating signal.

Refuse when

• fv-data-health returns 🚩.
• The measurable window is shorter than 90 days — say how long it is and re-run later.
• Fewer than 5 terms clear the thresholds. At that point this is manual work, not a sweep, and the report should say so rather than propose a token list.