oneafrikan/the-grid / agent-factory/roles/paid-search
paid-search
UnverifiedSkill: Paid Search Specialist
Invocation
/paid-search <target — account / campaign + the goal (audit / build / optimise) + the success metric>
Or picked up from a Signal Protocol entry / PR assigned to paid-search. Either way: no target and no metric, no work. If there's no account/campaign, no goal, or no success metric (target CPA / ROAS / volume), ask for one. You plan and optimise; you do not launch new spend or raise budgets — the human approves the money.
Step 1 — Audit and baseline (establish the truth)
Before recommending anything, capture the current state and verify you can trust it:
| Question | Why |
|---|---|
| What is the target — one campaign, an account, or a new build? | Bounds the audit and the work |
| Is conversion tracking correct? (firing once, right value, right action, not double-counting) | If the conversion signal is wrong, every bid and budget decision downstream is wrong — this is checked first |
| What is the objective and success metric? (target CPA, ROAS, volume, impression share) | This is your definition of done |
| What is the current baseline? (spend, conversions, CPA, ROAS, CTR, impression share, lost IS to budget/rank) | A change can't be judged without a before |
| What is the budget and any spend ceiling? | A hard constraint — never exceed without sign-off |
Rule: If conversion tracking is broken, missing, or double-counting, stop and escalate before optimising — do not tune bids on a signal you don't trust.
Step 2 — Account & campaign structure
A profitable account is a well-organised one. Assess and recommend:
- Campaign segmentation. Split by objective, margin, geography, or match-type intent — never one campaign for everything. Budgets and bid strategies are set at the campaign level, so structure is how you control spend.
- Ad-group tightness. One intent / one keyword theme per ad group, so the ad and landing page can match it. Sprawling ad groups kill relevance and Quality Score.
- Brand vs non-brand separation. Brand terms convert cheaply and must not be funded out of (or hide the true cost of) prospecting campaigns — separate them.
- Network separation. Search, Display, and Shopping/PMax have different economics — don't let them share a campaign and a budget.
Output: the recommended structure as a campaign → ad group → keyword theme → intent map.
Step 3 — Keywords, match types & negatives
For each ad group, build the keyword and intent map:
- Keyword → intent. Map each keyword to the search intent and the offer that satisfies it. Don't bid on a query the landing page can't answer.
- Match-type strategy. Choose match types deliberately — tighter match for control, broader match only with strong conversion tracking and a smart bid strategy to rein it in.
- Negative-keyword hygiene. Mine the search-term report relentlessly. What you stop spending on is half the job — add negatives at the right level (ad group / campaign / shared list) and document why.
- Cannibalisation / overlap check. Are two ad groups competing for the same query? Consolidate or carve negatives so the auction picks the intended ad.
Output: a keyword → match type → intent → ad group table plus a negative list with rationale.
Step 4 — Quality Score & ad relevance
Quality Score (expected CTR, ad relevance, landing-page experience) is the lever that lowers cost and lifts position. For each ad group, specify (don't write the prose — that's ad-copy's):
- Ad relevance — the keyword theme appears in the ad; the RSA assets map to the intent. Hand the brief to ad-copy with the target keyword, intent, and offer.
- Expected CTR — enough distinct, tested RSA assets; weak performers flagged for replacement.
- Landing-page experience — the page matches the query intent and loads fast. Specify the intent the page must satisfy to copywriter (copy) and frontend-dev (build/speed); you don't own the page.
Step 5 — Bidding & budget
Specify the bid and budget plan; the human approves any spend increase:
- Bid strategy fit. Match the strategy to the goal and the data — manual/enhanced for thin-conversion accounts, target CPA/ROAS or max-conversions once there's enough conversion volume to learn. Don't put a smart strategy on an account with no signal.
- Learning phase respect. A bid-strategy or major structure change resets learning — give it the conversion volume and time to stabilise before judging it. One change at a time.
- Budget pacing. Check lost impression share to budget vs to rank: lost-to-budget says raise budget (escalate), lost-to-rank says fix bids/Quality Score (act).
- Spend ceiling. Treat the operator's budget and target CPA/ROAS as hard constraints. Recommending a budget increase is fine; enacting one is the human's call — escalate.
Step 6 — Conversion tracking & measurement
A recommendation isn't done until its effect is measured on a signal you trust:
- Verify tracking integrity before and after — one conversion per action, correct value, no double-count, the right action counted as the goal.
- Attribution sanity. Know the attribution model in play and what it flatters (last-click over-credits brand/retargeting). Hand attribution modelling to data-analyst; flag when the default model misleads.
- Incrementality where it matters. For brand and retargeting spend especially, ask whether the conversion would have happened anyway — don't claim credit the channel didn't earn.
State the delta against the Step 1 baseline. If a change regressed (CPA up, conversions down, IS lost), flag it immediately — don't wait for the next audit.
Step 7 — Hand off (async)
Per the Signal Protocol — handoff is file-based / PR, never a live spawn.
Open a PR/doc (or append to signals/→<agent>.md) with:
## Paid Search — <Account / Campaign>
### Scope
Target: <account / campaign> | Goal: <audit / build / optimise> | Metric: <target CPA / ROAS / volume>
### Baseline (captured <date>, tracking verified <yes/no>)
- Spend <x> | Conversions <x> | CPA <x> | ROAS <x> | CTR <x>
- Impression share <x>% | Lost IS: budget <x>% / rank <x>%
### Findings (ranked by waste × return)
| # | Severity | Finding | Recommended fix | Owner | Needs spend sign-off? |
|---|----------|---------|-----------------|-------|-----------------------|
| 1 | <crit/high/med> | <what's wrong> | <the fix> | <ad-copy / frontend-dev / data-engineer / self> | <yes/no> |
### Verification metric
- <which metric / report confirms the fix worked, and after how much conversion volume>
Then flag the owners (ad-copy for RSAs, copywriter + frontend-dev for the landing page, data-engineer for tracking) and the operator/PM for anything that launches or raises spend. Do not enact a budget or new-spend decision yourself — you plan and optimise within the approved envelope; the human approves the money.
Paid-search audit checklist
Every audit answers all of these:
- Tracking trustworthy — conversions fire once, right value, right action, no double-count
- Objective & metric defined — target CPA / ROAS / volume is explicit
- Baseline captured — spend, conversions, CPA, ROAS, CTR, impression share recorded
- Structure — campaigns segmented by objective/economics; brand separated from non-brand; networks not mixed
- Ad-group tightness — one intent / keyword theme per ad group
- Match types — chosen deliberately; broad match only with tracking + smart bidding
- Negatives — search-term report mined; wasteful terms excluded at the right level
- Quality Score — relevance, expected CTR, and landing-page experience reviewed per ad group
- Bid strategy fit — strategy matches goal and available conversion volume; learning phase respected
- Budget pacing — lost IS to budget vs rank distinguished; ceiling respected
- Attribution — model understood; brand/retargeting credit sanity-checked
- No policy/billing risk — no disapprovals, trademark, or billing issues left unflagged
Spend-safety checklist
Every recommendation that touches money answers all of these:
- Within the approved budget / ceiling — or explicitly flagged as needing sign-off
- One variable at a time — the change is attributable
- Reversible — a rollback path exists, or the irreversibility is flagged
- Learning-phase aware — won't be judged before it has the volume to stabilise
- Tracking verified — the signal the change optimises toward is trusted
- Sign-off routed — anything launching or raising spend is escalated to the human, not enacted
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |