--- name: monitor-and-maintain-bid-strategy-health description: >- The recurring six-part checkup for every active bid strategy in the account: learning status (hold flags on anything still calibrating), conversion-volume sufficiency against per-strategy thresholds, target-vs-actual variance scoring with a root-cause diagnostic, gradual target corrections (10-20% max per move), migration-readiness up the automation ladder, and portfolio hygiene (grouping, forgotten CPC caps, stale bid adjustments). Run it monthly on stable accounts, bi-weekly on active ones, and within 48 hours of any major change. It is the maintenance loop, not initial selection (select-a-bidding-strategy), not target derivation (calculate-bid-targets), not deliberate scaling (scale-bids-and-budgets), and not the dedicated budget-cap rescue (handle-budget-limited-campaigns). --- # Bid Strategy Health Check and Maintenance ## Purpose Bid strategies degrade silently: volume drifts below what the algorithm needs, a target set last quarter loses touch with breakeven, a CPC cap from an old emergency quietly throttles bids, and leftover manual-era adjustments clutter campaigns that ignore them. This skill sweeps every active strategy on a schedule, catches those regressions early, and applies only gradual, documented corrections. ## When to run - Scheduled cadence: monthly (stable accounts) or bi-weekly (high spend / recent changes). - Within 48 hours after any major change: strategy switch, budget move >30%, targeting overhaul, conversion-action edits. - Immediately on a detected performance decline. - First week after onboarding an account; at each seasonal transition; within 3 days after a learning period completes. ## When NOT to run - Choosing a first strategy -> select-a-bidding-strategy. - Deriving targets from economics -> calculate-bid-targets. - Healthy strategy, deliberate growth push -> scale-bids-and-budgets. - Single campaign pinned by its budget cap -> handle-budget-limited-campaigns (this skill only flags the condition). - Broken conversion tracking -> set-up-google-ads-conversion-tracking; no bidding maintenance is meaningful on bad signal. ## Prerequisites - Breakeven CPA/ROAS/profit-per-spend per strategy (from calculate-bid-targets). - Known conversion lag; last 30 days of data beyond it. - The change log from previous health checks (what was adjusted, when, why). ## Procedure 1. **Session guardrail.** Call `gads_policy_guardrail` before the session's first write. 2. **Inventory + learning status.** Pull all strategies via `gads_list_bidding_strategies` (portfolios) and `gads_get_campaign` / `gads_run_gaql_query` on `campaign.bidding_strategy_system_status` (campaign-level). Classify each: fully operational; learning (recent change, HOLD, no changes of any kind until ~two conversion cycles pass); learning-limited (cannot exit, diagnose: volume too thin -> step 3, budget too small -> budget fix, tracking broken -> fix tracking, target too aggressive -> flag for step 5); budget-limited (note for step 4); misconfigured (fix immediately, this is the one exception to the learning HOLD). 3. **Volume sufficiency.** For each non-held strategy, pull 30 days of conversions (window extended by conversion lag, trailing lag days excluded) via `gads_run_gaql_query`. Judge against the Decision-rules thresholds (portfolios judge the pooled total). For shortfalls, in priority order: verify tracking (`gads_list_conversion_actions` + GAQL by conversion action); consolidate 2-3 same-goal campaigns via `gads_create_portfolio_bidding_strategy` (previewed, approved); temporarily drop the target (tCPA -> Maximize Conversions, tROAS -> Maximize Conversion Value) via `gads_set_campaign_bidding_strategy`; consider optimizing on a higher-volume upstream event while data builds; or raise budget in ~20% steps via `gads_update_budget`. Every write: preview -> user approval -> apply. 4. **Variance scorecard.** For each healthy-volume strategy, compute (actual - target)/target from `gads_get_account_performance` / GAQL. Classify per Decision rules. For underperformers, walk the diagnostic in order and stop at the first hit: recent structural change still settling; junk queries or new competitors eating spend; market contraction; over-broad negatives blocking good queries; evaluation window shorter than the conversion cycle; short-term seasonality; conversion-rate drop on the landing page/offer (not a bidding problem). Only if none apply is the target itself the suspect. Budget-limited entries: IS lost to budget 10-25% -> raise budget if growth matters; >25% -> raise budget or, if capped, set the target 5-10% looser than recent actuals to recapture volume within spend (a tighter target under a binding cap buys fewer, pricier auctions). 5. **Gradual target corrections (write).** For flagged strategies, adjust via `gads_update_campaign` / `gads_set_campaign_bidding_strategy` (portfolio targets via the portfolio resource). Loosening: max 15-20% per move. Tightening: max 10-15%. Anything needing >20% total gets split across multiple moves, one conversion cycle apart. Preview each change with old/new/reason; apply after approval; log to the change record with a next-review date. 6. **Migration readiness.** Score each strategy against the automation progression: manual -> Maximize Conversions once ~30 conversions have accumulated and tracking is clean; add a CPA target once 30+/month is sustained for 2 months with stable CPA and known breakeven; move to tROAS once 50+/month carry genuinely varied values; move to profit-based values once profit tracking is live and validated. Ecommerce naturally climbs to value/profit bidding; lead gen usually ends at tCPA unless imported deal values vary meaningfully. Execute migrations through migrate-from-manual-to-smart-bidding or set-up-value-based-bidding (experiment-first), not inline here. 7. **Portfolio hygiene.** For each portfolio (via `gads_list_bidding_strategies` + GAQL): members all one campaign type and one efficiency goal; no paused/ended stragglers attached; no member contributing <5% of pooled conversions as dead weight. CPC caps: expected OFF; if ON, the cap must sit at >=3x the average CPC of top converting terms, top-keyword CPCs must not plateau at the cap, IS lost to rank must not have climbed since it was set, otherwise raise or remove it (previewed, approved). Any minimum-CPC floor: remove. Shared budgets: no member consuming >60% disproportionately; daily budget >=10x target CPA for CPA portfolios. 8. **Bid-adjustment cleanup (write).** Smart Bidding ignores every percentage modifier except device -100%. Query existing adjustments via GAQL; zero out stale non-exclusion device / location / schedule / audience / demographic modifiers on Smart Bidding campaigns with `gads_set_bid_adjustments` (keep deliberate device -100% exclusions). Preview the cleanup list, apply after approval, these leftovers do nothing but mislead future readers of the account. 9. **Close out.** Update the change log (date, entity, old, new, reason, next review). Schedule the next check per cadence. 1-2 days after any target change: spot-check spend/volume via `gads_get_campaign`; one conversion cycle later: evaluate against the new target. ## Decision rules - **Volume thresholds (30-day, per strategy):** Maximize Conversions / tCPA, 15 absolute floor, 30 functional, 50+ comfortable. Maximize Conversion Value, 15/50/50+. tROAS, 30 absolute, 50 functional/recommended, values must vary (identical values -> tROAS adds nothing over tCPA; use tCPA). - **Variance bands:** within +/-10% -> healthy, hands off. 10-25% worse -> diagnose. >25% worse -> diagnose, escalated. 10-25% better -> check whether volume is being sacrificed. >25% better -> target likely over-tightened; loosen. - **Adjustment limits:** loosen <=15-20%, tighten <=10-15%, one conversion cycle between moves; >20% in one move restarts learning. - **Learning HOLD:** no target/budget/structural changes on strategies in learning; sole exception is fixing a genuine misconfiguration. - **Seasonality:** for known 1-7 day conversion-rate events use a seasonality signal / set-up-data-exclusions route; for multi-week seasonal shifts adjust targets directly (loosen 5-10% pre-peak, tighten back after). - **Offline-import accounts:** thresholds apply to the imported action, and import lag (often 7-30 days) extends every evaluation window. ## Common failure modes - **Editing a learning strategy.** Resets the clock, extends volatility, and invites the next panic edit. Hold flags exist to be honored. - **Lag-blind evaluation.** Scoring the last 7 days with a 14-day conversion cycle manufactures fake underperformance; trim the window every time. - **The forgotten CPC cap.** Set in an emergency, correct for a month, wrong for a year. Check every cap, every cycle. - **Treating demand problems as bidding problems.** Seasonal decline or a landing-page regression will not be fixed by target surgery; the step 4 diagnostic runs before any adjustment. - **Undocumented changes.** Without the change log, the next session re-adjusts the same target and compounds the move past the 20% learning-reset line. ## Related skills - Upstream: select-a-bidding-strategy, calculate-bid-targets, set-up-conversion-based-bidding / set-up-value-based-bidding (the setups this loop maintains). - Downstream on findings: scale-bids-and-budgets (healthy + headroom), handle-budget-limited-campaigns (cap-bound), set-up-portfolio-bid-strategies (consolidation), migrate-from-manual-to-smart-bidding (ladder moves), set-up-google-ads-conversion-tracking (signal repairs), allocate-budget-across-campaigns (structural budget rethink).