--- name: plan-and-execute-seasonal-adjustments description: >- End-to-end management of a Google Ads account through a dated peak event (Black Friday, Christmas, an industry sale): builds the year-over-year baseline with gads_run_gaql_query, stages seasonal keywords, promotion/sitelink/callout assets ahead of review deadlines, raises budgets and loosens bid targets at event start (gads_update_budget, gads_set_campaign_bidding_strategy), monitors live capture, and reverts everything within three days of the event ending. Start it 6-8 weeks out. Do not use it for the retrospective and post-event query cleanup one to two weeks later (run-post-peak-season-normalization), for non-seasonal growth pushes (scale-bids-and-budgets), or for everyday budget rebalancing (allocate-budget-across-campaigns). --- # Prepare, Run, and Revert a Seasonal Peak ## Purpose Peak events compress a quarter's opportunity into days, and the two classic ways to lose it are being unprepared (assets stuck in review, budgets capped on the big day) and being sloppy afterwards (inflated budgets and loosened targets left running into normal demand). This skill manages the full arc, baseline, build, execute, revert, with every account change staged ahead of deadlines and unwound on a fixed timer, so automated bidding stays stable on both sides of the peak. ## When to run - A confirmed peak event is 6-8 weeks away and the business will run offers or expects a demand surge. - At least one prior year (or comparable period) of data exists to baseline against. - The immediate post-event days, for the built-in reversion phase. ## When NOT to run - The event ended 1-2 weeks ago and you need query cleanup plus the retrospective, that is run-post-peak-season-normalization. - You want a durable, non-calendar growth push, use scale-bids-and-budgets. - You are redistributing budget between campaigns in normal conditions, use allocate-budget-across-campaigns. ## Prerequisites - Prior-year daily performance data for the equivalent period (pullable via VigilDog). - The confirmed commercial plan: offer type, discount depth, promo calendar, and an approved peak budget. HUMAN STEP (outside VigilDog) to obtain, VigilDog cannot decide the offer. - Current daily budgets and tCPA/tROAS targets for all active campaigns, recorded so reversion has exact restore values. - Seasonal keyword list from last year, or fresh seasonal keyword research. ## Procedure Consult `gads_policy_guardrail` before the first Google Ads write of the session. Every write below is previewed first (validate_only=true default) and applied only after explicit user approval. **Baseline (6-8 weeks out)** 1. Pull last year's equivalent window with `gads_run_gaql_query`: daily campaign metrics, cost, clicks, CTR, average CPC, conversions, conversion value, conversion rate, and both impression-share-lost columns (budget and rank). Exclude brand campaigns from the main cut and read them separately. 2. Split those days into three bands, the two weeks before the event, the event itself, and the week after, and compute band averages. The peak-versus-before deltas quantify what the season does to CVR, CPC, and volume. Note that Performance Max exposes no auction/impression-share data at campaign level; read PMax at spend/conversion/ROAS level only. 3. Set the peak plan with the user: peak-period targets (typically 10-20% looser than standard), peak daily budgets (last year's spend plus the growth goal; sanity check that daily budget is at least 5-10x target CPA), an event revenue goal, and a CPC ceiling (prior-year peaks usually ran 25-50% above normal CPC). Record current budgets/targets as the restore point. **Build (2-6 weeks out)** 4. HUMAN STEP (outside VigilDog): seasonal landing pages built and speed-tested; Merchant Center feed work for Shopping/PMax, seasonal title terms via supplemental feed (3-5 business days processing), sale-price annotations (base price must have been stable ~45-60 days), and Merchant Center promotions with hard end dates. 5. Stage seasonal keywords ~3 weeks out with `gads_add_keywords` (and re-enable last year's paused ones). Tag every seasonal keyword with a shared label or naming convention so the whole set can be paused in one filtered pass later. Preview first, apply only after user approval. 6. Stage seasonal assets ~2-3 weeks out, each with explicit start/end dates: - `gads_create_promotion_extension` carrying the event occasion and the offer. - `gads_create_sitelink` pointing at deal/category pages. - `gads_create_callout` for urgency and value messages. Seasonal RSAs, images, and video are handled by the creative skills; coordinate so ad review finishes in time. Submit everything at least 2-3 business days before the event, review queues are slow exactly when you need them fast. Preview first, apply only after user approval. **Execute (event window)** 7. HUMAN STEP (outside VigilDog): if the event is a sharp 1-7 day spike, configure a seasonality bid adjustment in the Google Ads UI (bid strategy advanced controls) sized to the expected CVR lift (CVR historically doubles → +100%), dated to the exact event window. VigilDog has no tool for this control. Never stretch an existing adjustment's dates, create a fresh one. 8. Raise budgets at event start with `gads_update_budget` to the approved peak levels. For PMax, step up gradually (about 20-30% per day), a sudden jump pushes spend into low-intent placements. Preview first, apply only after user approval. 9. Adjust targets only for multi-week ramps: `gads_set_campaign_bidding_strategy` to lower tROAS or raise tCPA by 10-20%, ideally tiered into 2-3 campaign groups by margin/priority. For short spikes, leave targets alone, the seasonality adjustment from step 7 covers it; doing both double-counts the lift. Preview first, apply only after user approval. 10. Monitor live with `gads_run_gaql_query`: CPC is the primary real-time signal (conversions lag); check budget-lost impression share frequently on the peak day, any budget loss on peak day is revenue left on the table; compare pacing against plan and shift budget intraday from laggards to leaders via `gads_update_budget` (preview → approval → apply). **Revert (0-3 days after)** 11. Restore budgets to the recorded pre-peak values via `gads_update_budget` (PMax stepped down gradually) and targets via `gads_set_campaign_bidding_strategy` if they were changed. Preview first, apply only after user approval. 12. Pause the labeled seasonal keyword set via `gads_pause_keyword` and confirm via a label- filtered `gads_run_gaql_query` that none remain enabled. Seasonal assets should expire on their end dates; verify, and update any stragglers via `gads_update_campaign`-adjacent asset tooling as needed. Preview first, apply only after user approval. 13. HUMAN STEP (outside VigilDog): for sharp spikes, add a data exclusion in the bid strategy advanced controls covering the event window so automated bidding does not learn from anomalous days, include the conversion-lag tail (a 3-day sale with a 5-day conversion window means excluding roughly 8 days), and keep exclusions short (about 3 days typical, 7 max). Also revert Merchant Center feed changes. Then schedule run-post-peak-season-normalization for 1-2 weeks out. ## Decision rules - Sharp event (1-7 days) → seasonality bid adjustment, targets untouched. Multi-week ramp → targets loosened 10-20% starting 2-4 weeks out, no seasonality adjustment. Never both at once. - Separate seasonal campaigns only when the promotion has a genuinely different audience or economics; otherwise run seasonal copy inside existing campaigns and keep their bidding history. - Peak daily budget floor: 5-10x target CPA per campaign; below that, automated bidding starves. - PMax budget moves in either direction: ~20-30% per day, never a single jump. - No discount does not mean no participation: shipping, bundles, and added-value angles convert seasonally without margin erosion, decide in the offer-validation step, not by default. - All seasonal elements carry explicit end dates at creation time; reversion should be mostly verification, not work. ## Common failure modes - Assets submitted the day before the event and stuck in review through the peak. Stage everything 2-3 business days early; feed changes a week early. - Peak settings left live after the event: loosened targets plus normal demand equals a quietly terrible week. The revert steps run on a 0-3 day timer regardless of how busy the week was. - Budget exhausted in the ramp-up so the peak day itself is capped, pace conservatively before, front-load the event day. - Doubling the lift by combining a seasonality adjustment with loosened targets, then wondering why CPCs exploded. - Skipping the restore-point recording in step 3, making reversion guesswork. ## Related skills - run-post-peak-season-normalization, mandatory follow-up 1-2 weeks after the event. - scale-bids-and-budgets, the non-seasonal version of raising volume. - allocate-budget-across-campaigns, cross-campaign budget logic used during intraday shifts. - set-up-ad-extensions, mechanics for the asset types staged in the build phase. - run-a-daily-account-health-check, run daily through the peak window.