--- name: set-up-performance-based-shopping-segmentation description: >- Restructures Shopping spend around historical product performance: products get tiered by trailing ROAS into a custom label, and each tier gets its own campaign bucket with its own budget but a shared ROAS target. Use it when a mature catalog (90+ days of conversion data, 50+ SKUs) shows clear winners and losers and budget is being dragged down by the losers. It reads the ROAS baseline, verifies label sync, builds the bucket campaigns and product-group filters, and sets uniform tROAS. Do not use it for multi-factor prioritization by margin/stock/price (set-up-scoring-model-segmentation), for the initial feed build (set-up-and-optimize-product-feed), for a first unsegmented launch (launch-standard-shopping-campaign / launch-pmax-feed-only-campaign), or for routine tuning of an already-segmented setup (run-shopping-campaign-optimization-cycle). --- # Segment Shopping Campaigns by Product Performance Tier ## Purpose Split one undifferentiated Shopping budget into tiered campaign buckets so proven products get most of the spend, chronic money-losers get contained, and never-shown products get a deliberate testing budget. The tier assignment lives in a feed custom label maintained by an external labeling tool; this skill wires the Google Ads side: baseline math, bucket campaigns, listing-group filters, uniform ROAS targets, and overlap exclusions. ## When to run - 90+ days of Shopping conversion history exists and product-level ROAS varies widely. - Budget is constrained and a meaningful share of spend sits on products that never pay back. - A large slice of the catalog gets essentially no impressions and needs forced exposure. - A labeling tool (or feed-rule pipeline) is available to write a tier value into a custom label. ## When NOT to run - Tiering should weigh margin, stock, or price competitiveness, not just ROAS -> set-up-scoring-model-segmentation. - The feed itself is weak or brand new -> set-up-and-optimize-product-feed first. - No campaign exists yet -> launch-standard-shopping-campaign or launch-pmax-feed-only-campaign. - Under ~50 SKUs or under 90 days of data: tiers would be noise -> keep one campaign and run run-shopping-campaign-optimization-cycle instead. - Products all perform in a similar band: segmentation adds structure without payoff. ## Prerequisites - Merchant Center linked to the Google Ads account, feed approved and stable. - Conversion tracking with values verified (set-up-google-ads-conversion-tracking if unsure). - A labeling tool or feed pipeline able to sync a tier value into `custom_label_0` daily. - Current account tROAS and 30-90 day average ROAS known or computable. ## Procedure 1. **Session guardrail.** Before the first Google Ads write of this session, call `gads_policy_guardrail` and honor its output. 2. **Pull the baseline.** Call `gads_get_shopping_performance` for the trailing 90 days, and `gads_run_gaql_query` on `shopping_performance_view` for the trailing 30 days to compute: account-level average ROAS, total Shopping conversions/month, and the product-level spread (spend and ROAS per item). Record average ROAS, the tool threshold derives from it. 3. **Classify the catalog (analysis only).** From step 2 data, bucket products into four neutral tiers: consistent high-ROAS sellers, moderate-but-profitable products, high-spend/low-return products, and near-zero-impression products. Report the counts and spend share per tier to the user, this drives the bucket-structure decision. 4. **Choose the bucket structure** with the user, per Decision rules: two buckets (growth or cost-control variant) or one campaign per tier if every major bucket clears ~30 conversions/month. 5. **HUMAN STEP (outside VigilDog).** Configure the labeling tool: set its ROAS threshold to average ROAS x 0.75 (see Decision rules), map output to `custom_label_0`, enable daily sync to Merchant Center, and confirm the first sync in the Merchant Center UI. VigilDog has no tool for labeling-tool or Merchant Center feed configuration. 6. **Verify labels landed.** Call `gads_run_gaql_query` on `shopping_performance_view` segmented by the product custom attribute to count products per label value. Compare against the sanity ranges in Decision rules; if the top tier is nearly empty or the bottom tiers dominate, the threshold is set too high, send it back to step 5 before building anything. 7. **Plan the buckets.** Call `gads_plan_shopping_campaign` once per bucket to assemble a coherent plan (naming, budget share, bidding, filters) and present the full plan to the user. 8. **Create budgets and campaigns (write).** For each bucket: `gads_create_budget` then `gads_create_shopping_campaign`, budget split per Decision rules. Preview first (validate_only=true default), show the user each campaign's name/budget/target, apply only after explicit approval. 9. **Filter each bucket to its tiers (write).** In each campaign, use `gads_create_product_group` to subdivide by `custom_label_0` and include only that bucket's tier values, and `gads_exclude_product_group` to shut out every other value. Preview first, apply only after user approval. Verify the resulting tree with `gads_list_product_groups`, each tier value must be servable in exactly one campaign. 10. **Seal the legacy campaigns (write).** For every pre-existing Shopping/PMax campaign that could still serve these products, apply `gads_exclude_product_group` on the now-segmented label values. Preview first, apply after approval. Skipping this creates self-competition and splits the data. 11. **Set bidding (write).** Call `gads_set_campaign_bidding_strategy` to apply the SAME tROAS on every bucket. Optional exception: the zero-impression bucket may run Maximize Conversion Value with no target to force delivery. Preview first, apply after approval. 12. **Day-1 and week-1 validation.** Via `gads_run_gaql_query`: every bucket serving, spend pacing close to the intended split, the zero-impression bucket actually accruing impressions, and no product serving from two campaigns. Fix filter gaps immediately. 13. **Monthly maintenance.** Re-run steps 2 and 6: confirm products migrate between tiers over time (a frozen distribution means the sync or threshold broke), and re-check that the tool threshold still sits ~25% under the live tROAS. After ANY tROAS or budget change, recompute the threshold and have the human update the tool (step 5). ## Decision rules - **Tool threshold formula:** labeling-tool ROAS threshold = trailing 30-90 day average ROAS x 0.75. The ~25% buffer below the campaign tROAS keeps a product that dips briefly from being demoted and stranded. - **One tROAS for all buckets.** Differentiate buckets by BUDGET, never by target. A lower target on the weak bucket means a recovering product never clears the bar to climb back up. - **Volume gate:** every bucket bidding to tROAS needs ~30 conversions/month of its own. Any bucket below that gets merged into a neighbor. The zero-impression bucket is exempt when it runs Maximize Conversion Value without a target. - **Structure variants:** growth focus -> bucket A = winners + moderates + losers (70-80% of budget), bucket B = zero-impression products (20-30%); cost-control focus -> bucket A = winners + moderates + zero-impression (70-80%), bucket B = losers (20-30%); full split (one campaign per tier) only when volume allows, roughly 50-60% / 20-25% / 20-25% across winner/loser/dormant lines. - **Label distribution sanity check (step 6):** top tier ~5-15% of catalog, moderate ~15-25%, low-return ~25-40%, near-zero-impression ~30-50%. Outside these bands, suspect the threshold, not the catalog. - **Patience windows:** label sync 24-48h; bidding re-learn 7-14 days after the restructure; do not judge bucket performance before ~30 days. ## Common failure modes - **Different targets per bucket.** Feels intuitive ("go easy on the weak ones"), quietly pins products in the bottom tier forever. Uniform target, tiered budget. - **Threshold never revisited.** The tROAS gets raised later but the tool threshold stays put, so the buffer collapses and tiers thrash. Re-derive the threshold after every target change. - **No exclusions on legacy campaigns.** The old catch-all campaign keeps bidding on segmented products; both campaigns learn from half a signal. Step 10 is not optional. - **Dormant bucket starves.** A tROAS target on products with no history means the algorithm never risks an impression. Give that bucket Maximize Conversion Value (no target) or a bid floor. - **Judging tiers on a week of data.** Tier migration is the health signal, and it plays out over weeks. Monthly review cadence, not daily. ## Related skills - Before: set-up-and-optimize-product-feed (labels ride on a healthy feed), set-up-google-ads-conversion-tracking (value data must be trustworthy). - During: launch-standard-shopping-campaign or launch-pmax-feed-only-campaign (per-bucket build details). - After: run-shopping-campaign-optimization-cycle (recurring tuning), set-up-scoring-model-segmentation (graduate to multi-variable prioritization), handle-budget-limited-campaigns (if a winning bucket hits its budget ceiling).