--- name: scale-bids-and-budgets description: >- Pushes a stable campaign or portfolio beyond its current plateau by locating the net-profit optimum on the target/volume curve and moving toward it deliberately, via a 50/50 experiment for big moves or 10-15% increments for small ones, raising budgets first whenever budget, not the target, is the binding constraint. Use it when performance is stable and the goal is more volume or more profit, when you suspect the target is over-tightened and choking volume, or on a monthly/quarterly optimization pass. Not for first-time target math (calculate-bid-targets), initial account budgeting (allocate-budget-across-campaigns), rescuing a capped campaign (handle-budget-limited-campaigns), or diagnosing an unhealthy strategy (monitor-and-maintain-bid-strategy-health). --- # Scale Bids and Budgets Toward the Profit Optimum ## Purpose Between "bidding so aggressively you hand all margin to the auction" and "bidding so tightly volume dries up" sits a net-profit peak. This skill finds where the campaign currently sits relative to that peak, models what a target or budget move would do, executes the move with validation proportional to its risk, and iterates, because scaling is a loop, not an event. ## When to run - Learning is long done, performance is stable, and the business wants more volume or more profit. - IS lost to budget is high on a campaign hitting its efficiency target. - Actual performance beats the target by a wide margin (a sign the target is over-tightened and volume is being left behind). - Scheduled monthly/quarterly target-and-budget review. ## When NOT to run - The campaign has never had an economics-derived target -> calculate-bid-targets. - Whole-account budget distribution is the question -> allocate-budget-across-campaigns. - The campaign is pinned by a fixed budget cap that cannot rise -> handle-budget-limited-campaigns. - The strategy is in learning, misconfigured, or missing volume -> monitor-and-maintain-bid-strategy-health; never scale an unhealthy strategy. - Migrating strategy types -> migrate-from-manual-to-smart-bidding. ## Prerequisites - Stable performance: learning complete, no major change in the last two conversion cycles. - Known breakeven CPA/ROAS/profit-per-spend (from calculate-bid-targets). - Current target, trailing 30-day actuals, and the account's conversion lag. - Growth and efficiency goals to score outcomes against. ## Procedure 1. **Session guardrail.** Call `gads_policy_guardrail` before the session's first write. 2. **Baseline pull.** Via `gads_run_gaql_query`, capture the last 4 weeks (excluding conversion-lag days): cost, conversions, conversion value, actual CPA/ROAS, IS lost to rank and to budget. Compute net result = conversion value - cost (lead gen: conversions x value per conversion - cost). This is the comparison anchor for everything that follows. 3. **Locate the position.** Compute the profit-reinvestment share implied by the current target: - CPA strategies: share = target CPA / breakeven CPA. - ROAS strategies: share = breakeven ROAS / target ROAS. - Profit-per-spend: share = 100% / target. Read the zone from Decision rules and derive the scaling direction (loosen, tighten, or budget first). 4. **Model scenarios.** Where available, pull `campaign_simulation` rows via `gads_run_gaql_query` for estimated conversions/cost/value at several target levels between the current target and near-breakeven; supplement with `gads_keyword_forecast` for volume-at-cost sanity. Compute net result per scenario and find the peak. HUMAN STEP (outside VigilDog): Performance Planner in the Google Ads UI for budget-scenario curves if deeper modeling is wanted. Apply a 10-20% haircut to every optimistic projection, simulators assume stable competition and no creative fatigue. 5. **Write the plan.** One short block for user approval: current target/budget, proposed target/budget, the modeled peak, expected volume and net-result impact, and the chosen execution method from Decision rules. 6. **Execute, experiment path (preferred for moves >25% or high-stakes campaigns).** `gads_create_experiment` with the only variable being the new target (or target + budget), 50/50 split; `gads_schedule_experiment` for 30+ days. Preview each write, apply after approval. Touch neither arm during the run. 7. **Execute, incremental path (moderate, proven-direction moves).** Targets: change 10-15% per step via `gads_set_campaign_bidding_strategy`, one full conversion cycle between steps. Budgets: raise 15-20% per step via `gads_update_budget`, never more than 30% in one change, spaced 1-2 weeks. Every step is previewed (validate_only=true default) and applied only after approval. 8. **Execute, direct path (small, low-stakes moves <10-15%).** Single previewed-and-approved change via `gads_set_campaign_bidding_strategy` and/or `gads_update_budget`, then monitored through the learning wobble. 9. **Evaluate.** After the experiment matures (30+ days, significance ~80%+) or increments stabilize, re-pull step 2's metrics excluding learning and conversion-lag windows. Score against the baseline per the Decision-rules matrix. 10. **Close the loop.** Experiment won -> `gads_promote_experiment` (previewed, approved). Baseline won -> end the experiment and keep the original. Then document the new baseline and either iterate (another pass from step 3) or hold at the optimum. Re-check the optimum monthly: it moves with competition and seasonality. ## Decision rules - **Zones by reinvestment share:** >=0.75 -> near breakeven, maximum growth, minimal retained profit, consider tightening; 0.50-0.75 -> balanced, hunt the optimum; 0.25-0.50 -> conservative, volume likely left on the table; <0.25 -> over-tightened, volume-starved, loosen now. - **Budget before target:** if IS lost to budget > ~15% at acceptable efficiency, budget is the binding constraint; raising the target first just pays more per conversion for the same cap. - **Method by move size:** <10-15% -> direct; up to ~25% in steps -> incremental; >25% or high-stakes -> 50/50 experiment. - **Step limits:** targets 10-15% per move; budgets <=30% per move (larger single budget jumps can re-trigger learning), 1-2 weeks apart. - **Outcome matrix:** net result up + growth goal met -> lock in, new baseline. Net result up, goal not yet met -> iterate same direction. Net result down but volume up -> only keep if the growth goal explicitly prices that trade. Net result down, volume flat -> revert and investigate. Still budget-capped after a raise -> raise again within step limits. - **Diminishing-returns tell:** each successive loosening buys less volume per point of margin given up; when the modeled curve flattens, stop, spend the effort on conversion rate, offer, or new inventory instead. - **Evaluation hygiene:** exclude learning windows and the trailing conversion-lag days from every comparison; use value (value-based strategies) or conversions (volume strategies) as the primary metric and net result as the tiebreaker. ## Common failure modes - **Scaling by feel.** Changing targets without scenario modeling is guessing with money; pull the simulation/forecast first, even if crude. - **One giant jump.** A >25% target move triggers a long relearn and makes the result unattributable. Increments or an experiment, always. - **Trusting the projection.** Simulator curves are best-case; the 10-20% haircut is the difference between a plan and a disappointment. - **Scaling the wrong constraint.** Raising budget when rank (not budget) limits impressions, or loosening the target when the budget cap binds, spends more without buying volume. Read both IS-lost metrics first. - **Judging too early.** Evaluating inside the learning window or before lagged conversions land reliably makes the change look worse than it is. ## Related skills - Before: calculate-bid-targets (breakeven + starting target), allocate-budget-across-campaigns (initial allocation), monitor-and-maintain-bid-strategy-health (confirm the strategy is healthy enough to scale). - After: monitor-and-maintain-bid-strategy-health (routine watch on the new baseline), handle-budget-limited-campaigns (if a cap re-emerges), set-up-value-based-bidding (if scaling motivates a move to value/profit bidding).