Start with the P&L, not the keyword list
Before touching campaign settings, write down the business segments that carry different contribution margins: product categories, service lines, new customer versus repeat, or region. Those segments are the only defensible campaign boundaries, because they are the boundaries at which a budget decision actually gets made.
Google's recommendation card optimizes for platform-level efficiency. It has no view of your fulfillment cost, your intake capacity, or the fact that one category is strategic and another is being harvested. That context has to come from structure.
Brand, non-brand, competitor, remarketing
- Separate these four at the campaign level. Blended into one budget, brand's cheap conversions mask non-brand inefficiency and true incrementality becomes unmeasurable.
- Brand exists to defend the SERP and should be evaluated on impression share and defensive cost, not on ROAS.
- Non-brand is where growth is bought; it earns its own budget, its own CPA target, and its own reporting line.
- Competitor terms are a strategic spend with a separate tolerance. Remarketing is a frequency and recency problem, not a keyword problem.
Alpha/Beta: control where it pays for itself
The Alpha/Beta pattern still holds at high spend. Beta campaigns run broader match types with tight CPA guardrails and act as a discovery surface. When a query proves it converts at an acceptable cost with enough volume, it graduates into an Alpha campaign on exact match with its own budget and bid strategy.
Negatives do the enforcement: every graduated query is negated out of Beta so the two never compete. The result is a clean separation between proven revenue, which gets protected, and exploration, which gets capped.
Shopping, Search, and PMax as separate lanes
- Give Shopping its own campaign structure segmented by product margin or price band, not by an arbitrary category tree.
- Run PMax with brand exclusions and a defined product or audience scope so it cannot quietly absorb demand Search was already capturing.
- Keep reporting boundaries explicit: if you cannot say which lane produced an incremental order, the lanes are overlapping.
- Review query and placement overlap monthly. Overlap is the most common cause of rising blended CPA at flat volume.
Multi-location grouping
Location-level businesses fail in one of two ways: a single national campaign that averages high-performing and low-performing markets into one number, or 300 campaigns that fragment conversion volume until no bid strategy can learn.
The working middle is grouping by market economics — demand density, competitive cost, and local capacity — into tiers, then applying location-level budget controls within each tier. That preserves enough conversion volume for bidding to work while keeping spend attributable to a P&L owner.
Maintenance is part of the structure
A structure is only as good as the hygiene around it: shared negative lists organized by theme and updated weekly from search query reports, naming conventions that survive an account handoff, and budget pacing rules that surface a runaway campaign the same day rather than at month end.
If a structural change cannot be explained in terms of the decision it makes possible, it is reorganization, not architecture.
Questions we get
- Should we consolidate campaigns for Smart Bidding?
- Consolidate only where the segments share economics and conversion volume. Consolidating segments with different margins teaches the algorithm to average across them, which quietly funds the weakest one.
- Does Performance Max replace Search campaigns?
- No. PMax works as a lane alongside Search and Shopping, with brand exclusions and clear reporting boundaries so its performance is not being credited with demand Search already owned.
- How many ad groups per campaign?
- One to three at this spend level, so budget and bidding decisions stay attributable. More than that and the campaign becomes an average rather than a control.