In short: every account architecture solves the same conflict: control versus volume. The rule is as consolidated as the economics allow and as separate as the data demands — and each piece only exists on its own if it has enough data (30-50 conversions per strategy). Scaling means adding within the architecture, not rebuilding it every six months.
A small account can be organised on instinct; a growing one gets reorganised every six months unless it was designed to grow. This module collects real architectures by type of business. This first lesson sets out the common principles — the ones that stop you rebuilding the account every time the budget doubles.
Consolidate campaigns or split them?
Every structural decision sits somewhere between two extremes:
- Control: many small campaigns and ad groups, each with its own budget, its own target and its own report. Upside: you know what every euro is doing. Cost: each piece has little data; Smart Bidding learns badly; managing it is expensive.
- Volume: a few large campaigns with plenty of traffic. Upside: the algorithm learns fast; management is light. Cost: you do not know where the money is going; a bad piece hides inside the average.
In 2026 the optimum has shifted towards volume (Smart Bidding needs dozens of conversions per strategy), but not all the way: separating by economics is still non-negotiable. The rule: as consolidated as the economics allow; as separated as the data you need to decide demands.
Which principles govern account architecture?
1. Separate by economics, not by taxonomy
Different campaigns when the budget, the target (acceptable CPA/ROAS), the margin, the region or brand/generic change. Never because of how the website or the catalogue is organised, if the economics are the same.
2. Consolidate so the learning has data
A campaign with 3 conversions a month is not a campaign: it is noise. Group things together (campaigns, portfolios) until each strategy has 30-50 conversions a month. If an economic separation leaves pieces without volume, use portfolios to share learning while keeping budgets apart.
3. Protect the brand
Your own brand campaign, brand exclusion in PMax, brand negatives in generic campaigns. Brand is the piece that distorts everything else most if you mix it in.
4. Universal things, at account level
Universal negatives, placement, app, content and IP exclusions, audience lists, conversions and values: at account level, once. Campaigns inherit; nothing is duplicated by hand.
5. Identify by ID
Every automation, sheet, report and rule references campaigns, ad groups and ads by ID. Names are labels that change; keep a clear naming convention (type · target · region · extra) for humans.
6. Measure before structuring
Structure exists to help you decide, and decisions are made from data: clean conversions with values (Measurement II) before splitting campaigns by margin or by lead quality. Structuring on top of dirty measurement is just tidying up noise.
7. Document
An architecture document: which campaigns exist and why, portfolios and their targets, budget reserves, what is universal and what is not, and the rules for creating a new campaign. It is what stops the next manager (or you, a year from now) undoing what works.
How many conversions does each campaign need?
| Piece | Minimum data to stand alone | If it falls short |
|---|---|---|
| Bidding strategy (campaign or portfolio) | 30-50 conversions/month | A portfolio with its siblings |
| Campaign | A budget that buys ≥ 10-15 clicks/day | Merge with its economic sibling |
| Ad group | Enough impressions for the RSA to learn (thousands/month) | Merge groups with similar intent |
| Asset group (PMax) | Its own conversions within weeks | Fewer groups |
| Product group (Shopping) | Enough clicks to judge it | Move one level up the tree |
How do you scale an account without rebuilding it?
- You add (campaigns, regions, products) inside the architecture; you do not split up what already exists.
- Every new piece arrives with its own budget and an observation period (module 9 of Intermediate).
- Portfolios absorb new pieces that have no data yet.
- The architecture review is quarterly: is any piece chronically short of data? Is any piece now big enough to stand on its own?
💡 Ninja trick: the architecture is also what makes automation possible: the SBNS judge needs campaign targets that make economic sense; Guardian needs clear reserves; the Shield needs account-level exclusions; Lead Rating needs stable IDs. An account with a good architecture is an account where the scripts pay off from day one; in a messy one, you have to tidy up first.
What you should remember
- Control versus volume: as consolidated as the economics allow, as separated as the data demands.
- Seven principles: economics, consolidation, brand, account level, ID, measurement, documentation.
- Each piece stands alone if it has data (30-50 conversions per strategy); if not, a portfolio or a merge.
- Scaling = adding within the architecture; a quarterly review.