The judge and the decision matrix were born in Search. Applying them to Demand Gen, PMax, Shopping and video calls for adjustments: the conversions are not the same, neither are the latencies, and what a script can change varies by campaign type. This lesson adapts the logic to each of them so the whole account lives under one system.
What changes by campaign type
| Type | Conversions to use in the judge | Typical latency | Particulars |
|---|---|---|---|
| Search | Click-through, reference ones | Short-medium | The system's benchmark |
| Standard Shopping | Click-through with real value (margin) | Short | Judge on ROAS/POAS by margin portfolio |
| PMax | Click-through; view-through kept apart; brand exclusion applied | Medium | The CPA/ROAS mixes channels: judge on the total and watch by channel (module 1) |
| Demand Gen | Click-through; view-through and engaged views kept apart | Medium-long (discovery) | Its own target, different from Search; fatigue as an extra variable |
| Video (awareness) | Not judged on CPA | — | Reach, frequency, CPV; outside the conversion judge |
| Display remarketing | Click-through, 1-day view-through | Short | Budget tied to the size of the list |
View-through conversions: the rule
For the judge, view-through conversions (post-impression, engaged view) do not count in the deciding CPA/ROAS; they are recorded separately as a contribution. The reason: if they count, PMax and Demand Gen always "hit" their target and the budget flows towards the campaigns best at claiming sales, not towards the ones generating them. The conversion action settings let you decide whether they feed Smart Bidding; the judge discounts them either way.
Different latencies, different windows
Every campaign with its own latency (bid strategy report): a consideration Demand Gen campaign can take 10-15 days to bring in 90% of its conversions; a brand campaign, 1. Windows per campaign, not one account-wide window. Without that, discovery campaigns always look worse than they are and lose budget unfairly.
What a script can change in each type
| Type | Budget | Target (tCPA/tROAS) | Notes |
|---|---|---|---|
| Search, Shopping, Display | Yes | Yes (campaign or portfolio) | Full control |
| PMax | Yes | Yes | Through the API; asset group structure editable within limits |
| Demand Gen | Yes | Yes, via the API (mutate) | On maximise conversions/value with no target, budget only |
| Video | Yes | Depends on the strategy | CPV/CPM do not enter the judge |
What stays manual: changing the type of strategy, switching AI Max on or off, restructuring asset groups.
One logic for the whole account
- Portfolios and targets by economics (lesson 1), with Demand Gen and PMax in their own portfolios/targets (not shared with Search: a different yardstick).
- A judge per campaign, with click-through conversions and its own window.
- Budget shared out with the matrix (lesson 3) across all campaigns, with reserves for brand, remarketing and tests, and one nuance: discovery campaigns with positive incrementality signals (module 7 of the Intermediate level) can receive more than their apparent CPA justifies; the ones with no proof cannot.
- Channel monitoring in PMax (module 1) and fatigue monitoring in Demand Gen (module 2) as extra variables in the verdict: a PMax campaign that hits its target but has shifted its spend into rubbish Display, or a Demand Gen campaign hitting its target on creatives that are burning out, gets no more budget until that is sorted.
Typical mistakes
- A single target for Search and Demand Gen "to keep things simple".
- Judging PMax with view-through included and giving it the whole budget.
- One account-wide window: the long-cycle campaigns always lose.
- Moving Demand Gen targets every week: its learning is slower.
- Forgetting that a PMax CPA without brand exclusion is fiction.
💡 Ninja trick: Smart Bidding (SBNS) brings Demand Gen campaigns into the same logic as Search and Shopping: it reads them through GAQL, includes them in the daily statistics and writes to them through the API (budget on all of them; target on the ones that have one). With a safety rule inherited from its design: in TEST mode it applies nothing and, if a write fails, the campaign is left exactly as it was.
What you should remember
- The same judge, a different yardstick per type: click-through conversions, view-through kept apart, a window per latency, its own target for Demand Gen and PMax.
- A script can move budget and target in almost every type; the strategy type and the structure stay manual.
- Budget shared across all campaigns with the matrix, with incrementality and channel/fatigue as extra variables.