Demand Gen gets measured badly for two reasons: its conversions include views and video engagements that Search does not have, and it gets judged with Search's CPA when its job is a different one. This lesson sets the fair yardstick and the bidding that goes with it.
The bidding strategies
| Strategy | When |
|---|---|
| Maximize conversions (no target) | Launch: accumulate data for 2-4 weeks on a contained budget |
| Target CPA | With 30+ conversions/month; the target set from the real figure, different from Search's |
| Maximize conversion value / Target ROAS | Ecommerce with a feed and values |
| Maximize clicks | Only for traffic to content; not for conversion |
Portfolios work just as they do in Search; and so does the rule on changes (small, spaced out) — Demand Gen restarts learning with every target change or mass creative change.
The conversions that inflate
- View-through: they saw the card or the video, didn't click, converted later. A generous default window.
- Engaged view: they watched ≥ 10 s of a skippable video (or finished it) and converted in the following days. Google counts it in "Conversions" for video depending on your settings.
Both are genuine signals of influence, and both claim purchases that would have happened anyway. The honest configuration:
- View-through window: 1 day (or the shortest available).
- Engaged-view window: short (3 days or less).
- In reporting: click-through conversions as the main column; view and engagement separately as an "estimated contribution".
- For Smart Bidding: decide whether view-based conversions feed the bidding (the conversion action's settings allow it); in lead generation, better not.
The fair yardstick
Demand Gen is not Search: the person who sees your ad was not searching. A click-through CPA 30-100% higher than Search's can be good if it brings new customers Search would never have brought. The right yardstick combines:
- Click-through CPA against a target of its own (derived from customer value, not copied from Search).
- Incrementality: holdout, geo test or reading the total (module 7 of the Intermediate level); Demand Gen has native A/B experiments (below).
- New customers: the % of conversions from users who were not on your lists (Google shows it when you supply customer data).
- Upper-funnel effects: brand searches and direct traffic during the campaign weeks (time series).
Native A/B experiments
Demand Gen allows experiments from within the campaign itself: two variants (creative, audience, bid) with the traffic split and a result with significance. Use them for the big questions: video vs image? narrow vs broad lookalikes? with expansion or without? Four weeks minimum; one variable per experiment.
Defending Demand Gen to whoever pays
One page: click-through CPA vs its own target · % of new customers · view-based contribution (separate, labelled) · incrementality test (if there is one) · the trend in brand searches. And the sentence that heads off the usual argument: "it isn't competing with Search for the same customer; it brings in the ones Search never sees" — if the data backs it up. If it doesn't, the honest decision is to cut it back to remarketing.
💡 Ninja trick: view-based conversions are also a lead quality problem: a form that arrives "because they watched a video" tends to be colder. Lead Rating reflects that in the score (layer C: the source channel carries weight) and sends the real value back to Google; that way Demand Gen gets judged on customers, not on forms.
What you should remember
- Launch with Maximize conversions; a Target CPA of its own once there is volume; value/ROAS in ecommerce.
- View-through at 1 day and a short engaged-view window; click-through as the main figure; view separately.
- The yardstick: your own CPA + incrementality + new customers + brand effects.
- Native A/B experiments: one variable, four weeks.