GOOGLE ADS DEMAND GEN CAMPAIGNS: FIRST LAUNCH, REAL SETUP

Every time I launch google ads demand gen campaigns in a brand-new account, the part that decides success or failure happens before anyone sees a single impression, and then again in the fourteen days that follow. People imagine the work is the creative. In my experience the work is the setup and the discipline to leave it alone while the system learns. I have launched enough of these to see the same handful of structural errors repeat across new accounts, and almost all of them come from impatience or from copying an old Search-campaign mindset into a machine that does not reward it. This article is the behind-the-scenes version: the real decisions I make at setup, what the first two weeks actually look like from inside the account, and the mistakes I now warn every new advertiser about before they spend a euro. If you want the conceptual grounding first, start with how Demand Gen works as a campaign type and then come back here for the launch playbook.

The real setup of demand gen campaigns google ads

When I open the campaign builder, the first real decision is the advertising goal, and it is not cosmetic. The goal you pick shapes the conversion goals available and quietly constrains where your ads can serve, so I match it to the actual outcome the account needs rather than to ambition. For most new accounts that means Sales or Leads, with conversion goals that fire on something real rather than a soft page view. The official Google guide to creating a Demand Gen campaign lays the six steps out cleanly, but the order of importance on the ground is goal, then budget, then audience, then creative, with everything bending toward giving the system enough conversion signal early.

On channels, I leave the default of all Google channels on almost every launch. New advertisers instinctively want to hand-pick surfaces, but restricting channels at setup starves the model of the inventory it uses to find converters, and the campaign learns slower for no real gain. The one setting I am deliberate about is targeting level: location and language can be set at campaign or ad group level but never both, and that choice is locked once you publish, so I decide it consciously instead of discovering it later. The rest of the build is audiences and creative, which I treat as inputs to a learning process rather than as the finished product.

What the first 14 days of the learning phase actually look like

The first fortnight is where most new advertisers lose their nerve, and where I do the least. Day one to three, the numbers are chaos: cost per action swings wildly, the conversion rate lurches, spend appears without conversions attached. This is not failure, it is the system buying impressions across YouTube, Discover, Gmail and the Display Network to find out who responds to your ads. Google's own advice is to let roughly 50 conversion events accumulate before judging the bid strategy, and on a fresh campaign that is exactly the volume that takes a couple of weeks to build. So I read the early days as exploration, not performance.

By the end of the first week, if the budget is sized correctly, the campaign has usually gathered enough conversions to start stabilising, and a pattern emerges in which audiences and placements actually convert. The temptation at this point is to start cutting, but acting mid-learning resets the very process you are waiting on. I keep a light hand: I confirm conversion tracking is firing, I check the budget is pacing toward meaningful volume, and I note where conversions are clustering without acting on it yet. Around day ten to fourteen the bid strategy label typically stops saying "Learning", cost per action settles into a believable range, and only then do I treat the data as a basis for decisions. The single most expensive mistake I see is judging a campaign on day-three clicks, because every change made in that window throws away the learning the spend already paid for.

The structure mistakes I see repeat in new accounts

The error I encounter more than any other is fragmentation. A new advertiser, reasoning by analogy with granular Search structures, splits one modest budget across four or five narrow campaigns, each chasing a different audience. Every one of those campaigns then starves: none reaches the conversion volume to clear learning, so they all sit in the learning phase indefinitely, burning money without ever stabilising. Demand Gen rewards the opposite instinct. One consolidated campaign with enough volume to teach the model will beat five thin ones every time, because the algorithm needs concentrated signal, not scattered fragments of it. When I inherit a messy new account, consolidation is usually the first and most impactful fix.

The second recurring mistake is structural in a quieter way: treating the new-customer-acquisition goal as if it works the way it does in Performance Max. It does not carry over the same, so prospecting in Demand Gen runs through lookalike segments and audience exclusions rather than a single NCA switch. The third is misunderstanding where the ads even appear and briefing creative for the wrong context. If you are unsure, it is worth confirming exactly which surfaces Demand Gen ads run across before you design a single asset, because a creative built for a feed reads completely differently when it lands as a skippable video. Structure, in other words, is not just how many campaigns you run, it is whether the architecture matches how the system actually serves and learns.

Bidding and budget: funding the learning phase properly

Bidding is where the setup either supports the learning phase or quietly sabotages it. I launch almost every new Demand Gen campaign on Maximize Conversions, then move to target CPA or target ROAS only after the campaign has at least 50 conversions behind it. This is also Google's recommended sequence, and the reasoning is simple: a target set before the system has any conversion history is a number imposed on a model with nothing to optimise against, so it either chokes delivery or chases the wrong users. Maximize Conversions lets the campaign gather data freely, and the target I set afterward is grounded in real performance instead of a hopeful guess. Choosing the right starting strategy is the same discipline I apply across the full range of Demand Gen bid strategies, where the launch choice and the scaling choice are deliberately different decisions.

Budget is the other half of the same problem, and it is almost always set too low. I size the daily budget against the conversion volume the campaign needs to exit learning, not against what feels safe. If the target CPA is around twenty euros and the system wants roughly 50 conversions to settle, the campaign has to be funded to reach that volume inside two weeks, which in practice means a daily budget of at least ten to fifteen times the target CPA. Underfunding a fresh campaign is the quietest way to kill it: too little budget means too few conversions, which means learning never ends and performance never stabilises. I would rather launch one properly funded campaign than three starved ones fighting over the same scraps of data. The whole machine runs on conversion signal, and the same signals that feed Demand Gen also power broader automated bidding across the account, which is why protecting that signal early pays off everywhere later.

Creative variety is a structural decision, not a finishing touch

I used to think of creative as the last step, the bit you bolt on once the structure is right. Launching enough Demand Gen campaigns cured me of that. Creative variety is structural, because the system can only mix and match the signal you actually give it. Upload a single video and you have handed the algorithm one note to play across every surface; upload a spread of videos and images with different focal points and contexts, and you have given it real combinations to test. The data backs this plainly: advertisers who supply both video and image assets consistently see more conversions at the same cost than those who upload video alone, and in my accounts that gap is one of the most reliable patterns there is.

So at setup I treat the asset spread as a deliberate part of the architecture. Each ad group gets several videos and several images, the full complement of headlines and descriptions the format allows, and at least one product feed where the catalogue justifies it. The point is not volume for its own sake, it is variety the model can learn from. A thin creative set is one of the hidden reasons a campaign lingers in the learning phase long past the point the budget should have carried it out, because there is simply not enough distinct material for the system to find what converts.

Reading the data once the learning phase ends

When the "Learning" label clears and the cost per action settles, the account finally hands you data worth acting on, and the discipline shifts from waiting to reading. The first thing I check is whether conversions are genuinely concentrating in particular audiences or placements, or whether the apparent pattern is still thin enough to be noise. With a campaign that cleared learning on around 50 conversions, I am cautious about over-reading a fifty-row report; I look for direction, not precision, and I change one variable at a time so I can attribute whatever happens next.

The honest read of early Demand Gen data is that it tells you whether the foundation is sound, not whether the campaign is optimised. If conversions came at a believable cost across a sensible spread of surfaces, the structure is working and the job becomes incremental: prune the weakest assets, reinforce the audiences that converted, and shift bidding toward a target now that there is history to anchor it. If the campaign spent through two weeks and produced almost nothing, the problem is upstream of optimisation, usually a tracking gap, a budget too small to reach volume, or audiences too narrow to contain buyers. I resist the urge to fix a structural problem with a tactical lever, because no amount of bid tweaking rescues a campaign that was never funded to learn in the first place.

After learning: the rhythm that keeps a campaign healthy

Once a Demand Gen campaign has cleared the learning phase and proven its foundation, the work becomes a rhythm rather than a relaunch. I move bidding from Maximize Conversions to a target only when there are at least 50 conversions to base it on, and even then I set the target gently, often a notch looser than the recent average, so the shift does not jolt the campaign back into learning. Every meaningful change I make is treated as a small reset, so I make few of them and space them out, letting each one settle before stacking the next.

From there the cadence is steady. I refresh creative before fatigue sets in rather than after, retire the assets that never earned their place, and feed the audiences that converted with fresh variations on what worked. The new accounts that thrive are the ones where the operator internalised early that demand gen campaigns google ads reward patience and concentration over fiddling: fund one campaign properly, give it varied creative, leave it through the first fourteen days, then read the data honestly and adjust in small, deliberate moves. The mistakes that wreck new accounts are almost never exotic. They are fragmentation, starvation, and impatience, and every one of them is avoidable if you respect what the system is actually doing in those first two weeks.