There is a persistent belief among marketers that Google's AI campaign tools are the tricky part - that once you've set up Performance Max or enabled Smart Bidding, the heavy lifting is done. The measurement infrastructure underneath tends to get treated as an afterthought. That belief is producing poor results across accounts, and it is worth being direct about why.
Smart Bidding, Performance Max, Demand Gen - all of these systems run on conversion signals. Not broadly. Not partially. Entirely. When those signals are incomplete, blocked at the tag level, or simply absent, the AI doesn't slow down and wait for better data. It optimises toward whatever it can see. And what it can see may bear very little resemblance to your actual business outcomes.
The AI Isn't Broken - Your Data Is
One of the more frustrating conversations in PPC right now is the one where a client questions whether Performance Max is working, when the underlying issue is a measurement gap that's been there for months. The campaign is doing exactly what it's been told to do - it's just been given the wrong instructions.
If your primary conversion action is a form submission but your tag is only firing on 60% of completions due to consent mode misconfiguration, Smart Bidding is pricing bids against a distorted view of value. It doesn't know what it doesn't know. The algorithm fills the gap with inferences, and those inferences compound errors over time as the model continues to learn from flawed inputs.
This isn't a niche problem. Consent frameworks across UK and EU markets have made accurate conversion tracking considerably harder to maintain. Browser-level restrictions on third-party cookies, ITP on Safari, and varying user consent rates all create conditions where a surface-level measurement setup will have material gaps. The consequence isn't a footnote in your reporting - it's the basis on which your AI campaigns are making every spend decision.
What 'Data Strength' Actually Means in Practice
Google's concept of Data Strength is an umbrella term covering how well conversion signals across your account equip the AI to optimise effectively. It encompasses conversion volume, signal completeness, recency, and the quality of what's being measured. A high-volume conversion action based on page views tells the algorithm very little about which clicks generate real business value. A low-volume but high-fidelity conversion action based on verified purchases tells it far more.
The practical implication is that you need to audit not just whether conversions are tracking, but what they represent and how complete the data is. That means checking tag firing rates against actual transaction or completion data. It means reviewing whether your primary conversion actions reflect genuine value or just activity. It means confirming that enhanced conversions are configured correctly so Google can match back to logged-in user data when consent is withheld.
It also means thinking carefully about conversion value. If you're using a flat value per lead, Smart Bidding treats a low-quality enquiry from a mismatched audience the same as a high-intent request from your ideal customer. Feeding in actual or modelled revenue values - even imperfect ones - gives the system significantly more to work with. The AI doesn't need perfection. It needs signal that correlates with outcomes.
Performance Max and the Signal Dependency Problem
Performance Max is more dependent on conversion data than any campaign type that preceded it. Traditional search campaigns could be managed through a combination of manual oversight, keyword control, and bidding adjustments that didn't require a rich data foundation. PMax operates differently. Its cross-channel inventory allocation, audience signal interpretation, and creative selection all flow from the conversion patterns it observes.
When conversion data is thin or noisy, PMax tends to lean heavily on the easiest-to-reach conversions rather than the most valuable ones. This often manifests as high spend on branded inventory, YouTube view-through conversions, or Display touchpoints that are difficult to validate - not because the algorithm is broken, but because it's found a path to hitting its conversion targets that happens to align with the data it can access most confidently.
The practical fix here isn't exclusively about more data - it's about better-structured data. Segmenting conversion actions so the algorithm can distinguish between a newsletter sign-up and a high-value purchase matters. Using customer lists as audience signals helps. And if you're running PMax alongside standard Search campaigns, making sure conversion actions are consistently assigned across both is critical to avoiding signal conflicts that confuse the system.
Demand Gen Has the Same Problem, Often Overlooked
Demand Gen occupies a different part of the funnel - it's designed to build demand rather than capture it. That framing sometimes leads teams to apply looser measurement standards, on the basis that upper-funnel activity is harder to attribute precisely. But Demand Gen still runs on Smart Bidding, and its optimisation logic still depends on the conversion signals you provide.
If you're running Demand Gen campaigns towards a 'Maximise Conversions' target, the AI needs enough conversion data to understand what a conversion looks like from a demand-gen pathway. If your attribution model doesn't credit Demand Gen touchpoints adequately - which is a common issue with last-click - the campaign is effectively training the model on an incomplete picture of its own contribution.
For Demand Gen specifically, data-driven attribution is worth prioritising even when it feels uncomfortable to move away from last-click. It provides a more accurate signal about which creative, audience, and placement combinations are contributing to downstream conversions. That improved signal quality feeds back into the algorithm and improves future decisions. The compounding effect is meaningful over a campaign's lifetime.
Measurement Is Now a Strategic Advantage
There was a time when measurement was a technical discipline sitting adjacent to campaign strategy. That separation no longer makes sense. If your conversion data is stronger than a competitor running the same campaign types in the same auction, your AI systems will systematically outperform theirs - not because of better creative or smarter targeting choices, but because the algorithm has better inputs to work from.
This reframes the investment case for measurement infrastructure. Server-side tagging, enhanced conversions, consent mode implementation, and conversion value rules are not compliance exercises or reporting improvements. They are performance levers with a direct line to Smart Bidding efficiency and PMax allocation decisions. The brands treating them as such will pull further ahead as AI campaign types become the default across Google's inventory.
A practical starting point is a signal audit: map every conversion action in your account, confirm it's firing accurately by cross-referencing with backend data, assign it a realistic value, and check whether it's contributing signal to the right campaigns. That audit will almost always surface gaps. Closing those gaps is, at this point, the highest-leverage measurement task available to most Google Ads teams.