The conversation in paid search right now is all about AI. AI Max, agentic search, automated bidding. The capabilities are genuinely impressive and the excitement is justified. But there is a precondition that rarely makes it into the conversation: none of it works properly if your measurement infrastructure is broken. Feed a smart bidding algorithm incomplete data and it makes bad decisions confidently. The AI performs exactly as designed. The problem is what you’re feeding it.
The scale of the infrastructure gap is bigger than most people realise. A recent Google study found that 84% of CMOs are making budget decisions based on ROI data, but only 28% of the agency teams running those campaigns believe they can measure ROI accurately. The money is moving based on data that the people producing it don’t trust. And the AI features sitting on top of that data are optimising toward a version of reality that doesn’t quite exist.
One in five conversions isn’t being counted
The most common source of measurement loss is also one of the least discussed. Safari and Firefox don’t allow Google Tag to run as a third-party script. For any user landing on a client’s website from either browser, the tag cannot collect conversion data. The visit happened, the purchase happened, but the measurement system didn’t see it. On average, that creates upwards of 20% data loss before any other challenge gets involved.
That missing 20% flows directly into how AI campaigns behave. Smart bidding algorithms making making choices about where to spend your budget are making those choices on incomplete information. Keywords get deprioritised. Audiences get undervalued. Attribution models show a picture that’s consistently darker than reality. The algorithm is doing exactly what it’s been asked to do – it just hasn’t been given the full picture.
Why this keeps happening
Measurement infrastructure rarely gets prioritised. Fixing it requires website tag access, development resource, consent frameworks, and a willingness to spend time fixing something invisible before optimising something visible. The path of least resistance, for most teams and most agencies is to press on with campaign work and treat the measurement as good enough.
The problem is that good enough is becoming harder to sustain. Signal loss has been building steadily for years, compounding with every browser update and privacy regulation. And as AI features take on more of the decision-making in campaigns, the quality of the data feeding them matters more, not less. A manual campaign run by an experienced practitioner can compensate for measurement gaps through judgement. An AI-powered campaign cannot.
What the fix looks like
Google has been positioning a set of tools under the umbrella name Data Strengths – the measurement infrastructure that should be in place before layering on AI-powered campaign features. The components are not new: Consent Mode, Enhanced Conversions, and Google Tag Gateway have been available for some time. What’s changed is the recognition that they need to be treated as a foundation, not an optional add-on.
Of these, Google Tag Gateway addresses the signal loss problem most directly. Think of it as a VPN for tags: instead of Google Tag loading from googletagmanager.com (which Safari and Firefox identify as third-party and block), it loads from your own first-party domain. The browser sees a first-party script, passes the data, and the measurement gap closes. Implementation is relatively fast once your development resource is involved, but someone needs to make it a priority first.
Advertisers who have implemented Google Tag Gateway are seeing, on average, a 14% increase in tracked conversions from the change alone. That’s not incremental optimisation. That’s your campaigns finally showing what they were already doing.
Why it matters beyond performance
There’s a commercial consequence to getting this wrong that goes beyond campaign efficiency. Last year, 61% of clients who left their agencies cited dissatisfaction with ROI. A meaningful portion of that dissatisfaction almost certainly had nothing to do with campaign performance. The gap was in the measurement, and nobody flagged it. If you don’t know the data is incomplete, underperformance looks like a strategy problem rather than an infrastructure one.
Treating Data Strengths as a standard audit is worth demanding from your agency. Running through your accounts to identify where signal is leaking, where consent frameworks are incomplete, where Enhanced Conversions hasn’t been deployed, is the kind of work that creates documented proof your investment is being managed carefully.
AI in paid search is only as good as the data it runs on. Fix the foundations before chasing the features, and you’ll get more from both.
