Why PPC measurement feels broken (and why it isn’t)

If you’ve been managing PPC accounts for any length of time, you don’t need a research report to tell you something has changed. 

You see it in the day-to-day work: 

• GCLIDs missing from URLs.
• Conversions arriving later than expected.
• Reports that take longer to explain while still feeling less definitive than they used to.

When that happens, the reflex is to assume something broke – a tracking update, a platform change, or a misconfiguration buried somewhere in the stack.

But the reality is usually simpler. Many measurement setups still assume identifiers will reliably persist from click to conversion, and that assumption no longer holds consistently.

Measurement hasn’t stopped working. The conditions it depends on have been shifting for years, and what once felt like edge cases now show up often enough to feel like a systemic change.

Why this shift feels so disorienting

I’ve been close to this problem for most of my career. 

Before Google Ads had native conversion tracking, I built my own tracking pixels and URL parameters to optimize affiliate campaigns. 

Later, while working at Google, I was involved in the acquisition of Urchin as the industry moved toward standardized, comprehensive measurement.

That era set expectations that nearly everything could be tracked, joined, and attributed at the click level. Google made advertising feel measurable, controllable, and predictable. 

As the ecosystem now shifts toward more automation, less control, and less data, that contrast can be jarring.

It has been for me. Much of what I once relied on to interpret PPC data no longer applies in the same way. 

Making sense of today’s measurement environment requires rethinking those assumptions, not trying to restore the old ones. This is how I think about it now.

Dig deeper: How to evolve your PPC measurement strategy for a privacy-first future

The old world: click IDs and deterministic matching

For many years, Google Ads measurement followed a predictable pattern. 

• A user clicked an ad. 
• A click ID, or gclid, was appended to the URL. 
• The site stored it in a cookie. 
• When a conversion fired, that identifier was sent back and matched to the click.

This produced deterministic matches, supported offline conversion imports, and made attribution relatively easy to explain to stakeholders. 

As long as the identifier survived the journey, the system behaved in ways most advertisers could reason about. 

We could literally see what happened with each click and which ones led to individual conversions.

That reliability depended on a specific set of conditions.

• Browsers needed to allow parameters through. 
• Cookies had to persist long enough to cover the conversion window. 
• Users had to accept tracking by default. 

Luckily, those conditions were common enough that the model worked really well.

Why that model breaks more often now

Browsers now impose tighter limits on how identifiers are stored and passed.

Apple’s Intelligent Tracking Prevention, enhanced tracking protection, private browsing modes, and consent requirements all reduce how long tracking data persists, or whether it’s stored at all.

URL parameters may be stripped before a page loads. Cookies set via JavaScript may expire quickly. Consent banners may block storage entirely.

Click IDs sometimes never reach the site, or they disappear before a conversion occurs.

This is expected behavior in modern browser environments, not an edge case, so we have to account for it.

Trying to restore deterministic click-level tracking usually means working against the constant push toward more privacy and the resulting browser behaviors.

This is another of the many evolutions of online advertising we simply have to get on board with, and I’ve found that designing systems to function with partial data beats fighting the tide.

The adjustment isn’t just technical

On my own team, GA4 is a frequent source of frustration. Not because it’s incapable, but because it’s built for a world where some data will always be missing. 

We hear the same from other advertisers: the data isn’t necessarily wrong, but it’s harder to reason about.

This is the bigger challenge. Moving from a world where nearly everything was observable to one where some things are inferred requires accepting that measurement now operates under different conditions. 

That mindset shift has been uneven across the industry because measurement lives at the periphery of where many advertisers spend most of their time, working in ad platforms.

A lot of effort goes into optimizing ad platform settings when sometimes the better use of time might’ve been fixing broken data so better decisions could be made.

Dig deeper: Advanced analytics techniques to measure PPC

---

What still works: Client-side and server-side approaches

So what approaches hold up under current constraints? The answer involves both client-side and server-side measurement.

Pixels still matter, but they have limits

Client-side pixels, like the Google tag, continue to collect useful data.

They fire immediately, capture on-site actions, and provide fast feedback to ad platforms, whose automated bidding systems rely on this data.

But these pixels are constrained by the browser. Scripts can be blocked, execution can fail and consent settings can prevent storage. A portion of traffic will never be observable at the individual level.

When pixel tracking is the only measurement input, these gaps affect both reporting and optimization. Pixels haven’t stopped working. They just no longer cover every case.

Changing how pixels are delivered

Some responses to declining pixel data focus on the mechanics of how pixels are served rather than measurement logic.

Google Tag Gateway changes where tag requests are routed, sending them through a first-party, same-origin setup instead of directly to third-party domains.

This can reduce failures caused by blocked scripts and simplify deployment for teams using Google Cloud.

What it doesn’t do is define events, decide what data is collected, or correct poor tagging choices. It improves delivery reliability, not measurement logic.

This distinction matters when comparing Tag Gateway and server-side GTM.

• Tag Gateway focuses on routing and ease of setup.
• Server-side GTM enables event processing, enrichment, and governance. It requires more maintenance and technical oversight, but it provides more control.

The two address different problems.

Here’s the key point: better infrastructure affects how data moves, not what it means.

Event definitions, conversion logic, and consistency across systems still determine data quality.

A reliable pipeline delivers whatever it’s given, so it’d be just as good at making sure the garbage you put in also comes back out.

Offline conversion imports: Moving measurement off the browser

Offline conversion imports take a different approach, moving measurement away from the browser entirely. Conversions are recorded in backend systems and sent to Google Ads after the fact.

Because this process is server to server, it’s less affected by browser privacy restrictions. It works for longer sales cycles, delayed purchases, and conversions that happen outside the site. 

This is why Google commonly recommends running offline imports alongside pixel-based tracking. The two cover different parts of the journey. One is immediate, the other persists.

Offline imports also align with current privacy constraints. They rely on data users provide directly, such as email addresses during a transaction or signup.

The data is processed server-side and aggregated, reducing reliance on browser identifiers and short-lived cookies.

Offline imports don’t replace pixels. They reduce dependence on them.

Dig deeper: Offline conversion tracking: 7 best practices and testing strategies

How Google fills the gaps

Even with pixels and offline imports working together, some conversions can’t be directly observed.

Matching when click IDs are missing

When click IDs are unavailable, Google Ads can still match conversions using other inputs.

This often begins with deterministic matching through hashed first-party identifiers such as email addresses, when those identifiers can be associated with signed-in Google users.

This is what Enhanced Conversions help achieve.

When deterministic matching, if this then that, isn’t possible, the system relies on aggregated and validated signals rather than reconstructing individual click paths.

These can include session-level attributes and limited, privacy-safe IP information, combined with timing and contextual constraints.

This doesn’t recreate the old click-level model, but it allows conversions to be associated with prior ad interactions at an aggregate level.

One thing I’ve noticed: adding these inputs typically improves matching before it affects bidding.

Bidding systems account for conversion lag and validate new signals over time, which means imported or modeled conversions may appear in reporting before they’re fully weighted in optimization.

Matching, attribution, and bidding are related but separate processes. Improvements in one don’t immediately change the others.

Modeled conversions as a standard input

Modeled conversions are now a standard part of Google Ads and GA4 reporting.

They’re used when direct observation isn’t possible, such as when consent is denied or identifiers are unavailable.

These models are constrained by available data and validated through consistency checks and holdback experiments.

When confidence is low, modeling may be limited or not applied. Modeled data should be treated as an expected component of measurement rather than an exception.

Dig deeper: Google Ads pushes richer conversion imports

Boundaries still matter

Tools like Google Tag Gateway or Enhanced Conversions for Leads help recover measurement signal, but they don’t override user intent. 

Routing data through a first-party domain doesn’t imply consent. Ad blockers and restrictive browser settings are explicit signals. 

Overriding them may slightly increase the measured volume, but it doesn’t align with users’ expectations regarding how your organization uses their data.

Legal compliance and user intent aren’t the same thing. Measurement systems can respect both, but doing so requires deliberate choices.

Designing for partial data

Missing signals are normal. Measurement systems that assume full visibility will continue to break under current conditions.

Redundancy helps: pixels paired with hardened delivery, offline imports paired with enhanced identifiers, and multiple incomplete signals instead of a single complete one.

But here’s where things get interesting. Different systems will see different things, and this creates a tension many advertisers now face daily.

Some clients tell us their CRM data points clearly in one direction, while Google Ads automation, operating on less complete inputs, nudges campaigns another way.

In most cases, neither system is wrong. They’re answering different questions with different data, on different timelines. Operating in a world of partial observability means accounting for that tension rather than trying to eliminate it.

Dig deeper: Auditing and optimizing Google Ads in an age of limited data

Making peace with partial observability

The shift toward privacy-first measurement changes how much of the user journey can be directly observed. That changes our jobs.

The goal is no longer perfect reconstruction of every click, but building measurement systems that remain useful when signals are missing, delayed, or inferred.

Different systems will continue to operate with different views of reality, and alignment comes from understanding those differences rather than trying to eliminate them.

In this environment, durable measurement depends less on recovering lost identifiers and more on thoughtful data design, redundancy, and human judgment.

Measurement is becoming more strategic than ever.

View the full article