I have been doing this for a long time.

Long enough to have made the case for Flash. Long enough to have then made the case against it. Long enough to have sat in rooms where mobile-first was a radical proposition, where cloud infrastructure needed a business case, where someone senior always asked why we could not just keep doing what we were doing.

Every few years something arrives that rewrites the operating assumptions. And every single time, somewhere in the resulting chaos, someone stands up and says the same thing. We need to make data-led decisions.

I have said it myself. Probably more times than I can count.

What I did not fully clock, until fairly recently, is that the data we kept referring to was never really ours.

Not in the sense that it was stolen or falsified. The numbers were broadly fine. The dashboards worked. The attribution models produced something coherent enough to justify a budget decision or defend a channel strategy in a quarterly review. Nobody was lying. The infrastructure was genuinely useful. We made it work and mostly it did.

But useful is not the same as independent. And that, I think, is where the problem actually sits.

It was never that companies had only one source of data. Most had plenty. Marketing had its numbers. CRM had its records. Finance had its reporting. Product had its own events and dashboards. Customer service had its own view of the world. The issue was that much of the commercial meaning inside those systems had already been shaped upstream. Source logic. Attribution language. Campaign taxonomy. Session assumptions. Acquisition metadata. By the time the data reached downstream tools, a great deal of the interpretation had already been inherited.

And because most organisations were siloed, very few stopped to question it. Each function worked with its own numbers. Each department had its own priorities. Then as businesses moved from fragmented local systems toward cloud platforms, unified reporting, and some version of a master data file, the focus was on consolidation, consistency, and usability. Not on asking whose logic was already sitting inside it.

The question I keep coming back to is not whether the data was useful. It often was. It is how so much of the grammar through which acquisition, attribution, and performance were understood ended up being inherited from the same place. UTM conventions, source and medium taxonomy, channel grouping logic, session models, conversion event definitions, attribution defaults. These became the shared vocabulary of digital performance not because someone mandated them, but because Google Analytics was free, well documented, and already integrated with the ad platform most of us were buying from. Every sensible infrastructure decision pointed in the same direction. So we followed it.

That grammar then became normalised in a way that is easy to miss because it looked practical, not ideological. Over time it stopped feeling like one company’s framework and started feeling like the natural language of digital success.

And then it travelled.

It went into the CRM through lead-source data. Into email platforms through lifecycle segmentation built on acquisition metadata. Into finance through CAC calculations and ROI narratives. Into executive reporting through dashboards nobody really questioned because the numbers moved in roughly the right direction. Into product teams through audience definitions that might start in Google Ads and end up as the operational definition of what a high-value customer looks like.

That was part of what made it so hard to see. Nothing about it felt dramatic at the time. It felt normal. The agency used the same terms as the client. The dashboard matched the campaign report closely enough. The CRM inherited the source logic without anyone stopping to ask where that logic had originally been set. It all held together well enough to keep moving. And in most organisations, well enough is more than enough for a system to become permanent.

There was also a more practical reason this took hold so deeply. For much of the last fifteen years, digital teams were expected to drive growth without being resourced like core infrastructure functions. They were often working with smaller budgets, less internal leverage, and a constant need to prove value. They did not need elegant theory. They needed tools that were free or cheap, easy to implement, easy to connect, and credible enough to support decisions. Google’s stack fit that reality almost perfectly. Analytics was free. Tag Manager was free. Search Console was there. The documentation was there. Ecommerce platforms and CMSs integrated with it. Agencies knew how to work with it, and those that did not often recommended a move in that direction anyway because it was the quickest and cheapest path toward something that looked like a unified data model. The pitch was usually some version of the same thing: everyone uses it.

What looked like a sensible operational choice at the time gradually became something deeper: the measurement grammar through which digital success was described.

By the time most organisations had a proper data stack, Salesforce and HubSpot and Klaviyo and BigQuery and whatever reporting layer sat on top, the Google logic was already load-bearing inside it. Not as a product. As an assumption. As the inherited grammar of how commercial performance gets described and understood. And because so much of the source data across departments stemmed from the same underlying definitions of acquisition, attribution, and success, the diversity of platforms was often less real than it looked. Different interfaces, different teams, different dashboards. But underneath them, much of the engine was still being shaped by metrics and logic first normalised elsewhere.

I do not think this was a plan. That is actually what makes it interesting rather than scandalous. Nobody at Google drew a diagram in 2008 and said in fifteen years we will own how businesses understand their own customers. It compounded. One reasonable tool choice at a time, across millions of organisations, across two decades. Each individual decision made sense. The aggregate outcome is something most people still have not properly looked at.

Then privacy regulations started to arrive, GDPR most visibly among them, and everyone exhaled. Finally, regulation. Finally, some rebalancing. Finally, users getting back control over their data and businesses being forced to think more carefully about what they collect and why.

Except that is not quite what happened in practice.

The compliance burden was real. The operational pressure was real. Most companies did not respond by reinventing measurement from first principles. They responded the way organisations usually respond under pressure. They reached for the tools that were already closest, most documented, and easiest to operationalise at scale. In a lot of cases, that meant going deeper into the same ecosystem they were already trying to reduce dependence on.

The consent infrastructure that emerged was often built to remain compatible with the same dominant platforms businesses were already relying on. The measurement workarounds for a more restricted, more regulated, more cookieless environment were often framed through Google’s proposals or through logic Google had already helped make normal. The first-party data strategies that compliance teams urgently needed were, in many cases, still being implemented through tooling, standards, and vocabulary shaped by the same ecosystem.

That is the part I find people still have not really sat with. The legislation was supposed to reduce dependency and force more deliberate data practices. In some ways it did. But in practice it also increased the value of scale, engineering depth, and operational convenience. And when those pressures hit, the companies best placed to make the new world manageable were often the same companies that had done so much to shape the old one.

The cure and the disease share a landlord. I am still not sure most people have looked at that directly enough.

This is also where the AI conversation gets more complicated than much of the current discussion suggests.

One assumption now doing the rounds is that AI interfaces will erode Google’s position because they intercept search intent before it becomes a Google query. That may happen in some categories. But the measurement dependency does not live in the search page. It lives in the infrastructure layer underneath commercial decision-making. If that infrastructure remains anchored to Google’s ecosystem, which it largely does and will for some time, then the centre of gravity does not move as much as the disruption story implies.

A world where fewer sessions start with a Google search is still a world where many organisations are reading performance through a Google-shaped interpretive framework. The synthetic layer changes the discovery surface. It does not automatically change the measurement language. And the measurement language is where the dependency actually lives.

I am not writing this as a complaint. Google built genuinely useful things. The convenience was real. The standardisation had value. I have spent a significant portion of my career making these systems work for organisations that needed them to work.

But I have also spent enough time inside these operating models to know that the data we call ours is not fully ours in any meaningful sense. Not because every dataset came directly from Google, but because the grammar through which success was measured and communicated had often already been set upstream. The definitions came preloaded. The segmentation logic inherited assumptions we did not consciously choose. The customer understanding we built strategy on was always, at least in part, one company’s infrastructure reflected back at us.