Here’s the uncomfortable truth about your reporting stack: you have less observable data than you did five years ago, but your dashboards look more confident than ever.
That gap is the subject of a sharp piece in Search Engine Journal on attribution and false precision. The core argument: a growing share of what shows up in your reports isn’t measured at all. It’s modeled — statistically reconstructed from patterns. And knowing which is which is, as the article puts it, the difference between a good decision and confidently making a bad one.
Signal loss isn’t one leak, it’s a hundred
Signal loss doesn’t arrive as a single event. It accumulates: consent choices, device switching, platform restrictions, and the gaps between systems that never quite talk to each other. Most teams only notice when the numbers stop adding up.
The SEJ piece uses a journey most of us recognise. Someone hears your brand on a podcast. Later they search the brand name from a work laptop. They read a couple of articles, get retargeted on mobile, come back via direct, and convert.
Now answer honestly: which part was discovery? Which part was persuasion? Which was just the last thing your tracking could see?
Attribution systems see disconnected fragments. The podcast is probably invisible. Some of the research gets bucketed as direct or organic. The retargeting click, the cheapest and easiest touchpoint to observe, walks away with the credit.
The real cost: defunding what works
This is where measurement error becomes a P&L problem. When upper-funnel activity appears to contribute nothing, teams cut it. The decision feels rigorous and data-driven. In reality it may just reflect the limits of what the measurement system could see.
If you’ve ever killed a brand campaign, a podcast sponsorship or an influencer test because “it didn’t show up in the platform,” you’ve made this mistake. Everyone has.
Modeled data solves one problem and creates another
Platforms responded to observability gaps with modeling. Google and others have baked machine learning into measurement: when the direct link between an interaction and a conversion can’t be observed, the system estimates the missing piece from observable and aggregated patterns.
Those estimates then flow into reporting, attribution, bidding and optimisation. Useful — but the interface presents modeled and directly observed outcomes with the same visual confidence. Same font, same decimal places, wildly different epistemic status.
The practical rule from the article: treat anything labelled modeled or estimated as directional, not definitive.
Stop hunting for the one correct model
There is no correct attribution model waiting to be discovered. Every model answers a slightly different question. The interesting information sits in the comparison.
- If a channel looks important across several models, that’s a strong signal.
- If its contribution vanishes the moment you switch models, that’s also a signal — of fragility, not truth.
- Disagreements between models are questions to investigate, not errors to resolve by picking a winner.
Triangulate. Look for convergence. That’s the job.
When GA4, your analytics tool and your CRM all disagree
The scenario every growth team has lived: GA4 says 150 conversions. Another analytics tool says 180. The CRM shows 120 new customers. Three systems, three realities.
That usually isn’t a data quality bug. GA4 might count purchases while the CRM counts approved customers. Ad platforms may include view-through. Refunds might be stripped from one system and not another. Date-of-click and date-of-conversion reporting differ.
The recommended fix is a reframe. Start with the system closest to the actual business outcome — CRM, order records, subscription data — and use analytics and ad platforms to understand the journey around it. Backend systems aren’t perfect attribution sources either; they carry missing acquisition fields, overwritten data and duplicates. Their value is confirming that the outcome genuinely happened.
The question shifts from “which platform reports the most conversions?” to “which outcomes actually occurred, and which touchpoints appear consistently around them?”
What to do this quarter
First-party data collection is now non-negotiable — for measurement quality as much as compliance. Server-side tracking improves reliability and control, but it won’t erase consent gaps or recreate interactions you were never permitted to observe.
That’s still progress. It moves you from “we don’t know what happened” to “we have a reasonable picture with known blind spots.”
And be explicit with stakeholders about what’s measured versus estimated. It feels risky. It’s safer than anchoring strategy to false precision. Attribution’s job isn’t ground truth — it’s reducing uncertainty enough to make a better call.
Source: Search Engine Journal
