Here is a situation that occurs in performance marketing teams more often than anyone likes to admit.
A campaign starts underperforming and ROAS drops by 17% over three weeks. The marketing team starts to take measures. They pause and relaunch ad sets, test other creatives or they argue whether the latest Google update changed something. Eventually they might escalate the issue to the agency.
Team members are busy and optimizing and then almost by accident, someone opens the browser console and checks the purchase event. It has been firing with value: undefined since a new site release three weeks ago. Three weeks of “optimization” was optimizing toward noise.
The uncomfortable truth is that many “performance problems” are actually not performance problems but they are data problems. If you can’t be sure that your conversion data is correct, you are not really doing performance marketing but you are mostly gambling with a very expensive dashboard.
Look into the mirror before you blame Performance Max
When the numbers drop external factors are often the first to be pointed to. Maybe Google rolled out an update, Meta’s delivery might be acting differently this week or the blackbox Pmax is acting in an unexplainable way.
Accusing the algorithm only makes sense if we’re sure that the data it is using is accurate. In reality however the data is not checked frequently enough to be sure. The team tends to overlook this crucial step and that’s a problem. No matter how sophisticated the algorithm, if the data is corrupted or incomplete, so will be the output. So it’s essential to regularly examine the data’s quality to ensure it’s reliable. By doing so, we can trust the algorithm’s results and make informed decisions.
So in daily practice, the team gets busy with the wrong things by adjusting bids, shuffling budgets and updating creatives. When was the last time you checked that your purchase event, your add-to-cart, your form submission and your enhanced conversions were all firing correctly, across every browser, device and consent state?
Why broken tracking is more dangerous than ever (and not just annoying)
There was a time when a broken tag was simply annoying and inconvenient. Your dashboard was wrong, you noticed eventually, you fixed it and you moved on. It was not a disaster. Those days are behind us now and it’s all because of artificial intelligence.
AI models completely depend on the data you feed them. The well known concept “garbage in, garbage out” (GIGO) is now more true than ever. The price of bad data has significantly increased. Where it used to be a misleading dashboard nowadays it is automated bid decisions made at machine speed against a false reality.
On top of the increased cost of bad data, issues are harder to detect. Think for a moment of the advancements from recent years which you find in any modern stack:
- Smart bidding and Performance Max optimize bids automatically based on your conversion signals. If those signals are corrupted the machine optimizes toward the wrong thing and you don’t see it happening.
- Server-side tracking improved data quality and reliability. This is great but also moved the point of failure further away from the marketer’s view. Things break in places you never look.
- Consent Mode keeps you compliant. This is of course non-negotiable but it also adds another layer where events can silently stop firing.
These advancements have a downside too. They make tracking issues more subtle and harder to identify.
Why a periodic audit says virtually nothing
A common response from responsible teams is “But we do QA, we have a tracking plan, we did a full audit last month and the dataLayer is documented.”
Tools like GTM preview mode and GA4 DebugView are fine for quick checks but not for keeping a live site healthy. They rely on someone clicking through manually, they only see your own session and they do not alert you when something breaks.
Regardless of the way you manually check your tracking, any periodical audit is like a snapshot. It only shows a moment in time, while continuous tracking shows you the whole story. Manual quality assurance is important, but in reality tags always break between audits, not conveniently on audit day. Moreover tags tend to break on device or browser segments you do not always test manually.
As a result in digital commerce we see more and more search queries asking “What are the best tools for automated tracking monitoring and analytics QA?” The increasing demand for solutions makes sense because manual QA stops being realistic once a site grows. Teams start looking for tools that audit tags automatically, run checks after every deploy and alert them the moment a critical event like add_to_cart drops to zero.
The comparison between manual QA and real-time monitoring looks like this:
| Manual QA / periodic audits | Real-time data collection monitoring |
| Point-in-time snapshot | Continuous, around the clock |
| Error detection in days to weeks | Error detection in minutes |
| Covers what you remember to test | Covers every monitored event, always |
| Nobody is watching at night or in weekends | Alerts fire 24/7 |
| High impact on capacity | No impact on capacity |
You are already monitoring four things. Likely you forgot the fifth
Most serious e-commerce operations monitor their stack carefully. They already keep a close eye on:
- Website uptime (downtime instantly negatively affects revenue, reputation and SEO).
- Server and database (when they’re slow or overloaded it causes errors and lost orders).
- Website traffic (gives insights into behaviour on the website which helps making informed decisions about user experience, content and media).
- SEO (because rankings, traffic sources and Core Web Vitals drive organic revenue).
Nobody questions whether these four pillars deserve monitoring. However there is a fifth pillar which almost everyone skips: data collection monitoring. It protects the integrity of your tags, your dataLayer and the events flowing into your marketing platforms. We wrote about the gap in one of our blogs “You’re monitoring everything except the one thing that matters most“.
To summarize, traditional uptime monitoring only alerts you when your website is down but it doesn’t notify you when your data is corrupted. As a result, data downtime can occur silently and you might only discover the problem when you receive a bad report, if you’re lucky enough to find out at all. This gap in monitoring can have serious consequences and it’s essential to address it to ensure the reliability and integrity of your data.
What an error actually costs: a real example
This isn’t just a theory, it’s a real-life issue that Goboony (together with its sister platform Yescapa the operator of Europe’s biggest motorhome rental marketplace) has faced.
A small change on the website caused a tracking error that caused a sharp increase in measured (not actual) conversions. The data said things were going great. They were not.
“If fake conversions go unnoticed, Pmax thinks they’re real and will try to build on that success. It will slowly increase budgets and bids to get more of those conversions, not realizing they’re fake”.
The problem is that such an error will not be obvious right away. It might take days to come to the surface and by that time potentially thousands of euros will have been wasted.
The opposite can happen too and it is just as costly. A conversion tag might quietly stop loading in certain browsers or on mobile. Pmax does not see those conversions so it thinks the campaign is doing less good and it lowers bids and budgets. In this case you do not see wasted spend, you just miss out on revenue you never knew you could have had.
Here’s what saved Goboony: a real-time alert by Code-Cube.io caught a tracking problem within minutes and it was fixed in just a few hours. This was crucial because it stopped the algorithm from making bids based on false information. Their website traffic and conversions are highest in the evenings and on weekends when no one is actively monitoring the data.
So they decided they needed a way to constantly keep an eye on the accuracy of their data. It turned out to be a great investment: just one incident that was caught and fixed ended up saving more money on ad spend than they paid for the entire year’s subscription.
How real-time data collection monitoring works
It’s not as complicated as you might think and pretty straightforward.
- You define what “correct” looks like. So which tags should fire, on which pages and with which variables (for example transactionId, value, pageType).
- The monitoring system is always on and checking how your tags, tag manager and dataLayer are behaving in real-time. It checks 24/7 their actual behavior against a predefined set of rules, making sure everything is working as it should.
- If something goes wrong, you’ll know it right away. If a tag stops working, a value is missing or incorrect or a script isn’t working well on specific browsers, you willl get an alert within minutes.
- You detect and solve the issue before it affects the algorithm. The purpose is to minimize the time between something breaking and the algorithm starting bidding on the false data.
Why monitoring your data collection pays off
Your ad spend will be protected as you will catch errors that otherwise will silently waste your budget. Automated campaigns optimize toward reality instead of false signals because your AI gets the clean data it needs. You can have peace of mind and confidence in the decisions you make. Debugging errors will be faster than ever before because clear alerts point to the actual problem. Fixing errors now will take hours instead of days.
This isn’t just another analytics tool, it’s a must-have for any organization. You wouldn’t think of running production infrastructure without uptime monitoring and the same applies here, it’s a fundamental component that helps you stay on top of things.
Conclusion
Data collection monitoring is not something to put off until later. In e-commerce and online advertising we see data collection monitoring becoming standard practice. Because of the clear added value and strong business case, within the next one or two years, data collection monitoring will likely become part of any professional marketing stack.
So before you blame the algorithm again, before you rewrite the creative one more time, make sure you continuously check the data underneath all of it. You probably do not have a performance problem. You have a data problem which you have been calling a performance problem.
About the author
Harm Linssen is Managing Partner at Code-Cube.io. a full-stack observability platform that protects your dataLayer, tags and conversion data.
Code Cube BV is located at Grebbeberglaan 15, 3527 VX Utrecht and can be reached by phone at +31(0) 85 747 04 11 or by email at hello@code-cube.io.