Measuring Properly with Google Analytics 4

Mastering GA4 Explore workspace, attribution reports, and cohort analyses for strategic growth.

Speakers: Zafer Yıldız, Ferit Berkay Kelecek

23 Jan 2024

Video Transcript

⏱ 21 min read• 3,656 words

Read the full speech or switch languages

— Hey everyone, welcome to Zeyot Walks. Today, Ferit and I are discussing analytics studios. We’re both on the analytics team. He’s a Web Analytics Specialist. I’m a Web Analytics Manager. We’ll chat about common mistakes we see with new clients. We’ll cover property optimizations and chronic issues. Welcome, Ferit. How are you doing?

— Thanks, Zafer. I’m good. How about you?

— I’m great, thanks. Let’s dive in. When we get a new Google Analytics 4 property, what’s our process? Universal Analytics is gone. We’re living in a Google Analytics 4 world now. Brands come to us for analytics services. They’re usually high-traffic, major brands. We see a lot of issues in this industry. What’s at the top of your roadmap? What do you spot first? That’s how we can kick things off.

— Sure. When brands first reach out, they usually have high traffic. They aren't beginners. They’ve already done SEO or PPC. They come to us specifically for Analytics services. As you said, they have high traffic. We follow a strict audit checklist. The biggest issue I see right away? They haven't added GTM and Google Analytics tags to every page. That’s where we find the most errors. We always start with the website.

— We check the foundation first. Before going advanced, is your Analytics even working? We often start a project only to find the base setup has been broken for years. We see this constantly. Sessions are tracked incorrectly. Even before moving to e-commerce or advanced concepts, these brands often had prior consultants. Yet, simple things get missed.

— Exactly. We’ve seen this even with companies that had previous Analytics consulting. Honestly, Zafer. It’s often easier to fix these issues for companies that never had consulting before. Companies with prior consulting often feel confused. Existing integrations are tough to fix. Data is already flowing somewhere. Telling a new client their data was wrong is hard. Pointing out specific tracking errors is tricky. Brands that used other services often have bigger issues. This is one of the biggest problems we face. Sometimes codes aren't on every page. The brand can't analyze their full site. Another common issue is duplicate page views.

Technically speaking, that’s my...

— Let me add that. It’s my area of expertise.

— Yeah, we always...

— Page view speaking.

— It usually happens to me by chance. Zafer always teases me about this. I’ll be debugging a page. I’m just looking at it. Then, a duplicate page view pops up. You can elaborate on this. Think of it as 2x sessions or 2x users. It totally skews your metrics. It ruins your bounce rate. It changes everything. When I see a duplicate page view, I’m stuck. It’s a massive problem. Missing codes is one thing. But duplicate page views? That’s a huge issue. It’s usually bad code implementation. Or sometimes, brands just forget and add it twice.

They double up on GTM or JForm'u.

— Or they have parallel setups. Like GTAG and GTM running together. Page view is the core event in Google Analytics 4. If that duplicates, all sub-parameters duplicate too. You’re firing a ton of events under one page view. Everything, including sessions, starts double-counting. It’s a fundamental setup error. A bad integration. We have to tell them. We say, "Look, your tracking is wrong." They ask, "So, what is our real data?" We tell them, "We don't know. We have to fix it first."

— Fixing this takes a month. You know how it goes. I remove duplicate page views today. Your sessions drop. Will they bounce back in two days? Or drop further? We can't predict your monthly trends yet. It’s a new brand. I feel embarrassed, even though I didn't cause it. I’m the one pointing out the old mistakes. It feels like it’s my fault because the issue is so huge. They ask, "Was all our past data wrong? Was all that ad spend wasted?" You have to say, "Yes, it was wrong. Nothing we can do."

— I don't feel embarrassed anymore. I didn't make the mistake. But here is the lesson for our listeners. Even the smallest, most basic error ruins your entire measurement. We see this with massive brands. I mean, million-dollar companies. Revenue-wise. The whole team is just stunned. Remember that client meeting? They were working with someone else. We showed them the data. They were shocked. Exactly.

— This is a household name in Turkey. A major institution.

— Yes, we showed them. And how did it happen? They just took our training. We were running a session. We spotted these errors while using examples. There wasn't even an audit involved. I’m not saying we’re perfect. We make mistakes too. We miss things and fix them quickly. But when the error is fundamental, things get complicated. Sure, we have "how did we miss that" moments. We fix those too. But this was a basic error. A brand paying for Analytics consulting shouldn't have this. That’s the real issue. That’s why we’re surprised or critical. We usually catch this during an audit.

It’s not just that. We’ve been talking about page views for a while. E-commerce sites track revenue. They track organic, paid, and social traffic. They see it all in Google Analytics. They track sales and product views. But e-commerce setups are often broken. I’m not talking about Inanç, but general e-commerce setups. There are simple, high-impact errors. For instance, revenue is often miscounted. Shipping isn't counted. Discounts are missed. Taxes get lumped into total revenue. Plus, there’s the transaction duplication issue. You actually caught this on one of our accounts.

The transaction ID must be unique for every order. Right? It’s usually a random number. If it appears more than once, the order is being duplicated.

— If you see the same ID twice, you likely have a coding error. It’s impossible to push two identical IDs at once. Even if a user buys the same thing twice, the IDs shouldn't clash. If you see both, there’s a setup error. This was on our agenda anyway. It’s like this. You set up JForm'u correctly. You set up purchase on Inanç.com, but it can still fire incorrectly. It’s a fundamental issue. When we built inanç.com, we set up purchase tracking first. We thought, "Get this working, then handle the rest." That’s actually a huge mistake.

— Let me add this. You mentioned setting up purchase first. We follow a specific roadmap for e-commerce integrations to avoid errors. We usually break the process into phases. When you document and send all events at once, the dev team struggles to act. Even if they do, you can't be sure it's accurate.

— Yeah, it takes forever.

— Exactly. It also makes it harder for us to verify their work. So, we start by setting up purchase. It’s the most critical event. We track revenue and simulate the flow. The brand gives us a test user or we use a dummy account. We place orders just like a real customer. If they can't provide a test user, we buy a cheap item. You know I’ve had products show up at my house that I couldn't cancel. We place test orders. We ensure the data layer, GTM, and Analytics debug mode are perfect. There are many scenarios here. Cash on delivery, wire transfers, and credit cards must work. All parameters must fire correctly.

Parameter values need the right data types. Don't send strings if you need numbers. Ensure strings are actually strings. There are many sub-parameters. Breaking this into phases makes our job easier. It ensures the process is finalized correctly. What’s next? Purchase is working. It starts tracking revenue immediately. This is great data. You see exactly which products sold and your total revenue in Analytics. Purchase is the foundation for most reports. Then we handle add payment info, add shipping, add to cart, and remove from cart. We document these and send them to the dev team. They usually set it up quickly. If they’re new to this, we guide them. We hold meetings and share Google resources.

It’s not an easy setup. If you’re doing custom work in Analytics, it needs heavy testing. That’s why I recommend moving in phases.

— I agree. Let’s move to the next question. We’re talking about purchase tracking. It’s also important to track sales from ads. But what if your bounce rates are through the roof? Three or four percent. This is a common issue I see. Brands have no clue what to mark as a conversion. I see this with 60 or 70 percent of brands. I’ve seen session starts marked as conversions. Every user event is counted as a conversion. Imagine a huge banner on the homepage. They mark the banner view as a conversion.

Every user triggers that event. You’re counting it as a conversion. Naturally, your bounce rate and other metrics get skewed. This is one of the biggest mistakes we see. Feel free to add to this. I’m constantly flagging this to brands.

— Exactly. You should limit your conversion definitions. The most basic one is purchase. You can't remove that. For e-commerce, a purchase is the only real conversion. It’s simple. That’s enough. Purchase is really all you need. If you track page views or session starts, you can't get clean data without filtering. You end up with a million conversions. It’s not healthy. It even hurts your ad performance. Advertisers have specific targets and campaign learning phases. Multiple conversions can confuse the algorithm. They often ask us which conversion to prioritize. Yet, we haven't even defined those conversions properly.

There are just too many conversions. What should happen? Outside of e-commerce, take insurance brands. They might have 10 or 15 different conversions. Think health insurance, auto, mobile, or earthquake policies. These are all distinct conversions. They must be tracked separately. We agree on that. You can fill that list however you like. But many brands mark default events as conversions. We disable those immediately during our audits.

— I have another point to add here. As we talk, it comes to mind. We check the domains. Which domains are sending data to JForm? This is one of the most common issues. Big brands usually use test sites, Zafer. They think, "We installed JForm on the main site. Let's add it to the test site too." They want to test data layer pushes and changes there first. Then they push those changes live. That is actually a solid approach. But we recommend separating the JForms. Developers often organize everything on the test site, which causes issues.

Then they start testing on the staging site. They test constantly. Over and over. Since they’re developers, they hit the site thousands of times a day. They close it and reopen it. They debug something and refresh again. What happens next? If you connected the same JForm, you get thousands of sessions. Thousands of events flood in. You get events from various subdomains and domains. You get tons of fake users. Then you segment this in custom reports. Say you have 5 million monthly sessions. 1 million of those came from testing. From the test site. The brand pauses. That 1 million sessions is basically trash.

It’s just developer traffic. These aren't your target users. We have to delete all of it. This is one of the core problems we see.

— It pollutes the data. We see this more in large accounts. If you have a team of five, it’s no big deal. The data pollution stays manageable. But huge companies generate massive amounts of test data. Many websites don't even have that much real traffic. It’s just insane developer volume. Like you said, they trigger session starts constantly. They link test site data to the main property. So, what do we do there? We use a hostname filter or block by IP. It was easier in Universal Analytics. Remote work makes IP filtering less effective now. Hostname-based filtering is a real lifesaver these days.

You’re spot on. For audits, we must ensure we’re measuring correctly. This is vital for high-traffic brands with many employees. We need to optimize those properties. With that, I want to move on to consent mode. It’s a hot topic.

— We can’t skip this.

— Exactly, it’s impossible. Consent mode was huge last year. It’s been a major issue in the EU for years. Some listeners might not know what we mean by consent. Let me explain. You see a banner on every site. Accept or reject. We rarely see "reject" on Turkish sites. That’s the real issue. It’s about measuring data while respecting user privacy. Once they hit reject, you lose access. No remarketing. No tracking. Well, maybe some modeling. It gets complicated. It’s not that simple. It’s not just about accept or reject. You need Google consent mode for tracking to work. Google has official partners for this.

There are also local CMP providers. You can build your own consent setup. You don't strictly need them. But a professional dashboard makes everything easier.

— It boosts your data quality. You see who accepted and who rejected. You see their devices and locations. Usually, rejected users are invisible. We lose a lot of data there. CMPs prevent bad setups, Zafer. If you're setting up consent mode alone, it’s risky. CMPs help you avoid mistakes. But let me add a caveat. Using a CMP doesn't guarantee zero errors. Zafer knows this all too well. Cookie management platforms can have hidden bugs. You might also make mistakes yourself. Maybe you misconfigured a setup. Or you chose the wrong parameters. You might have hit snags while setting it up in GTM.

Some errors are really hard to find. I’ll use one of our brands as an example. Zafer and I spent about two weeks tracking one down. Remember that, Zafer? We had a brand where direct traffic spiked after a consent mode setup. The brand had steady traffic. Whether it was direct or organic, the volume was the same. It matched the previous months perfectly. But the traffic kept flooding into direct. We eventually traced it back to the consent mode setup. I think I remember it right, Zafer. The page would refresh. After the refresh, it sent the user straight to direct. That caused the direct traffic numbers to explode.

— Yeah, the user ID kept changing. It was a tough bug to catch. Like you said, we checked everything. But tracking that specific issue down was a real headache. And like you said, cookie management platforms aren't perfect. We work with most of them in Turkey. But that doesn't mean they're error-free. Many brands use them and still make obvious mistakes. Sometimes, getting support for these issues is a real struggle. You need an Analytics consultant. Or at least someone in-house to debug. This field updates constantly. We had Consent Mode. Now we have Consent Mode v2. New parameters keep getting added.

Are these active for you? It matters. There is a legal side to this.

— A bad setup can cause real headaches for a brand.

— Exactly. It’s not just about bad tracking. There are legal implications too. That’s not our expertise, though. We focus on the technical parameters. We handle the setup. We respect user consent. We model data in Analytics using machine learning. This is crucial for certain properties. Say, 20 people visit my site. Some reject, some accept. You might have a basic Consent Mode setup. It’s still a setup. You’re respecting user choice. But what happens? You might lose half your data or more. Google Analytics 4 uses modeling behind the scenes. You can recover 70% to 80% of your data. Only about 20% will vanish. This gives you a solid sample for your analytics.

It’s great for measurement. That’s why setting up advanced consent mode correctly is vital.

— You need consent mode to prevent data loss. It also keeps you safe from future legal headaches. It’s a critical control point for any business.

— Exactly. Major e-commerce sites in Turkey, what are you waiting for? This is standard in the EEA. They need to respect user data. We’re all consumers at the end of the day. If I check a site for sneakers, I don't want them everywhere. I don't want your remarketing. Maybe I was just browsing.

— It’s just not a priority in Turkey yet.

— Yeah, exactly. Big sites need to get their act together.

— Penalties need to get serious, Zafer.

— Agreed. Penalties should be a real deterrent. Things are different in Europe right now. It's the top priority for all our international brands. They check if Analytics is working right from day one. Things work differently over there. We make sure consent works perfectly alongside Analytics. Beyond that, regarding consent.

— We move forward if there are no issues.

— We looked at faulty e-commerce tracking. I'm checking our topics. Adil and Kamil are here. Yes?

— Let me ask you directly. That’s a mistake too. Zafer, I have a Google Analytics 4 account. I’m a brand. Should I use 360? Or BigQuery? What’s the indicator? I see my data in Analytics. My reports are crashing. They’re full of red exclamation marks. When should I use 360? Or when should I switch to BigQuery?

— I think moving to BigQuery makes more sense. But it can get pricey for high-traffic sites. If I recall correctly, there’s a monthly export limit on the free tier. You need 360 to bypass that. Unless things changed, that’s how I remember it. For smaller sites, BigQuery is actually quite cheap. It’s not as expensive as people think. It might even be free. Google Cloud gives you credits. When you connect BigQuery, data flows right in. What does this get you? Simply put, you get full control over your data.

Instead of relying on Google Analytics reports, you can query BigQuery. You get unsampled data. You see the raw numbers without any sampling. BigQuery integrates with so many tools. You can import Excel data. Match it with your Google Analytics 4 data. Create blended reports. Export data from your CRM. Match it in Google Analytics 4. Run queries in BigQuery. Send that data to Looker for visualization. 360 is super expensive. It's pricey even for huge companies. That's why 360 isn't common. But BigQuery solves those "other" rows or sampling issues. Exactly.

— We're discussing this for a reason. Accounts hit "other" row issues. Or cardinality limits. This shows up as sampling. Don't confuse it with thresholds. Sampling causes real problems. Remember that one brand? We couldn't add anything to reports. Add a parameter or metric. The report immediately hits sampling. It says it can only show 15%. Or maybe 10%. Use 360 or BigQuery to avoid these headaches.

— Exactly. Let’s talk about AI. You can’t have a conversation without it today. How do we use AI in Web Analytics? As a consultant, what are we actually doing here? Let me start before passing it to you. We use it a lot for GTM issues. AI writes great code. We often need custom JavaScript for GTM. What did we do before? We aren't developers. We don't code every day. We’d hunt for scripts or search Stack Overflow. We’d waste time fixing errors in variables.

Or maybe a custom HTML tag. We’d spend hours debugging JavaScript. Now, we just explain the problem.

— You just generate it.

— Yes, it writes the code for us. It usually works on the first try. Or maybe after a few tweaks. It all depends on our prompt. Better prompts mean better results. It saves so much time. I used to spend days on code. I’d bug senior developers for help. I’d ask why things weren't working. Now, I just paste my code. I explain my goal. I mention the error or the missing value. It gives me the fix immediately. Most of the time, it just works.

— I’m sure you’ve been there. You work with a brand. They send third-party code. They want it in GTM for tracking. Sometimes that code just fails. I’ve struggled to push it to GTM. GTM throws errors. It says the syntax is wrong. You know how GTM errors are. It points to line 12. It doesn't explain why. I use AI for that too. I paste the error. I ask what it thinks. It fixes it. It tells me what I missed. It even suggests better ways to write it. Without AI, you’d have to wait. You’d ask the third party for a rewrite. That takes forever.

I rewrote it. Can you push it again? Checking for errors takes forever. AI lets you fix it and keep moving.

— I solved a complex task with AI. I handled a third-party setup by pulling data. You still need some GTM knowledge, though. You must understand the logic. If AI suggests something, you need to verify it. You have to catch its mistakes. AI still hallucinates and makes errors. Don't hand over the whole process. Use it as an assistant. Combine it with your own expertise. AI won't do the whole setup alone. But it's a powerful assistant. It knows more than you. You just have to use it right. Anything else you want to add?

— We covered the main issues users face. Not every single one, but the core problems. If you check these, most issues will be solved.

— Exactly. We’re very active with this content at ZO. Check out our Digital Zone or blog. We have plenty of Analytics content there too. I highly recommend following the ZO community. Let’s wrap it up there. Thanks to everyone for listening. We appreciate your time. See you in the next ZO Talks. Goodbye.

— See you later. Goodbye.

This transcript was generated using AI speech recognition and verified against subtitle style rules.

Yiğit Konur

Content author

Yiğit Konur

both

Contact us

How we handle your data

Open as a full page