Data sampling is a hot topic when it comes to Google Analytics. Especially for websites with huge data sets (think of ecommerce with hundreds or even thousands of products), using the free version of Google Analytics often limits us to only be able to see a fraction of the data. 

What is sampled data?

Data sampling is a statistical analysis technique to identify certain trends, insights or patterns in a larger data set. In practice, sampled data is a small subset of your total data in order to get a glimpse of the full, larger, data set. 

Google's Universal Analytics generally uses data sampling for faster response. This will be used for both default and custom (or ad hoc) reports. When data sampling happens, GA takes a small fraction of the number of sessions and enlarges this fraction to the entire population.

The exact threshold is 500k sessions for Google Analytics Standard (or the free version of GA), and 100M sessions for GA360. However, we at Semetis have experienced this can be fewer when the complexity of your report intensifies, for example when adding additional dimensions, filters or segments.

This doesn't sound like a big issue initially, but the bigger your data set becomes, the bigger the discrepancy between your sampled data and your exact data becomes. When you need to report on exact metrics like Revenue, this can become problematic.  

How to avoid data sampling

In practice, there are two ways of avoiding data sampling:

1. You go for GA360, the paid version, and you practically never face data sampling.
2. You use third-party tools that work around data sampling.

These third-party tools (like Supermetrics or Funnel.io) break down your entire data set in smaller parts of your data set. Supermetrics breaks down your entire query into separate "subqueries", while Funnel.io continuously imports your most recent data.

Unsampled data in DataStudio

Most of the time, when using Google Analytics in DataStudio, we use a Google Connector. This is the native integration of all Google tools with DataStudio. This native integration also exists with Google Ads, DV360, Campaign Manager etc. 

When it comes to data sampling, Google DataStudio faces the same limitation to data sampling as Google Analytics does in-platform. DataStudio can "show" this data sampling as well. You can read more about that here. 

But, as explained above, third-party tools can work around that. The most straightforward way is to use the Supermetrics tool for Google Sheets, via which you can check a box to avoid data sampling in your query. 

When it comes to Google DataStudio, there is a Supermetrics Google Analytics connector. When adding Google Analytics to your report as a data source, you'll see you can select to avoid data sampling, just like you would do in Google Sheets.

You'll even be able to modify this with every chart you add to the report, under "Parameters". 

Bonus: via the Supermetrics connector you can also add some reports you aren't able to add via the Google Connector, like Multi-Channel Funnels to report on Assisted Conversions.

{snippet robbedesmyttere-en}

Why should I implement the Facebook Conversion API?

Facebook is pushing the market and the agencies to implement their Conversion API, but why? In this article, we will go through the advantages that it can bring to your business and how we can help you put it in motion. 

Before we go down to the set-up, it is important to reflect on what it is and how it can bring value. The reasons come from different corners, but it is mainly tight to the expected (3rd party) cookieless world that is ahead of us:

1. Browsers are becoming less reliable for conversion and traffic tracking due to the different privacy measures taken, ITP being the most popular one.
2. Regulators are putting - with reason - more and more pressure on the market to respect the user consent and privacy.
3. Ad Blockers keep growing in popularity and importance and can cause tracking issues.
4. iOS14 is the last example of privacy solutions built by technological providers (Apple in this case) that will challenge the current digital marketing operations.

Those different elements will for sure make your tracking imperfect and you are probably short of around 10 to 20% of your actual orders, for example, between your Facebook transaction data and your actual sales data. If those figures scare you, we’ve got you covered. Make sure you go through our vision on the future of measurement.

The good news is: using the Conversion API will allow you to improve the accuracy of your measurement while respecting the consent preferences of your customers. As the Conversion API works server-side, you should be able to cover for most of the data lost due to ther points mentioned above. Here is an example of the gap that the Conversion API is able to fill:

Another great reason to start getting familiar with the Conversion API is the ability to measure offline conversions. Indeed, where your pixel is limited to the measurement of your online selected events, the Conversion API allows you to go further than. Some examples of offline events are in-store sales, phone calls, qualified leads, etc.).

How does the Facebook Conversion API work?

The first important thing to mention is that the Conversion API is not a direct replacement of your existing pixel, but rather a complement. It will probably become the replacement of the pixel but Facebook advice to keep both of them running in parallel for now.

But then everything will be duplicate? You might ask. 

If set-up correctly, meaning if you send the right information to Facebook, every event should be deduplicated, except for the ones that the Conversion API was able to capture but not the pixel. This is precisely where we intend to remove the 15% gap in tracking mentioned above. Here is a representation of how it looks like:

In order to deduplicate the events sent via the pixel and the Conversion API, the more information you can share the better, but two are mandatory:

1. The event ID must be identical. For a purchase event for example, use the order ID.
2. The name of the event must perfectly match.

Some other elements such as the timestamp or user data (that need to be hashed) can be used to maximize the deduplication of the events you want to track. 

How can I implement the Facebook Conversion API?

There are typically three different implementation methods:

1. A direct implementation by your in-house developers.
2. A direct integration with partners such as Shopify.
3. A server-side integration with the appropriate tools (server-side GTM, CDP, etc.).

The advantage of the first option is that you have full control over the tracking in-house, but it also means you need to allocate the dedicated resources for implementation and maintenance of the tracking. In order to make that work, you’ll also need to make sure you have the right processes internally to make sure your tracking remains stable and reliable.

The second option has the advantage of being super easy. The Shopify integration, for example, is a flick of a switch. It is a great first step but the biggest constraint is that you are limited to the measurement of the transaction event (for the moment at least). It is therefore a good option if you want to test and measure the impact it can have on your measurement before implementing a full-fledged solution.

The last option, our preferred option, is to use third party tools such as the server-side Google Tag Manager (or sGTM). The biggest advantage will be to manage not only your Facebook events, but you can gradually start managing all your different tags server-side, which comes with a lot of benefits. Another benefit is that you will be able to measure all the events you want to track. Knowing that those events will be limited to 8 due to the last iOS14 update, you want to make sure those events are measured correctly.

If you want to go for the third option, get in contact with your dedicated semetis team or reach out to us.

{snippet gregoirelehardy-en}

This is the second article in a series of 7. Here we are building further on the article ‘Cookieless Series Part 1 - How ITP & GDPR are affecting your marketing results’, where we’ve set the scene on today’s constraints originating from ITP and GDPR. We will outline our vision on the future of advertising and how you should tackle it as an advertiser. The next articles in the series will go deeper into each of these topics separately.

The point-of-no-return

In the first half of 2020, Google made an announcement that created suspense among many advertisers. They are gradually working on a plan to phase-out 3rd party cookies from their browser, echoing the earlier decisions of both Safari and Firefox. With Chrome holding over half of the market share, we can therefore mark 2022 as a point-of-no-return. But let’s embrace the change. For all of their uses, third-party cookies have been an imperfect tool for marketers, leading to several recurring problems. As cookies are constantly deleted and there’s an inability to use cookies between devices, impressions are typically overstated and conversions undercounted. This leads to wasted ad spend and lost time in incremental frequency capping. Our way of thinking thus needs to pivot significantly.

Take action today by leveraging more first-party data

Let’s move forward:  what can your business do now in order to avoid a total cookie-disaster?

1. Collect first-party data: In the new marketing landscape, marketers and advertisers need to link and activate their customer data to more persistent IDs such as email, phone numbers, and customer IDs. This can be as easy as an email list coming from your CRM system, or more robust like a full Customer Data Platform deployment. The latter combines a wide range of data sources, both online and offline, in order to draw a 360° view of your customers. Consider the example of a website visitor who registered to your newsletter but is not yet a client. That person can be identified via his/her email, or any other unique id equivalent, and can already be mapped within your CDP. For a more detailed explanation, I suggest you read my colleagues’ article on ‘Why is the future of MarkTech Stack all about CDPs and not DMPs?’. One of the bigger challenges within larger organizations will be around data governance and responsibility. A strong collaboration between Marketing, IT, and Legal will be more crucial than ever before.
2. Give users an incentive to log in: In a world where everything is dependent on user data, we need the user to log in at some point. Only then we will be able to stitch that user’s behavior to his/her ID. So when redesigning your website, this concept always needs to be top of mind. It’s all about building better products.
3. Use Lookalike Audiences: as these are built 100% on first-party data and still leveraging the big data from walled gardens such as Google & Facebook. Lookalikes (or often called Similar audiences) are users that show similar traits to the people in your user base. They are created by smart algorithms and have proven to be very performant in the last couple of years. The more data, the stronger they get.
   Next to that, don’t forget that ITP has no impact on Contextual targeting, Geo targeting and device targeting as no personal identifier is needed to obtain this information.

Take action tomorrow by getting ahead on the following topics

We want to alert you on some advanced topics that will be pivotal for the future of marketing. Note that this is just an introduction. Each topic will get covered in more detail by a separate article in this series.

• Probabilistic measurement and conversion modeling: This is not new, but will become more important over time. Vendors such as Google and Facebook try to estimate how much you were not able to measure (using benchmarks, etc) and try to fill the gaps based on the data you were able to capture. In this way of thinking, there is no such thing as exact measurement anymore, but rather statistical correctness. Read more about this topic in the article ‘Cookieless Series Part 3 - Probabilistic measurement - A world where uncertainty is the new normal’
• Facebook Conversion API: This solution claims to safeguard data-driven marketing when the Facebook Pixel will become obsolete over time. It basically comes down to migrating our measurement from client- to server-side, making cookies redundant. Furthermore, it will enable us to measure data coming from offline channels and CRM. Read more about this topic in the article ‘Cookieless Series Part 4 - Facebook Conversion API’
• GTM Server-side Tagging: Server-side tagging means running a Google Tag Manager container in a server-side environment (f.e: Google Cloud). The concept already exists for a while on other tech vendors (such as Tealium) but Google might open the door to the broader public. To put things simply, this initiative is conceptually similar to Facebook’s Conversion API but instead applicable to your full existing tag structure. This solution is still in its infancy, but we expect it to grow a lot throughout 2021. Read more about this topic in the article ‘Cookieless Series Part 5 - GTM Server-side Tagging’
• CRO Impact and AB Tests: Today cookies are also leveraged by AB testing tools. What will the impact on those tools be in the future? Are there alternative solutions at hand? Read more about this topic in the article ‘Cookieless Series Part 6 - The impact on CRO and AB Testing’
• Google’s Privacy Sandbox: The Privacy Sandbox project’s mission is to ‘Create a thriving web ecosystem that is respectful of users and privacy by default.’ Their main challenge exists in finding an alternative for third-party cookie-based conversion measurement and targeting (f.e: remarketing). Important acronyms such as FLoC and TURTLE-DOV might grow to unorthodox solutions to improve tomorrow’s digital world. Read more about this topic in the article ‘Cookieless Series Part 7 - Google’s Privacy Sandbox’

On an ending note: much more is currently at stake between DSPs, SSPs, and DMPs regarding a cookieless world. We hear terminology such as ID consortiums, User-based tracking, First-party bitstreams & The Matching Principle. Currently, we can only agree that the future of ad tech is uncertain and will get clearer in due time. From our end, let’s focus on what we can change today: the way we treat first-party data and our mindset.

{snippet fredericpalella-en}

Privacy took a swing at measurement

The world of digital marketing is under scrutiny. For decades marketers, brands and tech vendors have been tracking internet users and their behavior without any form of transparency. Which data was collected and what it was used for was none of the user’s business. All of this is changing. Both governments and private actors are stepping in in order to protect the privacy of internet users. Governmental regulations such as GDPR (EU) and CCPA (California law) are impacting how personal data can or can’t be used. In the private sector we see initiatives such as Intelligent Tracking Prevention (ITP) developed by Webkit and deployed through Safari (Apple). ITP is protecting user privacy by restricting how cookies can be used for tracking. 

All of this means that users are more conscious about how their data is being used and that these practices are more regulated than they used to be. But how does that impact the way your data appears in your marketing platforms, and how does this impact your KPI’s?

Here’s why your KPI’s are shifting

As marketers we’ve been using something called deterministic measurement. Deterministic measurement is -- technically -- described as a measurement model where no randomness is involved. Everything that serves as an output was actually measured and used as an input. In the world of marketing deterministic measurement can be explained as simple as taking only the data into account that has actually been measured. And -- as a marketer -- you might understand, this is where it becomes both tricky and interesting.

KPI’s are based on measurement. In order to be allowed to measure, collect and handle data users have to provide consent. Users who do not consent are not being tracked. This impacts deterministic measurement as instead of tracking 100% of the population, we’re now tracking only the part of the population that has opted in. This means that KPI’s will oftentimes be impacted negatively. However, we’re still spending the marketing, communication, media and operational budgets. As a result your KPI’s get impacted. But are they really? It’s the story of a tree falling in the woods but nobody's around, does the tree still make noise when it comes crashing down? Similarly you can ask the question -- If conversions aren’t being measured in the context of consent, does that mean conversions didn’t happen? If conversions can’t be measured, but they’re still happening, do they really impact my KPI’s? Using a deterministic approach requires you to answer -- yes, your KPI’s are shifting. What can’t be measured, can’t be reported. So does this mean all KPI’s are going downhill from here?

Enter probabilistic measurement

Not really. Tech vendors such as Facebook and Google are trying to provide clarity on what’s really going on with your KPI’s. They do this by introducing probabilistic modelling on top of deterministic measurement models. In short -- The vendors try to estimate how much you were not able to measure (using benchmarks, etc) and try to fill the gaps based on the data you were able to capture. They do this by looking at your deterministic data and extrapolate from there. This is where elements such as Google’s Consent Mode come in play. By creating an overview of the proportion of what can’t be measured and what can and by combining this with deep analysis of the measured data a full image can be constructed. The objective here is to help the entire advertising and marketing ecosystem as a whole. Users should have a good experience, marketers should still be able to optimise their campaigns and publishers should be able to have control over their inventory to match the users with the right content. In the end this is what allows the internet to be (at least for the most part) free to users. 

A new mindset

All of this requires a new mindset. 20 years ago everything was done on either gut feeling or market research. Today many marketers won’t make a move unless data is 100% accurate and checked 15 times. Tomorrow we’ll have to let go of the obsessive behavior. Three years ago we wrote an article called “My web analytics data is not perfect, and I’m OK with it”. Today that idea needs to become part of the mindset of marketers. You should still strive for qualitative and perfect data, but in the context of measurement you have to be OK to take decisions knowing you don’t know everything. Otherwise you might as well stop your measurement activities all together. This evolution spells interesting times in the world of marketing. Elements such as attribution will need to move from an almost science to a more high level and global analysis form. The industry is changing and tech providers are trying to help marketers out. The real question is -- Can marketers change the way they are at ease with how their data is evolving?

{snippet glennvanderlinden-en}