The privacy changes facing digital analytics

This week’s “week in review” is a two parter. For Part One, enjoy my recap of “The privacy changes facing digital analytics”. Next week we talk solutions. No chapter this week as I’m working on consolidating the chapters to date in a fun recap on the blog. Stay tuned. 
 
Working in technology, you get pretty used to click-baity headlines that suggest “This [insert upcoming privacy change] is coming soon, beware!”
 
It’s a byproduct of the algorithms we’ve lazily come to rely on to serve us up content we’ll actually click on (which usually all have click-baity headlines because that’s what we click, the toxic cycle continues…). 
 
Of course, anyone who is close enough to the jobs to be done knows that it’s truly never as simple as something being dead, nor something being the best thing since sliced bread. 
 
There really is no such thing, and I’m sorry, but I now have to say the one thing everyone hates to hear…. 
 
“The truth is it depends”
 
But it does.
 
 
Here’s what it depends on. 
 
It depends on your use case; the specific question you are asking and the decision to be made.
 
It depends on your people, their skills, their appetite to learn, to adapt to change, the agendas they are all uniquely pursuing that may conflict with one another.
 
It depends on your processes and how well people can adhere to the process that was defined and where consensus was reached (how many people in your business do you KNOW adhere to their UTM tracking links in the way we all agreed to?) 
 
It depends on your technology and its capability to deliver on your needs. 
 
Finally, of course, it depends on your data. It’s configuration, quality, completeness and accuracy. 
 
So with that disclaimer now announced, let me share that this week, after receiving the recent news that Apple’s will soon be stripping identity parameters in Safari – my anxiety peaked yet again. Call me triggered. In short, this impacts some of the granularity we can otherwise generate to conduct customer journey analytics at an ad level (linked to a customer profile) and to collect browsing behaviours to a user ID. As one of my colleagues put it, it’s not the end of the world but it is still yet another reminder of the path we are all still on to seeing larger and larger gaps forming within our previously easy-to-track digital world. 
 
Now, as we’ve covered above, shouting “beware” is often a gross exaggeration. Our digital analytics will still be useful though as changes that are made to protect privacy continue to progress. We do however need to consider the limitations of the data we are able to access in our sample to ensure they are representative enough to offer a fair interpretation of channel and omnichannel interactions. This sample must be representative enough to afford us the ability to meaningfully optimise based on within-channel granularity too (that is, spend between campaigns, ad types, audiences targeted, placements, creative impact and more). 
 
When announcements like this one (no matter how small) continue to add to the privacy protection timeline marketers can take another little hit on that underlying sample and the granularity – and we still know that there will be more to come and the ask of analysts will be to keep abreast of the changes and understand their impacts to their reports so we can be better informed. 
 
So the question I pose today is, are you informed? 
 
Today’s newsletter offers a quick history lesson and a reminder on some of the major events in tracking history. Then we can pull apart how these moments have impacted the world of analytics as we know it. 
 
 
The Privacy Timeline 
 
1994 -Netscape Navigator introduces HTTP cookies: Netscape Navigator, one of the earliest web browsers, introduces HTTP cookies as a way to store user data locally. This allows websites to track user activity and personalise their browsing experience. 
 
Late 1990s to early 2000s – Widespread cookie adoption: Cookies become widely adopted by websites for various purposes, including user tracking, ad targeting, and session management.
 
August 2010 – The Wall Street Journal publishes a number of articles bringing widespread attention to tracking on the internet and related privacy concerns 
 
February 2011-  Mozilla Trailblazes introducing the “Do Not Track” (DNT) header, allowing users to express their preference not to be tracked. Microsoft follows in March of 2012. 
 
April 2013 –  European Union’s ePrivacy Directive comes into effect, requiring websites to inform users about the use of cookies and obtain their consent.
 
January 2017 – Apple introduces Intelligent Tracking Prevention (ITP) in Safari 11, significantly limiting cross-site tracking by reducing the lifespan of first-party cookies and blocking third-party cookies.
 
2019 – Mozilla turns on it’s Enhanced Tracking Prevention
 
Jan 2020 – Google Chrome announces the Privacy Sandbox initiative, an effort to develop privacy-preserving alternatives to third-party cookie
 
April 2021 –  Apple rolls out the App Tracking Transparency (ATT) framework with the release of iOS 14.5
 
June 2021: Google delays its plan to phase out third-party cookies in Chrome, extending the timeline for its implementation with no market 
 
June 2022 – Mozilla strips clickID Facebook, Marketo, Olytics, and HubSpot
 
June 2023 – iOS 17 strips tracking link parameters of all customised detail
 
Today’s Context

As of today, Similarweb provide the following breakdown of browser share of market. 
 
Image item
 
 
Though it’s been crying wolf for years about progressing its privacy standards, Google is yet to find a solution that maintains its (and ours) insatiable appetite for data granularity that supports the most profitable arms of its business. It has announced however that it will at the very least, start testing what the impact will start to look like (for them and for us no doubt) very soon. So although we’ve been saying it for some time, change is still coming…
 
Image item
As of June, 2023 via Oberlo 
 
So what do all these privacy changes mean for the analytics world? If you’ve ever heard me bang on about this in some sort of speaking opportunity, I’m always reminding everyone as the ultimate level-set that all data is sampled. It might seem obvious but I do find that most people do forget that. 
 
This notion that we could ever get close to tracking every interaction and stitching them into one, consolidated journey is a lovely utopia to aim for though of course we know it’s not ever exhaustively likely. 
 
What about offline channels? 
What about word of mouth? 
What about the fact that the brand is so powerful, they were going to buy from you anyway before we intercepted them with paid activities that were not even really needed? 
 
Beyond the quant data collection alone, there are then the assumptions and business rules we apply to draw conclusions about these touchpoints.
 
This includes things like “On average it takes approx 7 days for people to make a purchase, let’s set that as our attribution window” to “Based on our strategy I’ll split my channel groupings by paid and organic social even though we’re boosting the same organic content to gain additional reach” to name only a couple.
 
Image item
 
Image item
Example Touch Attribution Configuration & Report Output from Amplitude 
 
So just like we referenced a few weeks back when discussing MMM,  “All models are wrong but some are useful”.
 
For this reason, touch attribution models with the sample data we can acquire have a track record of being immensely useful. One case study I found speaks to an e-learning company adopting a multi-touch attribution model to yield a saving of a quarter million dollars while managing to drive up their ROI by 30% (way to cut out the dead weight in your Google ads)! 
 
This alone speaks to how powerful digital analytics tools can remain for channels where there is significant spend. Why? Because we can get granular. The success in this case study and the successes I’ve achieved through adopting MTA outputs have been more powerful to optimise within-channel and through leveraging the power of digital tracking to provide granularity of insight at a campaign, keyword, placement, format, device, creative and landing page level. 
 
Ahh I remember the days sitting in my Google Search Console and Meta Ads Manager customising my UTM parameters and comparing the data within and outside of the platforms to optimise them…. (let’s tackle data discrepancies between walled-gardens and analytics platforms another day). 
 
[Actual evidence of me geeking out with Facebook nail art circa 2012. Cool or loser? You decide]
 
Image item
 
 
Our analytics platforms, whether Google Analytics, Adobe Analytics, Amplitude, Heap.io, Mixpanel or whatever else, show us what we have, not what we don’t have. 
 
For that, we need to get a little more clever to match the milestone events in the privacy timeline, to what we are seeing (or NOT seeing anymore) in our analytics platforms and what you aren’t seeing is impacting your analytics.
 
It doesn’t make it not relevant and not usable – but the changes in the data you can collect will impact you a little, or a lot depending on your business. I’d encourage everyone to go through their own exercise to match the timeline of events above to what you know about your customers and make notes that should be considered when using analytics outputs for decision making. I won’t map every historical event, but here are a few considerations to check whether there could be important missing data that is impacting you. 
  • You rely on cross-device interactions or know customers are likely to browse and buy across a longer path to purchase (note, if you’re on Google stack leveraging Google Signals, Google promises a bit of mitigation to this in GA4)
  • You happen to know a disproportionate number of your users use particular browsers or devices (e.g. you have an intermediated sales force with an app that only works on iOS but want to track your B2B paid and owned marketing)
  • You have a particular cohort of customers (with similar behavioural, attitudinal demographic, psychographic, geographic, and firmographic traits) who are opting out or not providing consent to track
  • You have a devastatingly low authenticated user count and the capability to capture customer data, uniting behavioural activity to a stored user profile (for those who consent to it!)
Overall, the privacy changes will change the underlying data available in our analytics platforms. Not to get too ahead of “The Solution” (coming next week) but are ways to bridge data gaps and mitigate fallout from cookie blocking and further limitations from things like parameter stripping in URLs. 
 
Regardless, the impact to our behavioural analytics platforms and the data we send them will remain. We just don’t have the volumes or the granularity we used to.
 
So what are the impacts at a glance? Why should we care anyway? Here’s what these privacy impacts all boil down to at a high level.
 
Total sample sizes and granularity has reduced and is reducing further
 
Seeing more (not set) in your analytics platforms across various dimensions? “Direct traffic” appear to be increasing? Seeing weird changes in your device and browser ratios?
 
To get a sense of the impact to your business, consider mapping the timeline above to better interpret the output of your analytics systems and temper your outputs to updated ratios based on what you know you might be missing. Then, always combine the output of with in-platform reports (e.g. in your buying platforms like Facebook and Google directly) and Marketing Mix Modelling (MMM) for the most comprehensive outlook. 
 
 
Bias is inevitable
 
When we cannot know, we imply. Or at least the algorithms do. When algorithms cannot tie hit and session data to the user level, we leave behind “deterministic” inferences towards implied “probabilistic” models. This is the difference between Google Analytics being able to go “I KNOW you were Kate Cook who clicked my Instagram ad on mobile and later purchased via your desktop computer”. 
 
For Google specifically (which we’ll use to follow the example) it will still struggle with this use case. 
 
It will however be able to bridge the gaps on our behalf for Google owned channels (e.g. you saw an ad on YouTube, owned by Google and later purchased on desktop where you are logged in on Google Chrome). 
 
Gaps are unknown
 
Without consistent and universal tracking, examples like the above leave it up to the analytics platform’s algorithmic approaches to link everything it can while leaving all that it cannot link in its black box. 
 
The gaps are widening in our attribution data and we lack the visibility to know exactly what. That is, unless you are building your own solutions (using tools like Snowplow and your own MTA solutions), there is a more compelling reason to, if only for the transparency and its ability to better inform the interpretation.
 

Hi I'm Kate! I'm relentlessly curious about the attribution and origin of things. Especially as it relates to being a corporate girly balancing ambition and a life filled with joy.

Leave a Reply

Your email address will not be published. Required fields are marked *