Moving Forward Without Cookies Part Two: Browsers Are the New Snack


September 22, 2020

Have you read the first article in this series from Adswerve Senior Data Scientist Anže Kravanja, “Moving Forward Without Cookies Part One: Using Tensorflow.js for Client-Side Predictions”? If not, be sure to check it out. Otherwise, read on.

We can all be drunk from the cookie-less future drinking game of the past year or so, but the endeavor is clear: it’s time to sober up and focus on your first-party data. Namely, how to create and/or utilize what you have to your advantage.

Our industry is in the midst of a proverbial caloric diet, and slowly but surely it’s working on giving up our beloved cookies, which have been so satisfying. So…what’s the solution? Behold! Client-side, machine-learning modeling! I know, that was a big bite.  

Let’s take a step back before we leapfrog forward. Quite literally, think about what you or any user does, while sipping your morning coffee, opening your laptop and reading your daily news or trade mags on your preferred site(s). You likely read the headlines on the front page, then move to your interest sections (for me, tech, fashion, real estate). What’s the acquisition of the day? What designer mask collab do I need to know about? How are the rich and famous decorating their apartments? These are all data points that we already have access to via browsers. Maybe not directly, but we can easily create proxy data points utilizing machine learning, coupled with elements like time on page, mouse overs, clicks, etc., to shift modeling capabilities into current browsing sessions, from which real-time decisioning can occur.

Sure, one could argue this data is less precise, or even myopic since there will naturally be fewer points from which to model and most importantly, predict. But the other side of the argument suggests that at least you can activate on it immediately, rather than waiting. This means having the ability to adjust elements on the page to be arguably more catered than what exists today. 

This POC would enable us to better understand how a user is moving through a site, and what they’re interested in, to make real-time adjustments in a single, or possibly subsequent, browser session(s). So now, not only advertising, but elements like content, headlines, offers, product offerings or form completions can dynamically adjust. The best part? It’s all based on your current actions and consumption habits rather than having to do with anything about you as an individual. Privacy-safe and more catered content/page elements? Sounds like the future to me!

At the risk of being repetitive, shifting modeling methodology toward client-side, using the browser as the conduit, is not necessarily a cookie replacement, but rather, could be deemed a healthier, lighter option that should help with any outstanding cravings, and provide immediate relief (to bring it back to our analogy above).

Now, to leapfrogging forward. Here are some media-focused use cases, bifurcated to single vs. multi-session examples: 

 Single Session Examples:

  1. Dynamic Content Optimization (oh goodie, another acronym). This is likely the most literal example. As a user is consuming content, the model can quickly learn what they’re interested in, skipping over, etc. to help update elements on the page, bringing it back to fundamentals of increasing time spent on site, and increasing average page views/user can likely be better influenced with this method.
  1. [Better] Informed Product Feeds, informed by the browser movement. This is not only a consideration for targeting products and offerings, but also for excluding them. Oooooh ahhhh, now you get to play in the Amazon sandbox. Or(provocative point ahead), if you’re losing money on Amazon due to their retailer fees, this may nicely complement your other marketing efforts, driving users to your own site for purchasing, which I assume yields better purchase margins, while still leveraging Amazon for the overall influence. In theory, this method can help complete a full eCommerce cycle.
  1. Form Completion Optimization. Now, page three can be informed based on the actions from pages one and two to help move users through forms more effectively. 

 Multi-Session Examples: 

  1. Remarketing is the first use case that comes to mind, and is arguably the most traditional one, taken from the olden [cookie] days – using past behavior to inform future behavior. Here, the browser-based model would learn based on previous browser sessions for the same user, to help inform the current browser session. This can be interwoven with eCommerce and retail goals as well, but also has implications related to what type of advertising (creative and/or messaging) to display to this user, and for what purpose. 
  1. A/B Testing – This can be used for both media activation (which campaigns, creative types, etc. are effective/not effective) but also, to be better informed about your landing page, form captures, etc. You can certainly create control and test groups, but now optimize while in session. Engagement tests, messaging or design tests, and even creative type tests can all be learned and predicted upon to help improve a user’s experience on your site, or as it pertains to your product offerings.
  1. Dynamic Creative Optimization – Quite literally, this refers to dynamically serving an ad based on the point of ad-serving so it feels even more ripe for this client-side technology, seeing as it’s currently reliant on cookies. It admittedly not be as precise as what is available in a cookie world, but, we’re dieting!

In order to take advantage of the above, you may not need to increase your investment in tech or resources, but rather, you may likely be able to simply amplify the technologies you already have by modernizing the plumbing, so the water flows in an evolved direction. Of course, this is dependent on your digital maturity, but if you have, as an example, GA360 + BigQuery, you’re all set to try this out. 

So pull out your wrenches and watch the cookies be flushed away to embrace in-browser data modeling and collection as a more privacy-centric, accurate and scalable potential approach to shift into this next phase of digital marketing and measurement.