A curated roundup of marketing analytics news, tools and ideas — with a focus on the Google analytics stack and the intersection of AI and analytics.
An honest semantic layer
There’s a thesis going around in analytics circles: to do AI-assisted analysis right, you need a good semantic layer. I think that’s half right and half dangerous.
The practical application of a semantic layer can vary quite a bit, but the overall idea is that,
“The semantic layer serves as an intermediary translation layer in the modern data stack, converting raw data into business-meaningful information. It creates a unified business view of data across an organization, regardless of where the data resides or how it’s technically structured.” – Databricks
Ok, that tracks. So let’s look at the metric “Active User” as an example. Google Analytics defines an active user as, “The number of unique users who engaged with your site or app in the specified date range.” There’s an article below by Juliana Jackson that goes into detail about why measuring unique users is basically impossible, but here are a few key points:
- The primary mechanism GA uses to identify a user is a cookie, which is tied to a browser/device. Therefore, when a person changes browsers and/or devices, they show up as two people.
- Privacy controls and consent mean we only see a subset of traffic/interactions.
- User counts always include some amount of bot traffic.
“Active” is similarly complicated and, frankly, weird. For example, a ‘first visit’ based on the problematic definition of a “user” automatically counts as active.
So, an honest definition of a Google Analytics active user would need to be something like:
“A count of human- and non-human-initiated, unique browser/device combinations that have not opted out of or blocked tracking cookies.”
I’ve left a lot out—the comprehensive, accurate definition of an “active user” is even more convoluted and problematic.
Using Google’s definition, an LLM is likely to suggest marketing or seasonal factors when the active user count drops. I know, because I’ve received just such “insights” many times. On the other hand, the accurate definition could point to a consent requirement change, which is often the real reason, in my experience.
If we train our AI tools on the business-friendly but dishonest version of this and other metrics, we will get dishonest results. LLMs don’t need help making shit up—if we are to get real insights from our AI tools, we need to give them the unvarnished truth about what the data actually represents.
Product updates
- Google Analytics: Task Assistant
There’s a new ‘Tasks’ link above Settings in GA that lists configurations for getting the most out of your Analytics property—sort of like a mini audit. It’s a little bit skewed towards features that support Google Ads, but I reviewed a variety of properties and on the whole I agreed with the recommendations. I find it a little annoying that you can’t dismiss tasks that are not relevant to a property, but only a little. - Google Ads: is streamlining the setup of enhanced conversions. Ordinarily, Google Ads’ “don’t-worry-your-pretty-little-head-about-it” philosophy of feature design annoys me, but the legacy setup process for enhanced conversions annoyed me even more. More about this from Anu Adegbola at Search Engine Land.
- Looker Studio is (renamed) Data Studio, Google Cloud Blog
There’s been a lot of knicker-twisting in the analytics community about Google’s see-saw on the name, and so far it really is mostly a name change. But it is also a strategic broadening of the positioning of Data née Looker Studio. For more on that, check out the next links. - From data to dashboard: Empowering BigQuery users with Data Studio, Google
This is a short clip from a session at Google Cloud Next describing Data Apps and the direction of Data Studio. Data Apps look to me like the result of Colab and Data Studio getting together and making a baby.
For more of the nitty gritty, here’s Google’s documentation: Use Colab Data Apps in BigQuery and Data Studio - Data Studio: Conversational Analytics with Data Agents
Last month, I wrote about Data Agents in BigQuery. Now you can converse with your Data Agent in Data Studio. It took a while for me to realize that this doesn’t mean that you can talk to data from a dashboard. Instead, there’s a separate UX with an interface that’s similar to the interface in BigQuery, but friendlier. To try it out, go to your agent in BigQuery, click ‘Share’, then ‘Copy Link To Agent in Data Studio’. Then paste the link in a browser to try it out. For most of us, “Data/Looker Studio” is synonymous with dashboards, but it appears to be evolving to be more than that. - Meta: Meta Is Launching An Easy Button For CAPI, Allison Schiff, AdExchanger
An auto-magical one-click setup for the Conversions API. In my experience, Meta regularly takes credit for more conversions than is mathematically possible, but at least now you don’t need to spend as much time to get fishy numbers.
AI-assisted analysis
- Data and AI 101, Madison Mae
Five pillars for safely using AI for data analysis within an organization. I found myself a bit conflicted as I read this. The author is clearly a proponent of centrally-controlled access to data, while I tend to advocate for looser restrictions around anonymized marketing data. Where I land is that the pillars are good, but I would implement them as a framework for education rather than control. - Five things I believe about the future of analytics, Tristan Handy
Prognosticating from the CEO of dbt. He didn’t say this exactly, but my takeaway is that AI is making the analyst less dependent on the technology organization – data engineers, analytics engineers, etc. Not an original take, but there are a lot of details and nuance in his POV that are worth reading. - Consulting the Oracle: Claude on the Future of Data, Jordan Tigani & Claude, MotherDuck
Jordan prompted Claude on the future of data engineering, the data stack and analytics. It’s a bit tongue-and-cheek, but also very real, and has parallels to Tristan Handy’s predictions. Favorite quote: “Human judgment applied precisely to the right question at the right moment — that becomes the only thing the machines cannot yet replicate at will.”
Data visualization & reporting
- Make the user to look where you want them to look: the guide on guiding attention, Martynas Jočys
The title kind of says it all – it’s specifically about presenting data in a dashboard or report. I thought it would be fun to turn the design principles into an interactive tool, so Claude helped me do just that: try it out. - SEO reporting outgrew Data Studio — here’s what comes next, Bruce Clay, Search Engine Land
This article is unnecessarily negative about Data Studio. Data Studio isn’t perfect, but really the criticisms are more about dashboards in general. It strikes a chord for me, because I’ve been realizing that the time I and my team spend building and maintaining dashboards probably far exceeds the time my clients spend looking at dashboards. And it turns out, we are not alone. The second half of the article outlines a solid reporting methodology that is made possible by LLMs.
Attribution & measurement
Ideas
- The Conscience of the Machine, Kevin Brackney
AI is making it easier to build software, and a natural consequence is that we spend less time thinking through what we are building and why we are building it. Kevin addresses this topic from a moral perspective. - We Are Not Data Accountants, Dana DiTomaso
Wisdom about how we think and talk about imperfect data, “imperfect” being superfluous when it comes to marketing data. - Anomalous State of Knowledge, Dorron S.
A big part of what attracted me to marketing analytics was a fascination with search behavior and what it can tell us about the human mind. To date, I’ve tended to think of search as a controlled process of knowledge acquisition/refinement. This post/video/paper turns that idea on its head and leaves me a little breathless.
Miscellaneous
- Google recently released an excellent series of videos on Meridian, their marketing mix modeling (MMM) platform. While focused on Meridian, the series is a decent primer on MMM in general.