My roundup of recent analytics news and ideas. I mostly work in the Google analytics stack and I’m a little bit obsessed with the intersection of AI and analytics.
We Can Touch Infinity
“In a world where we can measure everything, or we think we can measure everything, how wonderful it is that you could have poetry or music that actually makes you think you are touching infinity.” – Yo Yo Ma
While I’ve really been enjoying exploring how AI can improve my workflows and help me uncover insights, I’ve also been feeling a bit anxious about the broader implications of AI on the job market and income inequality. I think these are legitimate concerns, but some recent readings have tempered my concerns a bit:
- LLMs are extremely fallible.
OpenAI recently shared results from accuracy tests they ran on gpt-5-thinking-mini. TL;DR: it answered questions right 22% of the time and wrong 26% of the time. This jibes with my own experience as I’ve ventured beyond questions that mostly require regurgitation. - Fallability may be more of a feature than a bug.
There are inherent limitations to LLM’s ability to solve problems in unconstrained topic domains. - The impact of AI on productivity is mixed.
Many organizations are finding that the cost of amplified mediocrity is counteracting efficiency gains, and AI is actually making experienced software developers less productive! The latter is an important bellwether because of how quickly AI tools and processes have been incorporated into the dev world.
I recognize that it is early days yet and we continue to see improvements in AI models and applications, but these challenges run pretty deep.
The parlor tricks LLMs can perform resemble real intelligence because that’s more or less what they are designed to do. What they can’t do well, however, is to navigate poorly-mapped topic domains with no clear boundaries. Humans aren’t always perfect at that either, but it is more or less what we are designed to do.
I will continue to explore ways AI can help with data gathering, cleaning, modeling, routine analysis, and visualization, but I am not quite as concerned about impending human irrelevance as I was a few weeks ago. AI is giving me more time to do the things I love, which are synthesizing, teaching, and creating. Yo Yo Ma may be better than most, but we all have it in us to touch infinity.
Product Updates
- Looker Studio added conditional formatting capability to query results variables.
- Google Analytics added automated advertising cost importing from Pinterest and Snap. They already had support for Reddit, but unfortunately not LinkedIn nor Meta yet.
Workflow
- The Google Ads API now provides search term data for PMax campaigns! And while this data isn’t included in the standard Google Ads BigQuery transfer, you can add it to the transfer as a custom table.
I got these (and a few other) links from the amazing GA4BigQuery newsletter – which was recently revived by Balazs Vajna, and is packed with BigQuery tips and links to a variety of great analytics articles. - Related, Johan van de Werken and Simo Ahava released the long-awaited overhaul of their GA4 + BigQuery course. I found the previous version of this course invaluable, and this one covers a lot more.
- Jude-Nwachukwu’s Attribution Insights Tag GTM template now supports custom UTM parameters – this tag captures and saves URL parameter values and exposes them as variables for later use with other tags. It is great for getting everything you need to enable importing CRM conversions into ad platforms and Google Analytics.
- GA4 Annotations Looker Studio connector, Steve Lamar
Pull annotations into Looker Studio dashboards. - BigQuery AI.GENERATE tutorial: turn SQL queries into AI-powered insights, MeasureLab
The BigQuery AI.GENERATE function brings the power of LLMs to SQL in BigQuery. This tutorial shows how to use the Vertex AI model to generate insights from GA data and do sentiment analysis on customer feedback. Note that the first example references a GA event parameter, “link_domain”, but I think “page_referrer” would work better. “link_domain” doesn’t typically exist in the session_start event. - GA4 Unassigned Traffic: Rename Channels Instead of Explaining Them, Dana DiTomaso
A practical tip for reducing the amount of time you have to spend explaining “Unassigned” traffic in Google Analytics. There’s also a blog post. - Engaged Session tracking (GTM Custom HTML Tag), Daniel Perry-Reed
Recreate the GA4 engaged session metric in GTM – very handy as an optimization metric in higher-funnel marketing channels. - From Model Gardens to… walled gardens?, Juliana Jackson
Enterprise-level thinking on next-level marketing analytics in Google Cloud. The parts I found most useful were the perspectives and resources related to server-side Google Tag Manager (sGTM) and Pantheon. The latter is a set of tools for increasing the utility of sGTM.
Attribution
Ideas