Analytics Roundup – March 2026

My roundup of recent marketing analytics news and ideas that I find particularly useful and/or interesting. I mostly work in the Google analytics stack and I’m a little bit obsessed with the intersection of AI and analytics. 

Training AI models is not enough, we also have to invest in training humans

I’ve been elbows-deep in AI-assisted analytics for a while, and it feels to me like the latest AI models from Google, OpenAI and Anthropic represent a tipping point. LLMs still produce plenty of nonsense, but I think it is possible that more than 50% of what I’m getting back is useful. A lot of human analysts don’t meet that bar.

I’m going to be honest with you: I find it thrilling, exhilarating, exhausting and low-key terrifying all in a jumble.

Like a lot of people, one of the things that keeps me up at night is the impact AI is having on the job market. I’m not too worried about my own situation. I expect to be pretty busy in the near term helping organizations collect good data and navigate the dizzying landscape of AI platforms, features and tools. In the long term I may have to retire from this line of work sooner than I expected, but that is a choice I can afford to make. 

But I have kids, and I care about humanity. 

What I am hearing anecdotally and reading about in business media is a rapid shift towards “diamond-shaped organizations”. Historically, organizations have been pyramid shaped: layers of hierarchy that go from a single leader at the top to a large workforce at the bottom. The idea of a diamond-shaped organization is that AI eliminates the need for the lower-skilled, more entry-level jobs at the bottom layer.

This begs a fundamental question: do we really want the next generation to have less opportunity than we do? Our ancestors did everything they could to make a better life for us. It will be a shocking reversal if we choose short-term gains over a better life for our children.

It is undeniable that much of entry-level knowledge work is now automatable—businesses that ignore that will fail. But if we don’t invest in our future, we all fail in time. The pyramid organization was an organic and lazy solution to workforce development—most of us spent time in jobs that fell far short of our potential, but you had to do the time before you had a chance to move up. 

To create a similar opportunity for people entering the workforce in the AI era, we will need to provide training that accelerates new hires directly towards mid-level roles. We have a fair amount of work to do to get there. Some larger organizations already have decent training programs in place, but over years of evangelizing apprenticeship, I have found that most employers prefer to hire people who already know how to do a job and offer no on-the-job training whatsoever.

I understand why. At my agency, we have a new marketing data analyst apprentice starting today, and I’m pretty anxious about the burden of responsibility I’m taking on. Providing meaningful training and mentorship is a big investment. I also think that now, more than ever, it is an investment we have to make.

If you are thinking about expanding structured training in your organization and want to hear what has and hasn’t worked for us, ping me in the comments. I want all of us, including you, to succeed.

AI-assisted Analysis

• Agents, swarms, and the AI awakening, the measure pod (podcast)
  This feels a bit more like overhearing a conversation at a coffee shop than a podcast, but it’s a conversation I’d totally want to listen in on. It’s two unpretentious guys mostly talking about their own progress and experiences doing analysis with AI.
• The Price We Pay for Effortless Analytics, June Dershewitz
  Mostly emphasizes how the “horse has left the barn” – people throughout organizations and society in general are using AI to glean insights from data, whether we like it or not. She advocates that data leaders should “set the bar, model good judgment, and prevent weak work from eroding trust.” I agree, but I also think she is the rare leader with quantitative and emotional intelligence. If you are thinking about who you should hire, keep or promote, I recommend looking for that combo.
• Bye, Bye Human-Powered Marketing Analytics, Avinash Kaushik
  Where a lot of data leaders see AI leading to an inevitable democratization of analysis (for better or for worse) Kaushik focuses on ways in which AI can help qualified analysts generate more actionable insights with advanced AI techniques.
  [Tip: I used the Claude Chrome plugin to read this article with the self-promotion removed, “Recreate this article with self-promotion removed. Do not make any changes to content that is not self-promotional, unless necessary to maintain flow with the promotional language removed.”]

Broader Takes on AI

• Something Big Is Happening, Matt Shumer
  A very thoughtful essay from an AI-industry insider. There’s too much to summarize, but two takeaways for me were:
  • As a business leader, I should be giving ALL of my employees the support, licenses, etc. to do their work with AI. For their sake as much as mine.
  • Right now, our future is largely in the hands of a handful of tech CEOs whose job it is to maximize returns for their shareholders. Advances in AI could have massive benefits for society, but for that to happen we need political leadership to face it head on, with the right knowledge, wisdom, and urgency. I don’t see that coming from people who were born before computers even existed.
• The AI Vampire, Steve Yegge
  A bleak but interesting buildup to a coherent argument for shortening the standard work week. 
• A World Where All Is Free? That’s Elon Musk’s Theory of ‘Sustainable Abundance.’, New York Times
  If you found Brave New World hopeful and uplifting, this theory is for you. Otherwise, only noteworthy because this particular jackass has the President’s ear.

Attribution & Tracking

• Inside Bing’s New AI Performance Report: What 20,000 Copilot Citations Taught Us, Will Scott
  Bing Webmaster Tools has started reporting on Copilot citations and the associated queries. I was aware of this, but not particularly excited until I read this article. Bing/Copilot certainly has a small audience share compared to Google AI Mode and ChatGPT, but you get to see what real human beings are typing into AI. 
• If you’re tired of waiting for Google Search Console to add similar features, a simple idea from Marco Giordano will get you  part way there: just filter your GSC data for longer queries. The average query length in Google is less than 4 words, and the average prompt in ChatGPT is about 23 words, so IMO this is a valid way of looking at it. I created a GSC dashboard using his method and it’s pretty freaking cool.
• The Haystack Paradox: What Happens After AI Creates a Million Ads, Juliana Jackson
  This is a 30-minute talk from Superweek, which happened at the beginning of February. It covers two main themes, and both give me goosebumps they are so exciting:
  • How to measure the influence of individual ad components when you have massive numbers of ads. The statistical methods she uses are somewhat advanced, but if you are willing to work through them the payoff is big.
  • The idea that creative performance is the metric that matters most when marketing is hyperpersonalized and most of the customer journey is unmeasurable. I don’t fully agree (nor understand, frankly), but she does have me thinking about marketing measurement differently.

Privacy

• How US Companies Can Avoid Data Privacy Fines, MeasureMinds
  A thorough overview of risks and steps to mitigate. Bottom line? US regulations are not as strict as GDPR, but US civil law and hungry lawyers make the risks of non-compliance greater.
• I heard some grumblings recently that we may need to be opting-out US visitors by default because of a dubious court decision based on the California Invasion of Privacy Act (CIPA). I’m not a lawyer, but this seems like good evidence to the contrary: California Court Limits Website Privacy Claims: Key Takeaways for Website Operators and Business Owners. 
• 10 common ways your digital analytics implementation leaks personal data, Mikko Pippo
  A lot of organizations equate cookie consent with compliance. This article covers common scenarios where personal data collection happens in spite of cookie opt-outs. I’ve personally encountered at least half of these. I’ve probably also encountered the other half without realizing it.

Miscellaneous

• A free MCP server for Google Tag Manager, Paolo Bietolini
  This is an easier setup than the Stape GTM MCP. I gave it a test spin on an audit of a fairly complex container, and it did a really nice job.
• The Geometry of Inequality, RJ Andrews
  Ways to visualize and understand wealth distribution. Powerful storytelling.