Almost Timely News: 🗞️ How to Use Generative AI For Analytics (2025-06-15)

Almost Timely News: 🗞️ How to Use Generative AI For Analytics (2025-06-15) :: View in Browser

The Big Plug

👉 Download the new, free AI-Ready Marketing Strategy Kit!

Content Authenticity Statement

100% of this week's newsletter was generated by me, the human. You will see bountiful AI outputs in the video. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future.

Watch This Newsletter On YouTube 📺

Click here for the video 📺 version of this newsletter on YouTube »

Click here for an MP3 audio 🎧 only version »

What's On My Mind: How To Use Generative AI for Analytics

Happy Father's Day to all who fulfill that role in life.

Today, let's answer this question, which was the topic of a workshop I did at the Marketing Analytics Summit:

Can we use generative AI for analytics?

No.

See you next week.

Part 1: Why GenAI Struggles With Analytics

I'm just kidding, of course. But even in that joking is a grain of truth.

The grain of truth is this: generative AI cannot do math. Fundamentally, the underlying architecture that powers tools like ChatGPT is incapable of doing math, in the same way that a blender is incapable of pan frying a steak.

It doesn't matter how fancy the blender is, it's never going to pan fry a steak. Can it cook a steak? Sure, by turning it into a puree and heating it with friction. Would you want to eat it that way? Probably not.

In the same way, a token prediction engine - which is what generative AI is - does not do calculation. Here's why.

Language, the language that we speak and write (and speaking came long before writing) is inherently a probabilistic, predictive task. We human beings are prediction engines ourselves; our brains are constantly trying to evaluate whether something is good or bad, whether we're in danger or not.

It took nearly 100,000 years of evolution for us to go from speaking to doing math, because math isn't language. Math doesn't describe, not in the way language does. It's symbolic in nature. The number 3 is a symbol representing three things, and we conduct deterministic calculations with numbers.

It took another 30,000 years for higher forms of math like geometric and arithmetic to arise because math is so different than speech. In fact, somewhat entertainingly, math and writing appeared about the same time, around 5,000 years ago. That shows how much more cognition is involved with math than speaking.

All that is to say that math and language are not the same thing. Generative AI tools are language tools; it's literally in the name of Large Language Models.

And the root of analytics is mathematics, from simple addition to very complex mathematical equations. Generative AI simply isn’t up to that task, so you can’t just hand your data to ChatGPT and call it a day, unless you don’t particularly care whether the answer is right or not.

In fact, a brand new paper and benchmark from Google came out just the other day which showed that generative AI models can’t even reliably read spreadsheets.

Now, does that mean generative AI has no role in analytics?

There’s the question we want to answer today.

Part 2: What Do People Really Want?

I’ve spent 31 years in digital marketing. My first website went online in 1994 (it ceased to exist long ago, sadly). And for those three decades, everyone I’ve ever worked for has said they want robust web analytics, robust digital marketing analytics. That was the topic at many a staff meeting, many a customer meeting.

And everyone who said that was lying.

Well, not lying, but what they wanted and what they asked for are two completely different things. No one wants analytics. Not really, not if we’re honest. What everyone wants is answers - and even there, answers they agree with or answers that make them look good (or at a bare minimum, allow them to shift blame).

Is that glib and depressing? Sure. But it’s also the truth.

On top of that, almost no one does analytics, even people with analyst in their job title. What people mostly do is data regurgitation, commonly called reporting. We trot out pile after pile of data, dashboards that are so heavily loaded, they look like a desperate buffet offering at a casino, and we call that analytics.

That's not analytics. That’s just backing the truck up and pouring data all over someone’s desk and hoping they sift through it to find the things they’re looking for.

What is analytics? We’re going back to the well here, back to my roots for 15 years of my career. Analytics comes from the Greek word analyein, to unlock or to loosen. It’s taking something that’s locked up and unlocking it. Interestingly, in Ancient Greek, it was the branch of logic that distinguishes good arguments from bad arguments.

It was not, and is not, the process of dumping spreadsheets on someone’s desk. That’s not analytics. That’s indigestion. Aristotle would not be impressed with the modern descendants of his analytics.

So when we talk about generative AI doing analytics, there’s some nuance. Generative AI absolutely cannot do the math of analytics - but the art of using logic to distinguish good from bad arguments, the art of making decisions? Generative AI does that very well, if we provide it with the right materials.

Part 3: What is Analytics Today?

Okay, so what is analytics today? Glib talk aside, how do we do something productive here?

We first have to start by distinguishing the different kinds of analytics. Long ago, we posited the Marketing Analytics Maturity Model, which is a hierarchy:

• Descriptive analytics: the data. The stuff. What happened. This is the number of website visitors or the number of email opens or the CTR. It’s the raw numbers.

• Diagnostic analytics: the why. Why did we see the numbers we saw? How do we explain what happened? Website traffic was up 30%, great. Why? Click through rates on our Google ads were down 150 clicks this week. Why?

• Predictive analytics: what’s going to happen? Once we know what happened and we have a good idea of why, we can start to predict, start to forecast, if the data supports it. This is where classical machine learning and classical AI. (Read: not ChatGPT) come in, to help us forecast what’s likely to happen. Predictive analytics has been in a rough state since March 2020 because our data has been such a mess due to <gestures wildly at macro circumstances>.

• Prescriptive analytics: what should we do? Given what happened, why, and what’s likely to happen, what should we do about it? This is where we make decisions. Analytics without action is distraction. Data without decisions is decoration.

• Proactive analytics: can machines do it for me? Back in 2018, when Katie Robbert and I started Trust Insights this was largely theoretical, the imaginary pinnacle of analytics. Today with modern AI and agentic AI, you can actually have agentic analytics.

When it comes to proving the value of marketing, the deeper into the hierarchy you are, the more value you’re unlocking. If you’re just doing descriptive analytics, you’re basically looking in the rear view mirror all the time. That’s fine as long as the road has no turns or obstacles, but if it does, you’re in for a bad time.

As you progress through the Marketing Analytics Maturity Model, you spend less time looking at what happened and more time figuring out what should happen, until you reach a point where you’re making decisions or even handing off decisions to machines.

An example of proactive analytics is an AI workflow where an automation is ingesting Google Ads data, determining which ads are underperforming, which ads are overperforming, generating new ads that are experiments on the over performing ads, deploying them in market, and basically operating autonomously. All five layers of the Marketing Analytics Maturity Model are in play in a system like that.

Here’s the rub. Almost everyone, like 95% of all companies and people, are stuck in descriptive analytics at best. They’re stuck making reports and dashboards, and it takes so long and is so painful that stakeholders don’t use the data for anything. They listen to the report or nod at the PowerPoint and make gut instinct decisions that they would have made anyway without any data.

Well, now that we’re thoroughly depressed, let’s talk about generative AI’s role in helping us move past being data dump truck drivers.

Part 4: The Role of Generative AI in the Marketing Analytics Maturity Model

Of the five stages of the maturity model, generative AI should not touch two of them: descriptive and predictive. Why? These layers are largely math, and math is best left to traditional code and analytics tools. Generative AI can help us make the tools to do the math, but it should not be doing the math for us.

The other