Lately, I have been spending a lot of time thinking about what happens when AI moves beyond answering questions and starts making marketing decisions.

Here are three things that stood out to me.

For years, analytics process has been pretty much the same: open GA4, Amplitude, Mixpanel or another analytics tool, look at dashboards, notice something unusual, and then start investigating.

I think AI is going to flip that workflow.

Instead of us constantly looking for problems, agents will continuously watch the data, identify what's important, investigate possible causes and tell us what deserves our attention.

We're already starting to see this. Amplitude, for example, is building agents that monitor dashboards, detect meaningful changes, investigate what's driving them and recommend what to do next.

The interesting question for me isn't whether AI can summarize a dashboard. It is:

Can AI reliably decide what is important enough for me to pay attention to?

That's a much harder and much more useful problem.

We're going to see more people connect ChatGPT, Claude and other AI tools directly to marketing data.

But there's a problem.

What exactly is a conversion? What counts as revenue? What's a qualified lead? Which attribution model are we using?

I don't want an LLM making those decisions on the fly.

The measurement system should establish the facts. AI should reason over those facts.

I think this separation, trusted measurement underneath, AI reasoning on top, is going to become increasingly important.

This is the part I'm most interested in exploring myself.

Imagine an AI agent sees that revenue dropped 30% yesterday and recommends cutting Google Ads spend.

Sounds reasonable.

Except revenue didn't actually drop.

Someone published a GTM change yesterday and the purchase tag stopped firing.

The AI may have reasoned perfectly from bad evidence.

I've been working on this problem with GA Auditor, where we already monitor GA4/GTM implementations and identify measurement problems. I'm now experimenting with what happens if we put an AI reasoning layer on top of those trusted signals.

Instead of sending someone ten alerts, could it say:

“Here are the three things you should investigate today and here's why.”

And more importantly, could it distinguish between a marketing problem and a measurement problem before recommending an action?

That's an experiment I'm going to start running.

The bigger idea I'm exploring is this:

Observe → Understand → Decide → Act → Measure → Learn

AI is rapidly getting better at the middle of that loop.

But if Observe is based on bad data or we can't reliably Measure whether the action worked the smartest AI agent in the world can still make the wrong decision.

That's something I'll be digging into over the next few weeks.

Thank you,
Anil Batra
Founder, Optizent