Did you know you can now use Gemini as an AI assistant directly inside Google Sheets?

I did. Found out from a friend. He connected Gemini to a spreadsheet with his marketing data: spend, leads, sales, channels, campaigns – you know the kind, his own little data warehouse with all the bells and whistles.

My first thought? Finally.

No more waiting two weeks for a dashboard, no more pestering an analyst, no more juggling 14 browser tabs. Just sit in the spreadsheet and ask:

"Why did CPA go up?" "Where are we losing money?" "What should we turn off? What should we scale?"

And AI is supposed to figure it all out.

And it did. Almost.

To be honest – it really is simple.

You take data from your ad accounts, CRM, GA4, Stripe, dump it into a spreadsheet, connect Gemini, and start asking questions.

"Find the anomalies." "Explain the drop." "Which campaigns should we turn off?"

And it answers. Confidently. Smoothly. In plain language – sometimes even with a table, conclusions, and recommendations.

The magic lasts about five minutes. Then it turns out that chatting with data isn't actually something you want to do on an ongoing basis.

Let me explain.

My friend ended up trying it a couple of times and gave up. Said something like: "Maybe I just didn't have enough data." Then added that maybe Claude would work better – supposedly smarter.

So the problem was the tool again. Meanwhile, my friend had real things to deal with: work, plans, budgets, deadlines, and a founder asking why everything looked fine yesterday and today it's "we need to figure this out" again. But instead he went to chat with Gemini.

The thing about us humans is that we don't actually want to chat with data. We want the data to just tell us what to do.

But agents don't say anything until you ask them. And they answer exactly the question you asked – nothing more. Which is where things go sideways. What does an agent do with total freedom? Exactly: it tries to please you.

So instead of magic, we get "You're absolutely right." That's a dead end. The question "what's going on with our marketing?" sounds perfectly reasonable in your head. To an agent, it's just a great opportunity to generate tokens.

A useful question sounds different: "Which 3 ad sets over the last 14 days had CPA more than 30% above target, spent over $500, and generated zero sales?"

Now AI can actually help. Because it has a framework – not "look at the data and tell me what I see," but "help me make a specific decision."

And now imagine you give it proper scaffolding: a clean, consistent dataset, detailed instructions, example queries, the right output format, external context, and a usable interface.

That's a completely different story. The agent starts working toward real hypotheses – not toward output tokens.

The bad news: building that scaffolding takes real resources. I say this with confidence because I regularly work on AI analytics architecture as a fractional Head of Marketing Analytics.

Try starting:

– Not with a dashboard. – Not with an agent. – Not with "let's plug in AI and it'll figure itself out."

Questions first. Then decisions. Then data. And only then – AI.

Because AI doesn't fix a weak problem statement. It just shows you faster that you never had one.

Want all my posts in one place? I put 350+ articles on GA4, BigQuery, attribution, and metrics into one searchable library.

Go here to explore it for FREE.