We cut ad spend dramatically. Nothing happened.

There’s a wave of new MCP tools hitting the market right now — Shopify MCP, Meta MCP, email service provider MCPs. The pitch is compelling: connect everything to an LLM and just ask it questions about your business.

Is my marketing working? What’s driving my growth? Should I cut ad spend?

And the LLM answers. Every time. Confidently.

And that’s actually a problem.

An LLM connected to your Shopify store has no idea what your business looks like over time. It doesn’t know your seasonal cycles. It can’t separate growth you caused from growth that was already coming. It’s reading transactional data and pattern matching against general knowledge — NOT doing actual analysis of your specific business history.

That’s not a small distinction. Trusting those answers to make real spending decisions is risky.

Let me show you what the alternative looks like…

One of our portfolio brands runs a membership model. Like any subscription business, the big question is always growth. Are we adding members faster than we’re losing them? Is what we’re spending on marketing actually working?

I started modeling it properly using a tool called ARIMA_PLUS, built into Google’s BigQuery ML platform. ARIMA is a statistical method that’s been the standard for time series forecasting for decades. What it does is decompose your historical data into its component parts: the underlying trend in your business, the seasonal patterns that repeat year over year, and the noise in between.

It separates those three things from each other so you can see what’s actually happening underneath the surface numbers.

You train it on your own data. It learns YOUR business.

What makes ARIMA_PLUS especially powerful is that you can bring in external variables alongside your historical data. For this model I brought in ad spend.

Now the model can show me not just what the business is doing, but how much of that movement actually correlates with what we’re spending.

Two things became clear pretty quickly.

First, our ads are not driving membership growth. At least not yet.

Sure, sometimes we’d increase ad spend and see some growth. But the the model showed me that was noise, not signal.

So we cut ad spend dramatically. The business was essentially unaffected. Profitability went up. Growth continued.

Second, we’re heading into the toughest season of the year for this business. The model showed it clearly in the historical data. Even great marketing won’t change that reality. So we’ll keep testing and improving, but with the right context. Modest improvement this season is a win worth building on.

Neither of those insights came from asking an LLM a question. They came from a real statistical model trained on our own historical data.

ARIMA models is one example of what becomes possible when you own your data. It can be applied to almost anything you want to understand over time: sales velocity, inventory demand, customer growth, seasonal patterns across any part of your business.

I’ve been rolling these models out across our portfolio. If you’re curious what they might reveal about your business, reply and tell me what questions you’re trying to answer.

To your growth,

Deacon Bradley

Keep up with the latest E-Com Ops insights.