Ask Your Marketing Numbers Anything

Grant Calloway

Sunday, nine at night. You've got three CSV exports open, and no two of them agree. Somewhere out there, an agency is assembling a PDF that'll hit your inbox in, what, three weeks? A report about a month that's already, um, dead and buried. I want to retire that ritual today.

Grant Calloway

Because here's the question. What if you could just ask your marketing numbers a question, in plain English, and get the answer right now? Not next month. Now.

Grant Calloway

Let me start with a number, and I'll flag it's from a Small Business Expo article, so treat it as a direction, not a decimal. They say small business AI adoption jumped 41% year over year, from 39% to 55% between 2024 and 2025. And the part that caught me, 45% of small businesses are using AI for data analysis. Not just writing posts. Analysis.

Grant Calloway

Now, how does that work, mechanically? A Technology.org piece walks through it. There's a standard plug called the Model Context Protocol, MCP for short. It lets the AI connect to your actual ad accounts and sales records, instead of, you know, improvising. So you type something like, and this example is from that same Expo article, "Which ad campaign had the lowest cost per lead in Chicago last month?" Our outline version says qualified lead, which is the better question. And back comes a breakdown. No SQL. No dashboard you spent a weekend building.

Grant Calloway

Okay, little floor story. Back at the NYSE, we'd get rookie traders on a Friday afternoon firing off market orders blind. No book depth. They couldn't see who was standing behind the bid, so they just, they just hit the button and hoped. Market order without the book is like driving at night with the headlights painted over. Sometimes you get home. Sometimes you meet a tree.

Grant Calloway

And that's the thing. People think conversational analytics is a creative toy for writing captions. It's not. It's a live order book for your marketing budget. Is your $500 ad spend pulling in buyers who'll come back for years, or is it subsidizing tire kickers? You can see the depth before you hit the button.

Grant Calloway

And if you remember the micro-segments we carved out last time, the tranches, this is where they earn their keep. Ask for cost per customer by segment. Segment by segment. The blast email never could tell you that.

Grant Calloway

Now, the cold water. Because I've been through two crashes, and I get allergic when everybody's cheerful. Gartner has forecast that more than 40 percent of agentic AI projects will be cancelled by the end of 2027. That's per the same Technology.org piece. The reasons given are inadequate risk controls and unclear business value.

Grant Calloway

Why should a five person shop care? Because the danger isn't that the AI says nothing. It's that it says something, fluently, in a tidy little table, with confident formatting, and it's, it's, it's fiction. Say the model is reading a CRM that updates in batches, or data scattered across three tools that don't talk. It will still answer. Politely. Wrongly.

Grant Calloway

So how do you catch it? That Technology.org piece describes what it calls a three way query test, and I like it because it costs nothing. You ask the same business question three different ways. One, "Break down acquisition cost by ad channel." Two, "What is our CAC across Google and Meta?" Three, "Show total ad spend divided by new customer count last month."

Grant Calloway

Same question, three costumes. If the figures match, good, you're probably looking at a real calculation against real data. But if you get three different numbers, or you get hedged prose, a lot of "approximately" and "it appears that"... that's a model writing plausible fiction, not running a verified database calculation. Back away from the ticket.

Grant Calloway

Think of it as getting three quotes on a roof. If they come in at nine, nine, and nine and a half thousand, fine. If they're nine, thirty, and "it depends," somebody's guessing.

Grant Calloway

So here's a sixty minute audit. Call it a balance sheet check for your marketing. First, connect a conversational AI layer to your current sales and ad accounts. Current, not last quarter's export. Second, ask it for the exact return on your last $1,000 of marketing spend, and run the three way test on that answer. Third, cut the bottom 20% of your underperforming ad variations. Fourth, and this one's the discipline part, demand verified data lineage. Where did that number come from? Which table, which account? No lineage, no capital allocation for next month.

Grant Calloway

Sunday night spreadsheet dread, meet your replacement: one hour, one honest question, asked three ways. If the answers agree, spend the money. If they don't, you just saved yourself a month of tuition. Alright, that's me. Go ask your numbers something.