This week is about the measurable stack. Today we open with the part everyone fakes: analytics. Most dashboards are decoration. Pretty numbers nobody acts on. We are going to do the opposite and build a real measurement layer, the kind that tells you what to do next. By the end you will know how to pull your own numbers, clean them, and let a model read them back to you in plain words.

First move. Stop looking at the front-end dashboard and go to the source. Open Google Analytics 4 and turn on the BigQuery export in admin. Now every event lands in a table you own. The dashboard is somebody else's summary. The raw table is the truth. You cannot fix what you can only see through a filter.

Second move. Put a model on top of that table. Run Claude as a coding agent in the terminal and connect a BigQuery MCP server, a small adapter that lets the agent run queries for you. Now you ask a question in plain English and the agent writes the SQL, runs it, and hands back the answer. No more waiting on a report.

Third move. Pick the one number that pays the rent. Not pageviews. A booking, a signup, a paid order. In GA4 mark it as a key event so it is tracked on its own. Everything else is context. When you know the one number, every other metric is just there to explain why it moved.

Fourth move. Connect that number to where it came from. In the export, join the conversion event to the session that started it using the user and session id fields GA4 already writes. Now you can ask the agent which channel brought the people who actually bought. That join is the whole game. Most people never do it.

Fifth move. Clean the data before you trust it. Filter out internal traffic, bots, and your own office IP in the GA4 settings, then have the agent flag any session under two seconds with zero events. Dirty numbers feel like progress and lead you off a cliff. A small clean dataset beats a big lying one.

Sixth move. Ask the agent to compare this week to last week and tell you only what changed by more than it normally wiggles. That is a simple standard-deviation check it can write in a few lines. You are not hunting through rows. You are getting a short list of what actually moved, so your attention goes where the money is.

Seventh move. Before you change anything, save today's numbers to a plain file with a date on it. A snapshot. When you tweak an ad or a page later, you compare against that file, not against your memory. Memory lies to make you feel smart. The file does not care how you feel.

Here is the honest part. We run this exact loop on ourselves, The Big T-M. The whole calendar you are watching was built from one source file, and the agent reads our own traffic the same way it would read yours. We are not selling a screenshot. We are showing the machine doing the reading.

So the takeaway is small. Analytics is not a wall of charts. It is one owned table, one number that matters, and a model that can read both. Next step: export your raw data, name your one number, and ask one plain question of it. You can watch our whole stack run live at office dot temerarii dot xyz.