A report that collects its numbers, checks them, writes itself and sends itself, every morning.

The directors wanted to know how their LinkedIn ad campaigns were going. As I was going through the campaigns, I discovered that one of them had paused. LinkedIn had stopped allowing the way it was set up, and it paused it without me getting a notification.

I got it up and running again, along with the second campaign. Then, to prevent things like this happening, I built an automation that tracks Campaign Manager every single day and emails the directors a report going over the analytics of each campaign and how everything is doing.

A report that makes itself. Each morning a script reads the campaigns' own figures. A second check compares every figure with what LinkedIn shows on screen, to the cent, before anything else happens. An AI writes the analysis in plain English: what was spent, who saw the ads, what changed since yesterday and what to watch. The report is laid out as a four-page PDF in Cre8tive AI's own document style, each page is checked, and it is emailed to both directors.

I used the right model for each job. The strongest model leads the run and writes the analysis. Small, inexpensive models do the repeatable steps and check the pages. The collecting and the number checks are ordinary scripts, with no AI in them, because numbers shouldn't be left to judgement.

The footer says plainly that the figures are LinkedIn's and the commentary is AI analysis.

When I sent the first one, I told the directors to say what they wanted changed in how it's structured and what it includes, because I can change that in the automation.

On its first morning alone, the report didn't send. One of its own checks hadn't passed, so it held the email, left it as a draft and alerted my phone. The numbers were right. The check was too strict. I had the run reviewed and fixed that morning, and the next day the report went out at 8:05am with nobody at the desk.

That is how I want an automation to fail: it stops, it tells me, and nothing wrong reaches the people it is for.

They read it and they're happy with it. Two days in, one of them came back with a question it didn't answer yet: can we find out who is stopping on the ads? A four-page briefing answering that, in the same format, was with them the same afternoon.