The AI Weekly Operating Review

Here's what usually happens with business numbers.

You look at them. Sales are up — you feel good. Sales are down — you feel bad. Then Monday comes, the phone rings, and you go right back to whatever you were doing. Nothing changes.

The problem isn't that you need better reports. It's that nobody ever turned looking at the numbers into a habit that ends in a decision.

This takes thirty minutes on Friday. You do it the same way every week. It ends with one thing you're going to do differently.

What you need

ChatGPT or Claude.

A place to keep your numbers — a spreadsheet is perfect. It doesn't have to be fancy.

Step 1: Pick five numbers

Five. Not twenty. You're going to pull these every single week, so they have to be easy to get.

For most businesses:

1. How many new leads came in 2. How many calls or appointments got booked 3. How many jobs or sales you closed 4. How much money came in 5. One cost that matters to you — ads, payroll, materials

Write those five down along with where each one comes from. Then don't change the list. The whole thing works because you're comparing the same five numbers week after week.

Step 2: Set up the spreadsheet

One row per week. One column per number. Plus a column for the date.

That's it. Five minutes.

Step 3: Every Friday, fill in the row

Takes about ten minutes once you know where everything lives.

Step 4: Ask the question

Copy your last few weeks out of the spreadsheet and paste them in with this:

CopyHere are my weekly numbers:

[paste your rows]

Tell me:
1. What actually changed, and what's just normal week-to-week bounce
2. Where the biggest problem is in this chain — if a lot of leads come in but few become customers, say so
3. Three things I could do next week, ranked by which would help most, each one doable in under two hours

Rules:
- Don't make up any numbers or assume anything I didn't give you
- If you can't tell something from this data, say so instead of guessing
- If I only have a few weeks of data, tell me how much I should trust your read

Step 5: Push back on the answer

This is the part people skip, and it's the part that makes this worth doing.

The AI will find a pattern. It's good at that. It's also good at finding patterns in random noise and describing them confidently.

So argue with it:

CopyYou said leads are trending up. I only have four weeks of data. How confident should I actually be about that?

CopyWhat would I need to know to be sure about that? What's missing from what I gave you?

A good answer will admit when it's not sure. If it never does, you're not asking hard enough.

Step 6: Pick one thing

One. Not three.

Write it on a sticky note. Put it where you'll see it Monday.

The reason is simple. You will not do three new things next week. You know that and so do I. But you might do one.

Step 7: Next Friday, start with last week

CopyLast week I said I'd [the thing]. Here's what happened: [what happened].
Here are this week's numbers: [paste]

Now it's not just looking at data. It's a loop. You tried something, you saw what happened, you pick the next thing.

What changes over time

The first few weeks aren't that useful. Three or four numbers can't tell you much, and you should ignore anyone who says otherwise.

Around week eight or ten it gets good. You've got enough history that real patterns show up and normal bounce stops looking like news. Around week twenty you'll catch things you'd never have noticed.

Gotcha: The most common way this goes wrong is skipping weeks. A gap ruins the comparison. Put it on your calendar as a recurring appointment and treat it like a meeting with someone else.

What to do next

Set up the spreadsheet today, even if it's empty. Then put "Weekly review" on your calendar for Friday afternoon, repeating every week, and give it thirty minutes.

That calendar entry is the actual job. Everything else is easy once it exists.