A Worked Example: AI Ad Management Set Up in One Week

Can you get AI ad management working in a week? You can get a safe, working routine in place in a week. Whether it improves your results is something only your own numbers can tell you. This post walks through a seven day plan as a worked example.

In one week you can have read-only access, a defined result, a first review, a few approved changes and a spending cap, which is a sound base for AI ad management.

The imaginary business

Say you run a neighbourhood bakery called Rise and Shine. You sell bread, cakes to order and coffee. You run two kinds of ads: a local search campaign on Google for terms like "birthday cake near me", and a small social campaign promoting weekend specials. You have a modest monthly budget, you check the dashboards when you remember, and you suspect some money is wasted.

The lesson that matches this plan is Lesson 4.4: Run your ads with AI, safely. It teaches connecting an account read only, seeing what works, approving each change and setting a limit. The week below follows that sequence.

Day 1: Decide what counts

Before touching the ad accounts, write down the result that matters. For Rise and Shine it is cake orders placed through the website form, because cakes carry the best margin. Coffee foot traffic matters too, but you cannot measure it from ads, so it is left out.

Google's conversion measurement help page says you choose what you identify as valuable, such as a purchase, a sign-up or a phone call. So the first job is checking that the website form submission is recorded as a result in the ad account. If it is not, that gets fixed before anything else, because every later step depends on it.

Day 2: Restrict access and fix the ceiling

The owner gives the AI tool read-only access, or exports reports to paste in instead. Google lists a read-only level in which users can view campaigns and run reports, but not edit campaigns. On Meta, the analyst role can view ads and access reports.

Then comes the money. The owner looks at the Google budget and applies Google's own rule: actual spend on a day can reach up to twice the average daily budget, and the monthly maximum is 30.4 times the daily budget. The owner decides the worst day and the worst month are both acceptable, and sets a card alert as a second layer.

Day 3: The first review

The owner asks AI for a summary of the last 30 days, with these questions:

• What was spent per campaign?
• How many cake orders did each produce?
• Which search terms triggered the ads?
• Which ads had the weakest click rate in their group?

Suppose the answer flags that the search campaign shows for terms like "cake recipe" and "cake decorating course". Those people want to bake, not buy. The owner opens the search terms report personally to verify. They find the terms there. That check tells the owner the AI is reading the account correctly.

If the check had failed, the sensible move is to stop, fix the data or the brief, and not proceed.

Day 4: Review the suggestions

Now AI is asked for proposals only, with reasons. Imagine five come back:

1. Add "cake recipe" and "cake decorating course" as negative keywords.
2. Pause an ad whose headline mentions a discontinued product.
3. Rewrite the headline of the weakest ad.
4. Raise the search budget.
5. Split the social campaign by weekend special.

The owner approves 1 and 2, holds 3 until next week, and rejects 4 and 5 for now. Why reject the budget increase? Because the owner has not yet seen whether removing wasted spend changes anything, and a higher budget spends more money right away. This is the one change that should never be approved by default.

Google's guidance on the search terms report says you can select terms and add them as negative keywords, and that negatives added this way default to exact match. The owner knows exact match negatives do not cover variants, so "cake recipes" with an s may need adding too.

Day 5: Make the changes and log them

The owner makes the two approved changes personally and writes them in a note with the date. Google also keeps a change history of account changes, so the owner checks it shows both edits. Small step, big comfort.

Nothing else is touched. Google's help also notes that smart bidding strategies need a learning period and advises avoiding frequent manual changes to budgets or targets during it, which supports keeping the week quiet.

Day 6: Look at the landing page

Rather than editing more ads, the owner asks AI to read the cake ordering page and list anything that would make a first-time visitor hesitate. Suppose it points out that the form asks for eight fields, that the page does not say how far ahead to order, and that there is no phone number. These are suggestions the owner can judge. If the page is weak, working on it may matter more than any ad edit. The relevant lessons are Lesson 4.2, A page that sells while you sleep and Lesson 4.1, A live website in one afternoon.

Day 7: Write the next week's plan

The owner writes a three line plan:

• Check whether the new negatives reduced irrelevant terms.
• Decide on the headline rewrite.
• Run the same five questions as on Day 3 to compare.

That is the whole week. Notice what is missing: no dramatic claims, no big budget move, no hands-off automation.

1. Day 1Define the one result and confirm it is tracked.
2. Day 2Set read-only access and a spending ceiling.
3. Day 3Run the first review and verify two claims by hand.
4. Day 4Collect proposals, approve a few, reject the rest.
5. Day 5Make the changes yourself and log them.
6. Day 6Review the landing page for friction.
7. Day 7Write next week's plan and the comparison questions.

What could go wrong in the example

• The AI misreads the report. The Day 3 check catches this.
• The result is not tracked. Day 1 catches this. Without it, the whole setup optimises for the wrong thing.
• The week is too short to judge. Likely. One week sets up a routine, but it rarely shows a reliable trend, especially on a small budget. Give changes at least a couple of weeks.
• Overreach. The urge to approve everything on Day 4 is strong. Resist it.

Turning the example into your own

Replace the bakery with your business. Swap the result for yours, the terms for yours and the campaigns for yours. Keep the order. If you want the structure as a repeatable routine afterwards, see the RUN framework, and for the full beginner walk-through see the complete guide.

Common questions

Questions people ask

Is this a real case study?

No. It is a labelled example. We are not claiming any measured results for it.

Do I need a bakery sized budget?

No. The steps work at any size, but smaller budgets take longer to show any pattern.

Should I approve every suggestion on Day 4?

No. Approve a few small, reversible ones and reject anything you cannot explain, especially budget increases.

What if my result is not tracked?

Fix that first. Everything after depends on knowing what a good result is.

Take the next step

The week above is the same order Lesson 4.4 teaches, with screens to follow along: Run your ads with AI, safely.