🔌 Works best with your Meta ad account. Connect it and Jeremy AI pulls the last 30 days for you. It only reads. Nothing changes in your account.

Score every ad set. Decide Kill, Watch or Scale. Move the budget with the math checked.

You'll see the data and rules pasted in, the math check, the totals and the Scale, Watch and Kill calls. The optional image isn't in the video.

What this use case does and who it's for

You paste your ad set numbers and your rules. Jeremy AI checks the math first. Then it gives every ad set a decision and a one-line reason that names the rule. You get blended totals, and a budget move: the money from the ad sets you kill gets split across the ones you scale.

It's for agency owners, ad buyers and info-business owners who check Ads Manager by hand and argue about what to cut. The rules are locked, so the same numbers give the same decisions.

It's drafts only. If you connect your ad account, it only reads. Nothing changes in it.

What you need before you start

A Jeremy AI account. Nothing to connect to try the sample.
30-day numbers per ad set: spend, impressions, clicks, CTR, CPC, purchases, CPA and ROAS. Copy them from Ads Manager.
Your target CPA and what a purchase is worth.
Your own Kill and Scale rules. Be exact about them.

How to set it up

You use the default Jeremy AI worker. There's nothing to build.

Open Jeremy AI and start a new chat.
Check the worker selector says Jeremy AI.
Paste prompt 1. Swap in your own ad sets and rules.
Click Send message. A thread opens.
Wait about 1 to 2.5 minutes. You get a chat summary and a file card named ad_set_scorecard.md.
Scroll the math check, the decisions, the totals and the budget moves.
Optional: paste prompt 2 for a one-page scorecard image.

The prompts

Prompt 1 is the exact prompt from the walkthrough. Every ad set and number in it is SAMPLE data. Replace them with yours.

Score these SAMPLE (made-up) ad sets and decide Kill, Watch or Scale. Use ONLY the data below and the rules below. Don't invent numbers.
FACTS: target CPA $40. Each purchase is worth $100 in revenue, so ROAS = purchases x 100 / spend.
RULES (locked, apply in order):
- KILL: spend at least $120 and (0 purchases or CPA above $80).
- SCALE: CPA $35 or lower AND at least 25 purchases.
- WATCH: everything else.
DATA (ad set, 30-day spend $, impressions, clicks, CTR %, CPC $, purchases, CPA $, ROAS):
SAMPLE Testimonial-Cold, 1200, 80000, 1600, 2.00, 0.75, 40, 30.00, 3.33
SAMPLE Demo-Reel, 900, 60000, 1080, 1.80, 0.83, 30, 30.00, 3.33
SAMPLE Founder-Story, 800, 50000, 750, 1.50, 1.07, 20, 40.00, 2.50
SAMPLE Offer-Discount, 700, 45000, 630, 1.40, 1.11, 14, 50.00, 2.00
SAMPLE Listicle-Static, 600, 70000, 490, 0.70, 1.22, 5, 120.00, 0.83
SAMPLE Meme-Hook, 450, 90000, 450, 0.50, 1.00, 0, n/a, 0.00
SAMPLE UGC-Unboxing, 1000, 65000, 1300, 2.00, 0.77, 32, 31.25, 3.20
SAMPLE Retarget-Cart, 300, 12000, 300, 2.50, 1.00, 9, 33.33, 3.00
DO THIS:
1. Check the math first: recompute CTR, CPC, CPA and ROAS for every row and flag any mismatch.
2. A table: ad set, spend, CPA, ROAS, decision (Kill / Watch / Scale), one-line reason that names the rule.
3. Totals: total spend, total purchases, blended CPA, blended ROAS.
4. Budget moves: the spend of the Kill ad sets is freed. Split it across the Scale ad sets in proportion to their purchases. Show each dollar amount and the maths.
5. Drafts only. Nothing is changed in any ad account.

Prompt 2 is optional. It isn't shown in the video.

Make a clean 1-page scorecard image from the table above. Title: "SAMPLE Ads Kill/Scale Scorecard". One row per ad set with the decision as a colored tag: red Kill, amber Watch, green Scale. Show CPA and ROAS for each row. Footer line: "SAMPLE data". Keep the text short and spell every ad set name exactly.

Prompting tips and how to adapt it

Lock your rules. The sample says "apply in order." That's why the decisions are repeatable.
Use a minimum spend and a minimum purchase count. They stop you killing an ad set too early or scaling one on a lucky week.
Keep the math check. Step 1 makes Jeremy AI recompute every row and flag any mismatch. It catches typos in your paste.
Say what a purchase is worth. ROAS needs it. If your offer has different prices, use an average.
Change the split. The sample splits freed budget by share of purchases. You could ask for an equal split instead.
Keep "Drafts only." It makes it clear nothing gets changed in any account.
Use your own thresholds. Swap in your CPA target and your Kill and Scale lines.

The result to expect

A table with every ad set, its decision and a reason. Then totals. Then the budget moves, with the dollar amount and the math for each.

With the sample data you'll see 2 Kill, 3 Scale and 3 Watch.

Kill: Listicle-Static and Meme-Hook.
Scale: Testimonial-Cold, Demo-Reel and UGC-Unboxing.
Watch: Founder-Story, Offer-Discount and Retarget-Cart. Retarget-Cart has a great CPA but only 9 purchases, under the 25 you need.
Totals: spend $5,950, 150 purchases, blended CPA $39.67, blended ROAS 2.52.
Budget moves: $1,050 freed from the two Kill ad sets. About $411.76 goes to Testimonial-Cold, $308.82 to Demo-Reel and $329.41 to UGC-Unboxing.

These are sample numbers. Your ad sets will give different decisions.

Common mistakes and fixes

The decisions don't match your rules. Numbers can vary between runs. Check the table against your rules. If it's wrong, send it again.
A row is flagged in the math check. Look at your paste. A typo is the usual cause.
A winner is on Watch. Look at the purchase count. Scale needs enough purchases, not just a good CPA.
You expected it to change your budgets. It doesn't touch any account. You move the money yourself.
The image has a typo. Check each ad set name on the scorecard image and send prompt 2 again if one's wrong.

Use Case: The Ads Kill/Scale Scorecard (Works with Your Ad Account)