AI Blocks › Get a whole job done
See a whole job from start to finish
An owner's ad costs keep rising. Follow what their AI does with the job: the brief, the research, the plan built from our specialists, the times it stops for the owner and how results are read.
This is the worked example our tools give every AI as the standard to meet. The screens are built from the example itself, so they change when it does. The numbers are illustrative, not a real business's.
The owner's words
Our Meta ads CPA keeps going up and profit is shrinking
Their AI calls start_outcome with these words, then asks one batch of questions.
The brief
What they want: Bring Meta purchase CPA back under break-even and toward the target, without losing new-customer volume or margin.
What they get at the end: A checked plan, built and launched, with an honest readout
Done when:
- Measurement checked, or its limits stated
- Diagnosis with every hypothesis marked and evidenced
- Owner-chosen plan with reasons recorded
- Ads and page fixes built and checked
- Changes launched with owner approval and read back
- Experiments registered before launch and closed with a verdict
- Learnings saved for the next round
How experts do it
Practitioners treat a rising CPA as a measurement question before a media question: they confirm tracking and reconcile platform purchases with real orders, check the landing page, and only then break cost down by segment, format, creative age and auction cost. Fixes are launched as controlled tests with one change each, read on enough volume, and fed into the next round.
- Know the brand and offer. Creative and page judgements need the brand voice, proof and audience first. (Owner answers, Our library: Ad account bottleneck auditor)
- Pull comparable before and after data. A rise is only visible against the period before it, at ad level with breakdowns. (Our library: Ad account bottleneck auditor, Owner answers)
- Check tracking and reconcile with real orders. Lost or double-counted events make CPA look worse or better than it is. (Meta Business Help Center, Our library: Set up tracking you can trust before you scale)
- Check the landing page and ad-to-page match. Slow or mismatched pages turn good clicks into lost sales. (Web Vitals, Our library: Ad account bottleneck auditor)
- Break cost down by segment, format, creative age and auction. Each cause needs a different fix, and several can act at once. (Our library: Ad account bottleneck auditor, Meta Business Help Center)
- Refresh creative with distinct concepts. Worn-out or look-alike ads raise cost even when targeting is right. (Our library: Ad account bottleneck auditor, Meta Business Help Center)
- Launch changes as controlled tests. One change per test, created paused, read on enough volume. (Our library: Meta test campaign planner, Meta Business Help Center)
What usually goes wrong
- Rebuilding creative while tracking is broken, so every later number misleads.
- Calling segment winners on too little volume.
- Editing ad sets so often they keep re-entering the learning phase.
The plan
15 stages, run by 14 specialists using 44 skills and tools from our library.
Pin down the problem
The brief. Turn the owner's words into a measured problem, a target, limits and the data we can reach. Writes BRIEF.md.
Specialist: Media buyer.
Runs without stopping. You see what it found later.
Build or reuse the brand brain
Know the business. Know what the business sells, to whom, with what proof and in what voice, so every later output fits the brand.
Specialist: Brand file from website.
Uses: Website brand brain builder, Product page brand profile extractor.
Runs without stopping. You see what it found later.
Gather the data in one place
Gather the data. Pull ad results and orders for comparable periods into one working dataset where the owner wants it.
Specialist: Media buyer.
Uses: Meta results review agent, Ad sales drop analyst.
Runs without stopping. You see what it found later.
Check the numbers can be trusted
Check the data can be trusted. Verify Pixel and Conversions API, deduplication, event match quality and attribution, and reconcile platform purchases with real orders.
Specialist: Paid media tracking specialist.
Uses: Set up tracking you can trust before you scale, Meta server tracking checker, Meta event matching evidence reviewer.
Runs without stopping. You see what it found later.
Check the site and ad-to-page match
Check the data can be trusted. Find drop-off from slow pages or a broken checkout, and score whether each page keeps the promise its ads make.
Specialist: Cro specialist.
Uses: Page speed fixer, Landing page auditor, Ad and page match checker, Ad library teardown counter.
Runs without stopping. You see what it found later.
Find what is really driving cost up
Find the cause. Test the hypothesis library against the data: segments, formats, creative fatigue, concept diversity, timing, auction costs and the landing page. Keep competing explanations.
Specialist: Media buyer, helped by Paid media auditor, Creative evidence review agent.
Uses: Ad account bottleneck auditor, Six pattern ad performance diagnoser, Ad performance early warning reviewer, Winners and losers ad comparer, Meta creative similarity risk reviewer, Cohort funnel reader.
Runs without stopping. You see what it found later.
Rank what to change
Rank the options. Turn supported findings into 5 to 15 ranked recommendation cards, fewer if the evidence supports fewer.
Specialist: Media buyer.
Uses: Break even ad target finder, Next dollar budget allocator, Small budget ad test planner.
Runs without stopping. You see what it found later.
Owner chooses
You choose. The owner picks card IDs. Ask for the reason when they pass on a top card or pick a low one; record it as a decision.
Specialist: Media buyer.
Stops for your choice.
Fix measurement
Make the work. Prepare the chosen tracking fixes: Conversions API, deduplication, event parameters or attribution, as exact steps or a ready patch.
Specialist: Paid media tracking specialist.
Uses: Set up tracking you can trust before you scale, Meta server tracking checker.
Runs without stopping. You see what it found later.
Plan the creative
Make the work. Turn chosen audience and creative cards into distinct concepts and briefs, in the brand voice, for the right awareness stage and format.
Specialist: Paid media creative strategist.
Uses: Awareness stage ad framework chooser, Ad angle planner, Seven bucket ad concept brainstormer, Meta concept diversity test planner, Competitor ad theme and gap mapper.
Stops so you can look over the work before it is used.
Make the ads
Make the work. Write scripts, hooks, headlines and primary text, and make static images and video drafts for each concept. Check every ad before the owner sees it.
Specialist: Campaign copy draft agent, helped by Design specialist, Weekly image ad batch maker, Weekly video ad batch maker, Paid media creative strategist.
Uses: Image ad copy writer, Five tactic ad hook writer, Five part creator ad script writer, Twelve pattern static ad brief maker, Six image ad prompt set, Static ad critic, Video hook scorer, Meta ads compliance checker, Pre launch ad checker.
Runs without stopping. You see what it found later.
Fix the landing page
Make the work. Prepare the chosen page changes: speed fixes and copy that keeps the ad's promise.
Specialist: Cro specialist.
Uses: Ad matched landing page writer, Page speed fixer.
Runs without stopping. You see what it found later.
Launch, paused first
Go live. Create the campaigns, ad sets and ads paused, read them back, register each test, then switch on only what the owner approves.
Specialist: Media buyer.
Uses: Meta setup of new ad campaigns with new creatives planner, Meta adding new ads to an existing campaign planner, Meta test campaign planner.
Stops before each live action for your yes.
Read the results honestly
Read the results. Check results on a schedule against the registered metric and limits, flag fatigue and breaches, and propose budget moves within the owner's rules.
Specialist: Experiment readout desk, helped by Ad test sprint runner, Ad fatigue watcher.
Uses: Ad test budget calculator, Creative test reader, Keyed ad results log keeper, Ad results report writer.
Runs without stopping. You see what it found later.
Keep what we learned
Keep what was learned. Close each experiment, record wins and losses, propose durable rules with their scope, and plan the next round.
Specialist: Media buyer.
Uses: Image ad test learnings, Find out why your ad test did not win and plan the next round.
Runs without stopping. You see what it found later.
Where it stops for the owner
The owner approves the plan first. After that, it stops for them 3 times.
- Approve the plan Nothing runs until you say yes. Ask for changes and your AI rebuilds the plan.
- Choose what to do Stage 8, Owner chooses. You pick which changes to make from a ranked list. If you pass over the top one, your AI asks why and writes your reason down.
- Look over the work Stage 10, Plan the creative. You see the work before it is used, then approve it or ask for changes.
- Approve each live action Stage 13, Launch, paused first. Your AI asks before each change that goes live, one at a time. Nothing goes live until you say yes to that change.
Reading results and learning
- How long each test runs: At least 7 days per test, longer for low volume; compare like-for-like periods.
- Enough data to judge: At least 50 purchases per variant, or the budget the ad-test calculator sets for this CPA.
- When to stop early: Stop early only if a guardrail breaks; otherwise run the full window before a verdict.
The 11 possible causes it tests first
- Cost only looks higher because tracking lost or double-counted purchases.
- Spend drifted toward audiences that buy less.
- A profitable segment gets too little spend or unsuited messaging.
- Most spend sits on the weaker format.
- Top ads are worn out.
- Too few distinct concepts, so delivery treats many ads as one.
- Some days or hours waste spend.
- Frequent edits keep ad sets stuck in learning.
- The landing page breaks the ad's promise or loads slowly, so clicks don't buy.
- The auction got more expensive (higher CPMs), not the ads worse.
- An offer, price or stock change made the product harder to buy.