What this checkpoint covers
The questions apply the guide to AI in ad platforms and the Performance Max structure guide, with one drawing on the guide to Advantage+ audience reaching existing customers. Most describe a campaign that reports well and is quietly doing the wrong job.
Task
- Write your answer to every question before reading any explanation.
- Give the decision and the reason, as you would explain it to a business owner.
- Mark each answer right, partly right or wrong, and note what you misjudged.
- Where an answer changes your design, revise it in the garden tools retailer guardrails exercise.
Questions
- A kitchen fitter runs a search campaign on automated bidding that counts a brochure download as its conversion. Cost per conversion has fallen sharply, but the sales team says enquiries are worse than ever. What would you change?
- A team adds its best customer list as an audience signal in Performance Max and expects ads to show only to people on that list. What have they misunderstood, and which controls actually limit delivery?
- A homeware shop with similar margins across its range wants to split one Performance Max campaign into a campaign per category "to see what works". What do you advise?
- A Meta prospecting ad set uses Advantage+ audience. To keep it relevant, the team adds a Lookalike AudienceAudience similar to a seed list. of all buyers and the customer list as suggestions. A month later the customer share of spend has risen. Why, and what is the fix?
- A shop switches on the Conversions APIServer-side event sharing with Meta. beside its pixel. Reported purchases roughly double overnight while orders in the shop are flat. What is the likely cause?
- After Performance Max launches, the brand Search campaign's conversions fall and Performance Max reports an excellent return. The manager proposes pausing Search. What should happen first?
- A team adjusts its target cost per acquisition every day because results swing. Why are the results not settling?
- Platform return on ad spend (ROASRevenue generated per unit of ad spend.) rose after a move to automated bidding. The finance director asks whether the automation added sales. How do you answer?
Answers explained
- Change what the campaign optimises towards. The campaign is finding exactly what it was told to find: cheap downloads. Make a qualified enquiry the goal, report outcomes back from the CRM, and give values where enquiries differ in worth. The guide to AI in ad platforms calls bidding automation a measurement problem in disguise.
- Signals are suggestions, not targeting. Google documents audience signals as hints that speed up learning; the campaign serves beyond them. The hard controls are exclusions where available, brand and URL controls, and the customer acquisition goal, as the Performance Max structure guide sets out.
- Split by target, not by taxonomy. Categories with similar margins and goals share a target, so separate campaigns only divide the conversion data. Use listing groups, asset groups and product reports for visibility, and split only where a different target or budget rule is needed.
- Suggestions start delivery among buyers. Under Advantage+ audience, suggestions can be exceeded but are where delivery begins; excluded custom audiences are the input Meta documents as honoured. Move the customer list and a purchase audience to exclusions and remove buyer-based suggestions from prospecting, as the guide to Advantage+ audience reaching existing customers describes.
- Missing deduplication. Browser and server copies of each purchase need the same event name and event ID, or both are counted, as the Meta Conversions API guide explains. Fix the event ID before trusting any reported return.
- Compare at account level and protect brand. Performance Max may simply have taken brand queries. Add a brand exclusion, keep exact-match brand terms in Search, and compare account totals before and after launch, as the Performance Max structure guide advises.
- Each significant change restarts learning. Both platforms document a learning period after major edits, so a daily change means the result can never be read. The guide to AI in ad platforms advises measured steps, left long enough to judge.
- Only a comparison with a holdout can say. Platform reports count conversions the campaign touched, including ones that would have happened anyway. Record the business baseline and run a holdout by region or period, as the incrementality testing guide explains; failing that, watch the business KPI beside the platform figures.
Reading your score
A learner marks themselves out of 8. With 7 or 8 right, they move on to the automation and analysis module. With 5 or 6, they reread the sections behind the misses and revise their Spadewell guardrails. With 4 or fewer, they redo the exercise with both module guides open. Misses on questions 1 and 4 matter most, because they point the whole budget at the wrong outcome or the wrong people.
Where to go next
If signals or exclusions caught you out, redo the garden tools retailer guardrails exercise. Otherwise return to the intermediate AI in marketing level and start the automation and analysis module, whose exercise is automating report commentary for an accounting firm.
