A quick A/B test before the numbers. Choose freely.
Paid Social case, Tech Company
A read of Q1 2025 on Meta, and a plan for the $1,000,000 Q3 launch.
Prepared for DEPT by Martin Borovski.
Download the presentation and workbook
Present with → ← or scroll
The short version
Part 1. The report table
The client export had 55 rows, 31 of them empty, and a February 29th that never happened. This is the cleaned version. The table below is the same one that is in the Excel file.
Before trusting the table
Each of these changes the answer. This is what I checked, and what I did about it.
| The trap | What happens if you miss it | What I did |
|---|---|---|
| Q1 means January to March | The file runs to May 27. Use all of it and four more countries appear above the ceiling. | Report is filtered to Q1. April and May sit on a separate tab. |
| Incremental CPA goes up, not down | Multiplying CPA by 20% makes every country look five times better than it is. | Spend / (purchases x 20%). A $1,000 ceiling is a $200 platform CPA. |
| Totals are not averages | Averaging six country CPAs gives $419. The real account CPA is $62. | Every total is recomputed from summed spend, clicks and purchases. |
| CTR is already a percentage | Format the source column as a percent and CTR reads 16% instead of 0.16%. | Recalculated from link clicks / impressions. |
| Frequency is not in the file | It has to be derived, and monthly reach cannot simply be added up. | Impressions / reach, reported as average monthly frequency with a note. |
| Dirty rows | A February 29th, 31 empty rows, blank purchases, revenue on $0 spend. | Fixed or removed, each one logged on the Data Notes tab. |
| The obvious suspect is wrong | CPC and frequency look like the story. They explain almost none of the gap. | Walked the funnel step by step. See the next two sections. |
Who is above the ceiling
Incremental CPA by country, Q1 2025. Spend divided by the purchases that would not have happened without the ads.
Move the incrementality slider in the report section and this chart follows. At 40% incrementality Switzerland drops under the ceiling; Germany never does. The whole verdict rests on one assumption, which is why testing it is experiment number one.
What stands out
Each KPI for Switzerland and Germany next to the four efficient markets pooled. Everything before the click looks normal. Everything after it collapses.
Before blaming the market
Before Q3
The greatest issue is click to purchase conversion. More budget, new audiences or new creative will not move it. In order:
Reconcile Meta purchases with shop orders for Germany and Switzerland in March. Check pixel and Conversions API, consent banner behaviour and event match quality. In the same audit, confirm New Zealand's location targeting and country breakdown.If the shop saw orders Meta did not, this is a measurement problem and the fix is a week of work.
Language (German, French and Italian for Switzerland), currency in CHF and EUR, delivery time, duties and returns stated up front.
Local payment methods, guest checkout, full shipping cost before the last step. Hypotheses to confirm with site analytics: TWINT in Switzerland, PayPal and pay-by-invoice in Germany.
One consolidated German-speaking campaign instead of thin country tests. Optimise for add to cart until 50 purchases a week is realistic, then switch to purchase.10 purchases in a quarter cannot train a purchase-optimised campaign.
Hold both markets at test level with a $200 cost cap, which equals $1,000 incremental. Release launch budget only when the conversion rate holds for two weeks.
Part 2. Channels
A considered purchase at around $730 a basket needs reach first and a conversion engine second. Each channel has one job.
The conversion engine in every market.
Demand creation around launch week.
Where a tech device gets researched before it gets bought.
Local presence in Japan, Taiwan and South Korea, at test size.
Not in the plan: LinkedIn (wrong mindset for a consumer device), Snapchat and Pinterest (small in the three new markets). Shares are of the full $1,000,000; the remaining 12% is the unassigned reserve.
The rough media plan
Keep the products that already sell running, launch the new device where it can be measured, enter Asia with enough to learn, keep weak markets on a short lead and hold a reserve for September.
28% of the budget, in proven markets only.
$200k in July: $120k existing products, $80k new product.
Part of the core group, at run-rate until explained.
All shares are planning assumptions and sit in editable cells on the Q3 Media Plan tab of the workbook. Launch is assumed in early August. Australia moved from growth to gated: its April test ran at $1,943 incremental CPA.
Meta campaign build
The structure exists to concentrate signal. Every split costs learning, so each one needs a reason: language, market maturity or funnel stage.
| Campaign | Objective and optimisation | Budget and bid | Markets |
|---|---|---|---|
| Always-on sales | Advantage+ sales and catalog ads for existing products. Purchase. | Campaign budget (CBO), cost cap $200 | Core markets and Poland, all quarter |
| Launch tease | Waitlist sign-ups (or pre-orders), plus reach and ThruPlay. | Campaign budget, frequency cap 2 a week | All launch markets, grouped by language |
| Launch sales | Advantage+ sales for the new product. Purchase. | Campaign budget (CBO). Highest volume first, cost cap $200 after learning | UK, New Zealand, France, Belgium and Poland |
| Asia sales | Sales. Add to cart, switching to purchase at 50 a week. | One campaign per country, own budget so none is starved | Japan, South Korea, Taiwan |
| Gated markets | Sales. Add to cart until volume allows purchase. | Cost cap $200 from day one, budget gated on conversion rate | Germany, Switzerland, Austria, Italy, US, Australia |
| Retargeting | Advantage+ catalog ads. Purchase. | About 10% of Meta spend, capped | Sign-ups, video viewers, visitors, cart abandoners |
The extra 10%
$100,000 of test budget. Each one has a decision attached before it starts.
Next in the queue if budget remains: cost cap against highest volume bidding, and TikTok's incremental reach on top of Meta.
What the brief does not say
The plan works with assumptions, as the brief allows. These are the ones that would change it.
The link between both parts: Switzerland and Germany failed after the click because the visit was not local. Japan, South Korea and Taiwan are the same risk at seven times the budget.
One more thing
Forty-eight seconds. No sheep were retargeted.
Martin Borovski
The presentation covers the same analysis in a formal 19-slide format, with speaker notes.