When I took over on 28 August 2026, the shop (a Base44 store selling flowers and gifts across Jordan) had been running Meta ads for ten days on an account a previous team had built: seven campaigns at $3 to $12 a day each. It was spending $22.63 a day for fewer than 3 paid orders a day, the same rate the store had in the first half of August with no Meta ads at all.
10
paid orders a day over the last six full days (30 Sep – 5 Oct)
×4+
the daily orders from before I took over
8.2x
real return on Meta spend, last six full days
$4.98
Meta spend per paid order, down from $9.43
What I inherited
- Meta, Google Analytics and the store each showed a different number of sales, so no decision could be trusted. Google Analytics showed zero revenue since the store opened, because there was no purchase tag at all.
- The thank-you page never reported a sale, and the cookie banner ran on European rules (GDPR) in Jordan, so tracking stayed off until a visitor clicked accept.
- Two of every three paid clicks never loaded the page. A click looked like $0.08; a visit that actually landed cost $0.52.
- The retargeting audiences held about 20 people despite 2,000 add-to-carts, because browser events arrived with nothing that identified the person.
- The budget was split into $3 to $5 ad sets that never left the learning phase.
Phase 1 · Get numbers I can trust (28 Aug – 10 Sep)
The first two weeks were about measurement, not sales: every later decision depended on it.
- I rebuilt the measurement around the store's real orders, so every decision was judged on sales that actually happened.
- I added the missing GA4 purchase tag and tested it with a real order. Revenue started recording the same day.
- Before trusting any 'winning' ad, I checked whether its numbers were possible. Only two of the eight held up.
- I wrote every stop, keep and scale rule before launch, from the account's own medians.
Phase 2 · Pool the budget around the idea that sells (11 – 29 Sep)
- In Jordan, every flower shop answers "DM us for the price". This one shows it. The price-transparency images brought 52% of purchases on 41% of spend, so the shown price became the brand's position, not one angle among many.
- I moved the proven angles into one campaign with a minimum spend per ad set, instead of a dozen starving ad sets. The "no need to ask the price" image became the engine: by 26 September it brought 86% of the account's purchases.
- On Google Ads I fixed how conversions were counted, stopped Performance Max (it was spending on bonsai trees, chocolate and the shop's own name) and launched two tight Arabic search campaigns.
- I tested four occasion angles (birthdays, get well, graduation, two-hour delivery) with 14 creatives. None beat the price. That loss told me what the brand is.
Spend doubled to $41.69 a day and paid orders doubled with it, to 5.3 a day, while the real return held (5.23x before, 5.39x now).
Phase 3 · Feed the winners, test beside them (30 Sep – 5 Oct)
- The winners went together in the main campaign: the "no need to ask" image, the price image, the delivery video and the customer (UGC) video, running as the original posts so their likes and comments travel with them. Its budget went to $50 a day.
- A small test of new angles ran beside it, excluding recent buyers. One angle brought 7 purchases on $18.85.
The last six full days: 10 paid orders a day, $408 of revenue a day, an 8.2x real return, and $4.98 of Meta spend per order.
Paid store orders (cancelled and zero-value orders excluded), averaged per day in each phase. The last phase ends on 5 October, the last full day before the data was pulled.
Show the data
| Paid orders a day | |
|---|---|
| Before me, 18–27 Aug | <3 |
| Diagnose, 28 Aug–10 Sep | <3 |
| Pool & scale, 11–29 Sep | 5.3 |
| Winners, 30 Sep–5 Oct | 10 |
Weeks run Monday to Sunday. I took over on Friday 28 August, in the second week shown.
Show the data
| Paid orders | |
|---|---|
| 17 Aug | 15 |
| 24 Aug | 18 |
| 31 Aug | 20 |
| 7 Sep | 23 |
| 14 Sep | 39 |
| 21 Sep | 31 |
| 28 Sep | 62 |
Real ROAS = store revenue ÷ Meta spend. Meta's ROAS uses its default attribution (7-day click, 1-day view).
Show the data
| Meta dashboard | Store (real) | |
|---|---|---|
| Before me, 18–27 Aug | 4.77x | 4.75x |
| Diagnose, 28 Aug–10 Sep | 6.86x | 5.23x |
| Pool & scale, 11–29 Sep | 4.66x | 5.39x |
| Winners, 30 Sep–5 Oct | 7.78x | 8.2x |
Meta spend ÷ every paid store order in the phase. Some orders came from organic, Instagram messages or Google, so the true cost per ad-driven order is higher.
Show the data
| Cost per order | |
|---|---|
| Before me, 18–27 Aug | $9.43 |
| Diagnose, 28 Aug–10 Sep | $8.22 |
| Pool & scale, 11–29 Sep | $7.92 |
| Winners, 30 Sep–5 Oct | $4.98 |
40 days in four numbers
$1,436.89
Meta spend, 28 Aug – 6 Oct
$8,432.91
store revenue (5,979.09 JOD)
5.87x
real return on Meta spend
$41.75
average order value
Reading the result honestly
Store revenue includes orders that would have come anyway. Counting only the revenue above what the shop made without Meta (against two baselines: 1 to 17 August, and July), Meta's return is between 2.4x and 4.2x. Spend on the two small Google search campaigns isn't included in any of these figures.
Meta credited its ads with 181 purchases and the store received 202 paid orders, so every result here uses the store's number.
In a market where every shop says "DM us for the price", showing the price stopped being an angle. It became the position.
What's next
- The landing page: two of every three paid clicks still don't load it. That's the biggest free lever left.
- Retargeting from the store's own customer list instead of the pixel.
