My own money, a commodity product, and no one to blame but the data.
AlphaMart was a hair-styling-tools store I ran on Shopify with a partner, selling cash-on-delivery in Pakistan over an eight-month run. There was no client budget to protect, which is exactly why I went back afterward and audited it as hard as I'd audit a client's account.
Reconciled every platform export against the bank, then followed the money past the ad dashboard.
- Rebuilt the full funnel from Snapchat, Meta and TikTok exports and courier settlement files, down to the individual order
- Traced every order from click to cash — not just "purchases," but how many of those purchases actually survived confirmation, shipping, and returns to become a delivered sale
- Compared Snapchat, Meta and TikTok on cost-per-click, cost-per-thousand-impressions and real conversion, not just the ROAS each platform reports about itself
- Compared three products on real delivered performance — order value against delivery rate — instead of stopping at the listed price
- Split delivery failure by courier and by city to find out whether a bad delivery rate was the courier's fault or the market's
1,092 orders were paid for by the ad accounts. 523 became cash.
Ad platforms report an order the moment a customer checks out. In Pakistani cash-on-delivery, that's the start of the funnel, not the end. Nearly a third of orders were cancelled on the confirmation call before a parcel ever moved, and another fifth came back as returns.
Confirmation-calling every order before dispatch caught roughly 325 of the bad ones early, saving an estimated PKR 78,650 in courier fees that would otherwise have gone out and come straight back.
Snapchat bought attention 4–6x cheaper than Meta. TikTok bought nothing.
| Measure | Snapchat | Meta | TikTok |
|---|---|---|---|
| Impressions | 11,079,368 | 389,700 | — |
| Cost per 1,000 impressions | PKR 42.53 | PKR 174.65 | — |
| Cost per click | PKR 1.95 | PKR 11.44 | — |
| Click-through rate | 2.19% | 1.53% | — |
| Reported ROAS | 5.63x | 2.73x | 0.00x |
Meta ran roughly PKR 15,000 of budget on campaigns that never produced a sale — left on because nobody was checking below the account-level number. TikTok spent PKR 6,412 and produced about five orders, all cancelled at confirmation.
The highest-priced product also had the best delivery rate.
| Product | Order value | Delivery rate |
|---|---|---|
| 5-in-1 Hair Styling Tool | PKR 3,894 | 76.3% |
| Rechargeable Mini Comb | PKR 2,475 | 71.2% |
| Hair Straightening Brush | PKR 2,004 | 70.5% |
The pattern ran opposite to what we expected going in — the cheapest item did not deliver the best. A buyer paying more per order seems to think it through more before ordering, and follows through more often when the rider arrives.
Stock buying didn't reflect this at the time: the best-performing product was reordered the least, and the weakest was reordered the most. That's the gap the next round of buying is built to close.
The worst-performing courier just happened to ship the most parcels to the worst-performing city.
Read on its own, one courier looked far weaker than the others. Split by destination, every courier underperformed in the same city — the courier gap was mostly geography wearing a different name.
| City | Parcels | Delivered |
|---|---|---|
| Karachi | 47 | 85.1% |
| Islamabad | 20 | 75.0% |
| Rawalpindi | 23 | 65.2% |
| Lahore — second-biggest market | 78 | 59.0% |
Eight months of testing across three ad platforms, three shipped products and two couriers produced a working funnel model, a channel comparison worth acting on, and a clear read on which products and which cities were actually worth scaling.
- ✓The media buying was genuinely excellent — 11 million impressions for PKR 471,193, a click-through rate of 2.19%, and video ads that held attention through nearly 40% of a 2-second view.
- ✓Confirmation-calling every order before dispatch is a habit that transfers to any store — it saved roughly PKR 78,650 in courier costs that would otherwise have been spent and refunded.
- ✓We switched from dropshipping to owned stock mid-flight once it was clear dropshipping was capping how much of the operation we could actually control — the kind of call that's easy to avoid making and costly to delay.
- ✓Eight product types were tested across the run, not one — which is how the best-performing product got found at all.
- 01Never sell a product that's price-anchored on Daraz or Temu — if a customer can price-check it against ten identical listings in ten seconds, you're competing on price before the ad even runs.
- 02Judge a product on how it performs through the whole funnel — order value and delivery rate together — not on how good it looks on the product page.
- 03Track delivery rate by city from day one, not just by courier — a 26-point gap between two cities went unnoticed for months because nobody split the data that way.
- 04Offer a prepaid discount from week one — prepaid orders return at roughly a quarter of cash-on-delivery's rate.
- 05Buy stock in proportion to what's already proven to sell — AlphaMart's best product got the least stock and its worst product got the most.
- 06When a month works, put more into it — the best month of the business was followed by a 76% cut in ad spend, for reasons that had nothing to do with performance.
