Split Your Cancellation Rate Before You Try to Fix It

When I ask a brand owner for their cancellation rate, the reply is one number: "around 30%". The number is correct. It is also no use to anyone.

It is an average, and it blends things that have nothing to do with each other. A customer in Cairo refused because delivery ran two days late. A customer in Upper Egypt refused because the product looked different from the video. A third was called two days after ordering and had forgotten placing the order at all. Three problems, three fixes, one number hiding all of them.

Which governorates are driving your cancellations?

Start here, because it is the fastest split to return an answer. Export at least 30 days of orders with their final status, group them by governorate, and calculate the delivery rate for each one.

In most Egyptian stores the gap between the best governorate and the worst one is not a few points. It is often double. The same ad at the same cost brings an order whose real value changes completely depending on where it lands.

The first thing to do with this is not to switch the weak governorates off. First find out whether the problem is the courier there or the audience itself. The fourth split answers that.

Which products should you stop selling on cash on delivery?

The threshold I use: a product that stays below a 60% delivery rate for two months straight does not need optimising. It needs to be stopped, repriced, or sold prepaid only. Some products keep a low delivery rate whatever you do, because the price sits above the value people see in them, or the category itself carries a lot of hesitation at the door.

That product poisons the account in a way that is hard to see. It takes spend and it returns orders on paper, so Ads Manager looks fine while the cash never arrives. That is why the verdict on it comes from the back end, not from the ad account.

What do the refusal reasons tell you?

Sorted properly, the refusal reasons usually show that about 50% of cancellations are a marketing problem, not a shipping one. Most couriers record a reason for every refusal, and most brands never read it, because it is handwritten and messy. Put it into five buckets and it becomes a map of who owns what:

That is the opposite of what most owners assume. They blame the courier first, and half the time the cause was set before the parcel ever left the warehouse.

Does changing your courier reduce cancellations?

Sometimes, and you only know when you split by governorate and courier together. A governorate that delivers below 60% with one courier and well above it with another does not have an audience problem. It has a courier problem.

That changes the decision completely. "Stop advertising in Upper Egypt" and "change the courier that covers Upper Egypt" are the difference between closing a market and opening it. If the rate stays low with both couriers, excluding the region is a sound financial decision.

Which ads bring the orders that get refused?

This one is not in the core four because it takes more work, but it pays back the most. Join each order to the ad that brought it, over the same 30 days, and calculate the delivery rate per ad.

What shows up almost every time: one or two ads with double the refusal rate of the rest. Those ads are usually the same ones with the lowest cost per order, because they promise more. So in Ads Manager they look like the winners.

Does the pixel learn from orders that get cancelled?

Yes. The purchase event fires when the order is placed, so a refused order teaches the platform exactly as much as a delivered one. That is how an account I re-measured reported 5.48 ROAS in Ads Manager while the real figure was 1.80, once view-through, cancellations and unsettled cash-on-delivery orders came out. An ad that attracts refusers keeps getting more budget, because by the account's count it is selling. Until you judge every ad on delivered orders from the back end, the algorithm is optimising for people who say yes on the phone and no at the door.

What does a refused COD order actually cost?

More than the ad that brought it. If you pay 100 EGP per order and your delivery rate is 70%, your real cost per delivered order is about 143 EGP, plus the shipping you paid on the orders that came back. The cost of a placed order is not the cost of a paid one, and I go through the full math in why your COD ROAS is lying.

So the split is not an analysis exercise. Its output is one number that goes into every bid: your real target CPA. The calculation is a free tool, the COD break-even calculator: it takes price, COGS, shipping and delivery rate and returns the most you can pay per order.

Skip that step and the damage is quiet. On a 618-product catalogue I worked on, break-even was 1.77 ROAS on paper and 2.11 once cancellations were counted, so campaigns clearing 1.9 looked profitable and were losing money on every delivered order. The write-up is in break-even before scale.

Does asking for prepayment reduce COD returns?

It fixes the orders paid that way, and only a small share of the Egyptian market pays upfront. My Offer Builder models prepaid orders at a 97% delivery rate with a 2.5% gateway fee, which is why a small discount for paying online usually earns its keep. The mistake is blending them. Track prepaid and cash-on-delivery orders separately, because their economics are completely different and one average takes you straight back to the single-number problem.

How do confirmation calls affect the delivery rate?

Speed decides it. A WhatsApp message the moment the order lands, then a call within 2 hours, catches the customer while they still remember ordering. A call two days later reaches someone who has cooled off. For the "not there, not answering" bucket, the call is also where the address gets confirmed before anything ships, so the courier is not sent to a door that will never open.

In what order should you run the split?

Work through it on at least 30 days of final-status orders, in this order:

1. Export orders with their final status for the last 30 days at least.
2. Calculate the overall delivery rate. That is your baseline.
3. Split by governorate, then by governorate and courier together.
4. Split by product, and stop anything below 60% for two months.
5. Sort the refusal reasons into the five buckets.
6. Join orders to ads if tracking allows it, and stop the ads with abnormal delivery.
7. Feed the delivery rate back into your target CPA and reset the bids.

This split is step eight of the full audit in my method, and it comes after the courier review in the diagnostic phase. If you want it run on your own store, send the link to the free audit: three numbers within 48 hours, the most you can pay per order, ROAS after delivery, and the biggest leak.