Attribution for delivery and on-demand apps

Four things that set on-demand apps apart

A food delivery, grocery, ride or home-service app sells something physical, in a place, at a time. That changes what attribution has to answer.

• Location decides value. An install outside your service area cannot order.
• The sale can fail after checkout. A courier may be unavailable or a restaurant may reject the order.
• Profit comes from frequency. The first order is often sold at a loss through a promotion.
• There are two user groups. Customers and the people who deliver or drive are acquired separately.

Report by area, because demand is local

Run and name campaigns per city or zone, and keep that name in every report. A national average hides the fact that one city is saturated while another has couriers waiting and too few customers. Area-level reporting also shows waste quickly: installs that never reach a first order can come from places you do not serve.

Offline marketing is common in this category: flyers, bags, posters in partner restaurants, vehicle signs. Give each a QR code on its own tracking link, so a scan can be followed to an install and an order in the same way as an ad click.

Count the order when it is fulfilled

Send two purchase-type events. "Order placed" comes from the app at checkout. "Order completed" comes from your server when the delivery or ride is finished, with the final amount after discounts. Tie both to one order ID. Each is an in-app event in the attribution tool, even though the second one originates on a server.

Judge campaigns on completed orders. Report placed orders to ad platforms only if completed orders are too few for them to learn from, and move to the stronger event when volume allows.

Promo codes and the first order

First-order discounts are a common acquisition offer. The ad says "free delivery on your first order", and a new user who installs and finds no trace of that offer feels misled. A deferred deep link can carry the code through the store and apply it on first open, so the offer is waiting in the basket.

Discounts also distort results. A campaign with a generous code will win first orders at a low cost and may bring people who never pay full price. For every campaign, read two numbers together:

The second number is a form of retention, measured on orders instead of app opens. Over a longer period it becomes LTV. Lapsed customers are usually brought back with push, email and retargeting ads, and crediting those orders correctly depends on re-engagement attribution.

Measure the supply side separately

Couriers, drivers and service providers usually have their own app. Treat it as its own project in the attribution tool, with its own funnel: install, application submitted, approved, first job completed. The conversion that matters is the first completed job, and like an approval it is confirmed on your server. Do not mix supply-side installs into customer reports, since they would lower every customer metric and blur both pictures.

Where to begin

Add the "order completed" server event first, with an order ID and the final amount. Most other readings on this page depend on it. Without it, campaigns can be judged on orders that were never delivered.