Why the app report always looks like a win
A delivery marketplace runs a promo for your restaurant. The marketplace's report comes back showing orders up for the week. It looks like a clear win, and in the marketplace's own numbers it is one.
The trouble is what that report can't see. It only counts orders that came through its own channel. It can't tell a brand-new guest from a regular who usually orders on your website and switched this week because the app had a discount. Both show up as the same kind of order.
So the question isn't whether marketplace orders went up. It's whether the promo found new guests, or moved existing ones to a channel that costs you more. That's a question only your own numbers can answer. This isn't an argument against marketplaces. They reach guests your own site never would. It's an argument for checking before you run the next promo.
What follows is our practice framing and a rough read, not a controlled test. We won't quote typical lift or switch rates, because they depend entirely on your guests and your channels.
Check 1: the three-week channel table
Build one table with orders by channel for three weeks: the promo week and the two weeks before it. Use the same stores in all three weeks, and the same weekdays.
| Channel | Two weeks before | One week before | Promo week |
|---|---|---|---|
| Own website | |||
| Own app | |||
| Marketplace A (the promo) | |||
| Marketplace B | |||
| Phone | |||
| Total |
The shape of the table. Fill it from your own exports; the empty cells are on purpose.
Where the numbers come from:
- Marketplaces: each marketplace's order report or payout statement, filtered to the same stores and dates.
- Own website and app: your ordering platform's order export, not an analytics dashboard. You want orders, not sessions.
- Phone: the POS, if phone orders are rung in with a channel or order type that sets them apart.
Matching stores is where this usually breaks. Every system names locations its own way, so a store called one thing in the marketplace may be called something else in your ordering platform. One Store, One Key covers fixing that once.
Check 2: spotting a switch
Look at your own channels in the promo week. If marketplace orders rose and your own website and app fell by a similar amount, that's what a switch looks like. The orders moved. They didn't multiply.
A partial switch is more common in practice: marketplace orders rose, your own channels dipped a little, and the total edged up. Read that as some new demand and some moved demand. You can't split it exactly from this table, but you can see that the marketplace's report overstates what the promo added.
If your own channels held steady while marketplace orders rose, that's the best case: the promo probably reached guests you weren't getting anyway.
Check 3: total orders are the scoreboard
The marketplace line is the promo's scoreboard. The total line is yours. If the total across every channel didn't move, the promo moved orders without adding any, whatever the marketplace report says.
Put the total on the same page as the marketplace report whenever a promo gets reviewed. It's one more row, and it changes the conversation from "orders went up" to "orders went up there, and here's what happened everywhere else."
What a switched order costs
A guest who switches from your website to a marketplace for a promo week still orders the same meal. What changes is what that order returns to you. The marketplace usually charges a commission, and if the restaurant funded part or all of the promo discount, that comes off too.
We're not going to quote rates. They vary by contract and by promo. The point is that a switched order isn't neutral. It's an order you would have had anyway, now arriving with costs it didn't carry before, and usually without the guest contact you'd get on your own site. Your Own Orders vs Marketplace Orders covers that data gap, and Net Per Order by Channel is the weekly table that shows what each channel actually returns.
Limits, and a better test
This comparison is a rough read. It isn't causal, and a lot can move order counts in a three-week window:
- Weather, especially in delivery-heavy markets.
- Holidays and school calendars, which shift the base week to week.
- Overlapping promotions: your own email, a social push, another marketplace's offer running the same week.
- Menu or price changes that landed in one of the three weeks.
When a promo is big enough to justify it, a holdout is a better test: run the promo in some stores and not in a comparable set of others, then compare total orders across channels between the two groups. We're suggesting it as the stronger design, not describing a method we've run for you. It takes planning, and the store groups have to be genuinely comparable.
A three-question self-check
- Can you pull orders by channel, for the same stores and weeks, from your own systems?
- Did your own website and app orders fall during the last marketplace promo?
- Does anyone review total orders across channels when a promo is judged?
If you can't answer the first one, that's the gap to close before the holiday promo season. If you want help building channel-level reporting you can trust, book a data and tracking audit. And if you want to see what your restaurant site loads before a guest makes a privacy choice, request a free privacy scan. We review each request within 48 hours.