The column no partner can fill
Somebody asks for a marketing dashboard with impressions and return on ad spend. We went looking for where those numbers would come from, across 145 pages of a delivery platform documentation. They do not exist.
A screenshot arrived of a marketing dashboard for delivery platforms. Nice table: offer name, orders, average order value, return on ad spend, impressions. The obvious job is to build the same table.
Before building it we went looking for where each column would come from. Two of the five have no source.
What a delivery platform actually gives a partner
We read the whole published interface — 145 pages of it. There is a promotions capability: create an offer, choose the mechanic, pick the items, set the window, aim it at a customer group. Eleven kinds of offer, seven customer groups. That is real and it is well specified.
There is no advertising capability. Nothing for buying placement, nothing that reports how many people saw your listing, nothing that returns a spend figure to divide revenue by. Not hidden somewhere, not in beta — absent.
So where would the numbers come from?
They would have to be made up. Not maliciously — you can build something that looks plausible. Impressions could be estimated from orders by applying a conversion rate somebody chose. Return on ad spend could be computed against a spend figure typed in by hand, or assumed.
Both would appear on screen looking exactly like the three columns beside them, which come from real orders. That is the problem. A person reading that table has no way to tell which two numbers are measurements and which two are arithmetic performed on a guess, and they will make a decision using all five.
The columns that are real
The measurement half of that table needs nothing from the platform's interface at all. When an order arrives on a webhook it carries the discount applied, at the order level and per line. So:
- Orders on an offer — countable, from your own orders.
- Average order value with and without — countable.
- What the discount actually cost you — countable, and more useful than a ratio.
That is a table worth having, and every figure in it comes from something that happened rather than something inferred.
The general version of this
When a supplier shows you a dashboard, pick the column you would act on and ask where the number comes from. Not "is it accurate" — ask what system produced it and what happens when that system has nothing to say. A figure with a good answer to that question is worth ten with a confident tone and no provenance.
We would rather ship three columns that are true than five that look complete.