From e02330cfa16deb773ae6a6ecb3fa9a1d6b9bb33d Mon Sep 17 00:00:00 2001
From: benjaminsavage <52456851+benjaminsavage@users.noreply.github.com>
Date: Wed, 14 Jun 2023 17:12:21 +0100
Subject: [PATCH] Consensus write-up on "Optimization" Use Case

In our last meeting we reached general agreement that the group wants to support the "Optimization" use case. This is a write-up of that agreement.
---
 measurement-use-cases.md | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)
 create mode 100644 measurement-use-cases.md

diff --git a/measurement-use-cases.md b/measurement-use-cases.md
new file mode 100644
index 0000000..69e6a8e
--- /dev/null
+++ b/measurement-use-cases.md
@@ -0,0 +1,24 @@
+# Measurement Use Cases
+
+The group is currently working on developing a privacy preserving ad measurement system. 
+
+There is general agreement that we want such a system to satisfy the following use cases:
+
+# Advertiser Reporting
+
+TODO: fill this in
+
+# Optimization
+
+Ad selection systems often attempt to estimate the probability that if an ad is shown, it will lead to a conversion event. 
+More sophisticated systems may also attempt to predict conversion values / counts / categories.
+
+The information produced by "Advertiser Reporting" is already sufficient to make naive predictions (i.e. just estimate the average conversion rate for all users). 
+But there is interest in the group to intentionally design the private measurement system to do better than this.
+
+There is general agreement in the group that:
+
+1. This is a use case that we explicitly want to support
+2. Within our agreed privacy bounds we would like to design outputs which provide as much utility as possible for this use case
+3. We are flexible on the mechanism to support this, and potentially envision a private measurement system which supports multiple modalities intended to support this use case.
+