# The Advertising Commons

Before you say anything: digital advertising is about $600 billion dollars annually
(in 2023). At the current pace, the industry will overtake the US DoD's budget. We can
either compete with companies that are capturing that revenue, or we can seize it and
put it to work supporting the commons.

The rest of this document assumes that the latter is preferable.

## Architecture Overview

Buyers (who want to place ads) and sellers (who have an available inventory of ad
spaces) are connected for deals that may be mutually agreeable over a P2P protocol
(similar to Filecoin). Deals are not for single ad impressions (as is done today
over Real-Time Bidding) but likely for larger impression bundles (essentially they
discover potential automated Insertion Orders). This forms a set of
*potential matches*. This discovery mechanism includes the (first-party) segments
that the seller can offer based on its knowledge of its audience and contexts.

Potential matches enter an auction (per seller per unit time) which is orchestrated
through a mechanism with reliable and auditable properties, like
[PASTRAMI](https://research.protocol.ai/publications/pastrami-privacy-preserving-auditable-scalable-trustworthy-auctions-for-multiple-items/).

The buyers who win an auction with a given seller have their ad creatives set up
at content-addressable locations (as tiles) and the campaign parameters are entered
into the seller's ad server/yield management system. Ad creatives must have trustworthy
(signed) metadata about the party that is being advertised for.

At impression time, the publisher makes sure that creatives are selected to match
campaign parameters, ensuring that pacing is correct, etc. (This is the same as it
is today.)

Because the ad creatives are tiles, they cannot exfiltrate information. (This is
in the same vein as proposals to use fenced frames today.) That's highly desirable
for privacy purposes, but we need to address a few related issues:

- Private serving from IPFS to serve the ads.
- Some integration with the existing work on private retargeting. A lot of digital
advertising is about eg. getting people to complete a purchase, funnel-based
marketing in general.
- Viewability and generally certification that an ad was served and shown to a
real person.
- Aggregate measurement so buyers can know when and which ads work. Work like
[IPA](https://docs.google.com/document/d/1KpdSKD8-Rn0bWPTu4UtK54ks0yv2j22pA5SrAD9av4s/edit#) or
[ARA](https://developer.chrome.com/docs/privacy-sandbox/attribution-reporting/) might
help.

If the seller does not deliver the campaign according to the provided parameters,
there needs to be a form of automated make-good campaign or they need to face some
degree of slashing based on escrow. (Note however that this shouldn't be massive slashing
as is done for instance with Filecoin — predicting traffic volume and patterns in
advance is impossible and missing campaign goals is common. Recoding how often and
by how much a seller missed their commitments could be a valuable service.)

## Questions

- Is there an opportunity to ride the PATCG/Privacy Sandbox wave for this?

## Capture Threat Model

- [ ] Which decisions does each component make?
- [ ] Who is affected by that decision?
- [ ] How can those who are affected have voice in the process?

## Todo

- [ ] Look at the systems that do automated IOs today
- [ ] Look into FLEDGE as model for integration

---
<small>This work is part of [The Web Commons](/dFpEb1jeSrKp0Fx8IrdZAg).</small>