Privacy ​

Not intended for legal use

The following document is not a legal document, it is designed to give a simple and easy to understand high level overview of the privacy considerations of the AirGrid platform.

Design philosophy ​

The AirGrid platform is built around the idea that an individual's personal information should always remain personal, which simply means any data which can be tied to a single person should only ever be under the control of that person.

The platform achieves this by combining two technologies:

Edge computing allows us to store personal user information at the "edge" which essentially mean locally on the user's device. So instead of storing events such as ad clicks on the server in the cloud, we store that information locally in the browser using a technology such as localStorage or indexDb.

We can now use this information stored on the device to learn a model of user behaviour, by leveraging federated learning which involves training a machine learning model across a distributed dataset. This means rather than collecting all the data into a single store and allowing our model access to it, we send our model to the location of the data, which in our case is the browser of each individual who has interacted with our technology.

This approach means we can learn a generalised model of user behaviour without the need to collect and store any data tied to unique user IDs, cookie IDs, MAIDs etc.

NOTE: A deep dive into the specific technologies is out of scope for this document.

Data storage ​

We broadly split the data which is stored into two buckets; user level and non user level. All user level information is stored in the browser, and non user level data is collected on the server for fraud, analytics & reporting purposes.

User level ​

User level information can be summarised as page view events, these events are encoded into dense vectors and stored locally, they contain information such as:

  • Context of the page.
  • Imagery of the page.
  • Structure of the page.
  • User interactions on the page eg dwell, scroll & clicks.
  • Browser based API outputs eg Chrome topics.

Retention ​

Note: this section is related only to data stored / retained in the browser of the individual who has generated that event.

The AirGrid software development kit (SDK) limits data storage to a maximum of 90 days or 300 distinct events, neither of these limits can be breached. This means the oldest event a user would store is 90 days, unless they hit the 300 event limit. In most practical cases the browser will evict data from the cache much earlier. The browser has a mechanism designed to purge data which is not used / accessed frequently.

Non user level ​

Non user level information can be summarised as aggregated metrics on platform usage. We collect and store a number of variables, but none are tied to a unique user identifier:

  • Time
  • Device
  • Domain
  • URL
  • Location (zip code / post code)

IP Address

For any networked technology to function an IP address is received by the server, this is the address to which any requests are made, and subsequently returned. We use the client IP address only for blocking of malicious actors and geo resolution.

We do not use IP for any form of user identification.

Retention ​

We store raw server logs for 365 days, after which aggregates are created and archived.

The AirGrid platform is not designed to circumvent any consent frameworks. We firmly believe that internet users should be made aware of how technologies are using their information. We are a member of the IAB Europe's - Transparency & Consent Framework (TCF). We are listed under the vendor number 782.

Whilst most server side technologies require receiving a "consent string" which encodes user consent choices made in a Consent Management Platform (CMP) deployed by the publisher, AirGrid does not rely on this string due to our presence in the browser. We are able to make a binary check - "do we have consent?" and act accordingly. This is a significant improvement over the usage of the consent string, due to the fact that the amount of entropy inside a consent string is high enough to be used as a user identifier itself.

When a user is shown a CMP pop-up on arrival to a publisher's site, two options are possible:

  • Consent is granted: this will trigger for our code to run, storing information about the page on which the user is on in their browser & executing segmentation / audience modelling logic.

  • Consent is not granted: this will mean our code does not run, the only thing we do is check if we have previously stored any information (due to having previously had consent), and if yes we clear this - deleting the only copy of the data.

If a user later decides to opt out, their information will be removed from their browser and this means they have cleared the single source of this information. This is a huge improvement over the current industry norms such as "soft deletes" or trying to ascertain how many server side systems have stored, sold & propagated user information.

Browser storage

If a user decides to delete their browser storage all AirGrid set data will be cleared.

Summary ​

  • The AirGrid platform leverages on device technologies to store and model audience data.
  • Any user tied information i.e personal information is stored on the device, and ONLY on the device.
  • Aggregated information is collected on the server to power platform metrics.
  • We are a member of the TCF and run our code only after gaining consent from the user.
  • If consent is retracted, we purge all previously stored information from the device.
  • We do not rely on the consent string, but a much simpler binary consent check.
  • Clearing of browser storage will also clear our information.

FAQs ​

In which geographies are AirGrid servers located? ​

We route traffic from users to servers located in EU or NA depending on where the request originates from. However all data is stored in the EU after the request has been processed. UK users are routed to servers located in the EU.

Does AirGrid rely on legitimate interest? ​

No, we only ever process user information, or access the device once we have gained consent, in the publisher's CMP.

How can a user opt-out? ​

Opt-out is handled by the publisher CMP, since we are not storing any cross site cookie or personal identifiers we would not be able to support opt-out globally. There is no user ID to tie activity of an individual across sites. We respect any global browser settings such as Do Not Track (DNT).