Meeting minutes
<wseltzer> Valentino Volonghi] https://
<kleber> Hi folks, sorry I won't be in today's meeting, I'm driving my kid to summer camp
Wendy: Welcome folks; hope you are all in places where you are able to stay cool
Wendy: Looking at today's agenda, we will start with agenda curation and introductions
… then look at "PAURAQUE"
Agenda-curation, introductions
Wendy: Ad Topics from Ben Savage, then dashboard highlights
… Any other business?
… I saw on list request to talk about FloC and the end of the origin trial
… if we have time at end of meeting we can address that
… Looking to see if we have our proposers here on the call
… Start with any introductions; anyone joining the call for the first time
[please mute if not speaking]
… Do we have Valentino on the call?
… This is Jabari from NextRoll
Wendy: Thank you, Jabari
PAURAQUE: Profile-based API for User Attribute Queries [from
Wendy: let's move to Jabari and PAURAQUE
Jabari: yes, looking for share screen
<wseltzer> https://
Jabari: Hello everyone, I am a staff engineer at NextRoll
… want to talk about proposal called PAURAQUE
… user attribute target without cross-site tracking
… in browser managed @ database
<wseltzer> https://
Jabari: B2B based on user attribute targeting
… target job function or firmagraphic data, all considered user attributes
… mixture of IP address and domain name resolution to find name of employer
… and then linking cookies from third party sources
… allows you to leverage third party tech
[talking fast]
… the original motivating use case was the B2B advertising
… B2B differs from B2C, focused on identifying key business accounts and firmagraphic data
… and then advertising context based on business role
… makes B2B marketing fundamentally different from B2C
… makes it unnecessary to do behavioral marketing
… more interested in job title and employer
… core use case, people at companies, particular job function, report spend clicks and page views at a company level
… Our proposed solution centers around four things
… First is in-browser profile
… second is user control over profile
… third is which sites can access profile
… fourth is attributes only available within privacy sandbox
… Proposals like TurtleDove and FLoC are building behavioral profile based on user activity
… We propose extending...user can choose to share profile with other parties
… Chrome linked to Google; Safari to Apple; FF to Mozilla
… our concept standardizes profile and lets user share attributes about selves through API
… two objects: set of positive and negative interests, domains to access; and a profile with geo location, job title and company
… this could be more extensive set
… those are key ones for B2B advertising
… and would also contain a deny list to block sites from accessing this profile
… Aspect of user control
… user enters profile data through browser settings
… things like geo, company, job title
… three ways
… explicitly declare settings in browser setting
… second is site allows to ask user
… categories they are currently viewing
… interest categories subject to K@
… third way is to allow user agent itself
… interest based on browser history
… instead of opaque identifier setting cohort
… interest category would be transparent to user and sites consuming data
… user can be prompted
… or user can add without explicit permissions
… user can remove profile data at any time
… or add any site to a deny list to block access to their interest or profile data
… As mentioned before, access to user profile and interest would follow TD and FLEDGE model
… to prevent first party data from being tied to user's profile
… if campaign is targeting CEO of Microsoft inside the sandbox, @ could see the user
… with a minimum audience size, ad would not be displayed
… targeting much larger than one
… so maintain privacy
… through that mechanism
… We have questions
… trustworthiness of profile data
… user can put nothing or a lot into profile
… verified attributes that advertiser can be sure of
… might generate higher value during the auction
… Question of having a third party verifier
… service that would examine first party data of user and issue them some sort of certification or trust toek
<wseltzer> https://
<wseltzer> https://
<wseltzer> https://
s/token
… verify this attribute on certain data
… in case there is value for advertiser to guarantee that attribute is accurate
… Next is about how data flows in the auction
… depending upon if this flows in FLEDGE or PARAKEET
… to determine decisions
… passing data straight into @
… exposes data to be joined with other request data which would reveal too much
… one idea is to request another request, like user attribute request which contains this data
… and prevents from being joined in contexual request
… trusted server could act as @
… or create an IG type structure for a set of attributes
… seeing this user before
… and add them to an interest group like structure
… Another open question is reporting size of attributes
… reporting APIs
… establish size of groups being targeted
… essential provide some sort of reporting mechanism
… to estimate size of audience for a set of attributes or size
… Another question should there be a standard set of interests
… modelled after AI
… or be a string and allow market to establish what these categories are themselves
… that is all
Brendan_IAB_eyeo: we have gone to market with crumbs doing some of things what is described here
… with IG on the client side
… excited to see this discussion go further
… be supportive and be aligned with that
… we are observing user browsing behavior and challenges getting that data out
… in a private, trusted way as you outlined here
… IAB hat
… ways of getting data out should be looked at...@ defined audience and context signalling
… how to transport these blobs of user attributes
… really appreciate your presentation
… and getting some consistent thinking
Jabari: thank you
Wendy: Thank you
… how would you further follow-up?
… issues in your repository?
Jabari: yes, you can post issues in our repository
… in the AdRoll
… you posted link in the chat
Wendy: yes, it is there and already coming up in our dashboard
… thanks a lot for sharing this
Ben, FB: seems to me this proposal is very similar to FLEDGE
… seems that fact that web site is adding someone, like job description
… is like adding someone to an IG
… seems like the main difference is the possibility of someone going into browser to configure their job, company
… as well as this consent prompt, which I don't recall being a part of FLEDGE
… seems this is accomplished through FLEDGE
… skeptical about seeing people go in and create their profile for sake of advertising
Jabari: we are discussing; sometimes I use Brave, sometimes Chrome
… there are times when I am interested in targeted ads
… there is a benefit or a win/win
… proposition in online advertising for everyone involved
… people just have a problem with how adv has been done
… users will see a benefit
… and publishers could provide incentives to users who provide profile data
… the death of third-party cookies provides an opportunity
… where things from past now makes sense
… up to industry to figure out how to incentivize this behaviro
… did not want to include in the proposal
… industry has to figure out new tools and the economics of using those tools
… to first part of your statement
… as far as it being similar to FLEDGE
… IGs can be used to store attributes, but this is more transparent
… not based on adtech deciding what an IG should be an dropping someone into it
… here a user can control what IG is wanted
… other aspects of ID are similar
Wendy: I heard in your presentation some use cases that B2B advertising
… and the user as a business participant seeking business relevant advertising
… encourage you to look at use cases document to see if those are captured there
… or if additional information could be added there
… and to help tease out where this could be relevant; or additional use cases to consider
Ben: let me say I am deeply skeptical
… that a segment would tell people what their job function was to receive more targeted ads
… it would increase their value from targeted standpoint
… if job info is not accessible, outside of privacy sandbox, how would they know
… to show and reward them with that incentive
… do you have a mechanism to query if person has configured or not?
Jabari: That is a good question
… can we reveal that a person has a set of attributes without revealing the person
… a blind spot we missed to add to the puzzle
Wendy: thanks
Brian: with respect to last point, if you have someone validating
… what users entered in way of credentials
<wbaker> Sounds like a perfect use case for DIDs and VCs
Brian: you would do that without publisher knowing what credentials were
Wendy: thanks again Jabari
… you can stop screen share and we'll move on to the next proposal
… which is also looking at user managed information
Ad Topic Hints [from Ben Savage] https://github.com/privacycg/proposals/issues/26
btsavage: (FB): today I want to talk about ad topic hints proposal
… which I wrote and shared with Privacy CG
… incorporated feedback from @ and @
… let me give host feedback
… overall intent is to say rather than say relevant ads by collecting behavioral data about them
… different way
… is to ask them; put them more in control
… they can choose to share what they want to share or no
… eliminates behavioral
… storage of two lists
… this ad is a relevant topics; happy to see ads from this topic
… and not relevant; don't want to see ads on this topic
… this could be a UI that people could manage list or clear out entirely
… how to get incentive to provide feedback on those ad?
… proposing that the browser automatically provides some kind of clear UX
… that allows people to say I like this or don't like this
… say it's relevant or not
… not sure what it would look like
… Some browser has text field that allows resizing component
… maybe automatically add HTML elements
… only there to provide comms with browser in a standard button format
… or maybe you hide behind a gesture like long click or right click
… if no obvious way to provide feedback, people won't know they can
… also propose that web sites do portability
… FB allows...Google I think
… person can say what they do/don't like can tell the browser
… and add them to list
… should be gated with consent mechanism
… you ok with adding X
… you can say yes or no
… don't want to add this
… once browser has some concept that 'Ben finds these ads relevants and these not'
… can make an ad topic hints
… similar to FLoC, some kind of signal, not user identifiable to select the relevant data
… unlike FLoC with cohorts that are stable through time
… I am proposing a vector space
… imagine a three-dimensional sphere
… every ad is mapped to some point on the sphere
… load browser to load ad topic hint
… browser picks topic for today
… and you do what you want from that
… pick ad from library closeby on sphere to that point
… if explicitly preferences from people
… adds randomness
… 25 points added
… browser picks point on sphere and is more likely to pick things I like
… some amount of variety and random exploration
… and give people the ability to control
… give myself variety with 50% new topics over time
… see more random ad
… open for questions
Brian: I have a number of different questions
… how do you go about figuring out what the ad is about?
… who defines that?
Ben: We take a ML algorithm
… looks at ad and implants a vector
… anyone can add
… at FB we have a whole bunch of ML models
… an ad represents some vector...an embedding
… be an open source model
… same model of concepts irrespective of topic
… all mapped to same vector space
Brian: happens on the browser?
Ben: yes, could happen on browser or on data center
… seeing a C++ executable
… spits out a list of numbers
… can embed in browser easily
… 'I like this' and get the embedding out
… and possibly a permalink
… to render inperpetuity
… or run over a series of ads and compute the embedding
apascoe: I have two questions
… first, I guess there is user fatigue
… if you are constantly querying the user about what they like/don't like
… if you are looking for something more implicit
… every ad can describe what you like or don't like
… what is fundamental difference from ad choices
… and second question
… I am pretty skeptical that
… deriving implicit interests from behavior is something that can go away
… where user is looking at specific products
… maybe recommend some new product they have not seen
… so what do you really mean by browsing behavior
… some stuff is really, really indicative of user behavior even if user has not provided interest or disinterest
Ben: I don't understand user fatigue
… not force user to answer question
… don't have to do anything
… but if you dislike ad, you can tell browser I dislike
… or can say you like it
… just provide opportunity to provide feedback
… like the "Like" button on FB
… not required to provide feedback
Andrew: thought you said browser would prompt, so I may have misunderstood
… so what makes it different from ad choices
Ben: nothing in common
… ad choices gives you no control; this would give you control
… imagine an ad comes in with no third party cookies
… only thing available is context
… with ad hints, you have context plus ad hints
… if that is all you have, then ad networks would choose to use that
… people providing feedback would have some impact
… unless I misunderstand what ad choices does
Andrew: I am not deeply familiar with ad choices
… under the hood
… don't show me this ad anymore...in back end
… if you train a model like on FB
… getting indications that this browser doesn't want to see this ad
… kind of implicit
… not bother to show things like this anymore
Brendan: the main thing is the opt-out mechanism
… the implementations that are clever
… creative user feedback user solicitation does happen based on ad choices
… with that icon
… don't want user to opt-out, but have user tell us why they don't like that ad experiences
… leverages opp for ad choices
Ben: still requires some concept of third party identity
… if person told you they don't like topic
… what profile to lookup?
Brendan: yes, shift from browser to client side
… can model after that UI in a useful way
Ben: other big difference is the centralization
… if providing feedback, I am doing to a specific ad network
… rather than on a web site, where there may be 12 different ad networks
… user provides control to one, but doesn't reflect choices
… by centralizing in single place, browser can select...and give more control
<jrosewell> Why are we assuming there is no identifier available?
Wendy: have closed the large queue
Olaf: looking at proposal from user perspective
… does this change interest to user interests
… for example, I could be interested in McDonalds before lunch but not other times
… or interested once every three years in cars
… can your proposal handle this?
Ben: possibly
… status quo of ads
<Brendan_IAB_eyeo> Additional item: When you're looking at classifying ads, the vector sphere is one way, using explicit ad classification like available through https://
Ben: you are interested, make purchase, and then don't want to see anymore ads
… so topic keeps following you around
… let's say you have a traumatic life experience; like seeing baby ads after loss of child; or alcohol after stopping drinking
… you can blow away settings or reset them
… or say I don't like this ad; or change preferences
… you could build an API that does that
… ads that you dislike, and change it
… you can change it any time you want
Angelina: regarding the metadata coming to help categorize the ads
… I do recommend adopting the universal ad ID
… use or adopt universal ad IDs
… @.org allows use of metadata; different rendering that can be rolled up for recording
… at another level
… allows publishers to disallow ads
… for example, on parenting site, no alcohol ads
… Going back to ad choices
… the adoption or interaction with ad choices like functionality
… seems similar
… to ad choices, allow transparency to opt-out to ad or platform
… have to understand how that UI comes across; most people don't know choices they currently have
Ben: ad choices icon is rendered by ad network, not by browser in a standard way
… not to collected stated preferences
… it's today more of a veiled fig leaf
… about 10 pixel wide...allows multiple page loads
<Brendan_IAB_eyeo> That's an unfair characterization!
<Brendan_IAB_eyeo> of adchoices
Ben: to provide system at scale in an easy way
… provide in easy way in browser
… and not make it small
… have a centralized education about how it works
… you can influence what you see; agree you will need some type of user education
DavidDabbs: thank you, Ben
… not sure if I heard an answer to Angelina's question
… does proposal reuiqre a global ad identifier
… or a bag of attributes and embedder
… based on a standard taxonomy
… it has been a tough row to how to have unique SSPs and DSPs
… nice if it's standardized metadata
Ben: maybe you can educate me about a standardized ad identifier
… like IAB...gives CSV of ads you are talking about
DD: ads they provided feedback on
… come away with sense...it knows the ad identifiers
… it will associate that; look it up
… as opposed to metadata about ads and user has said i like or don't like
… not taking about long list
… could be like you said, on sphere
Angelina: two products
… how to categorize ads, being adopted by AdID.org
… advertisers register...all kinds of information
… if 15s, 30s, 60s, each has unique ad id
… registered platform, partnering at global level
… IAB is pushing companies to use universal AdID across initiatives
Ben: That sounds really interesting
… if ads are registered in a consistent way, that would reduce the amount of storage needed
… but hate to have situation where ad tech vendor doesn't allow feedback
… important for you to provide feedback on anything you see
… embedding effort...
… build an integration between two but don't see as a requirement
Rotem, eyeo: thank you for the presentation
… I just wanted to say
… there are a lot of similarities between this proposal and work we do with product CRUMBS
… providing user with variety of topics interested in and ability to turn on/off
… what I like is the on-going feedback of users
… see if we can contribute in this sense
… question on what can be done
… when you have users...data being shared
… with site or advertiser
… documentation of interest
… of what might be done in future
<angelina> here's the wiki on Universal Ad-ID: https://
Ben: I did not understand whole question
… for logged in user
… and they are already providing feedback about which ads they like/don't, find relevant
… I like this idea that you can add this to the ecosystem
… going through user consent
… that could be an important way to kick-start the ecosystem
… if people don't provide feedback
… then it has no bearing
… this data portability idea with sharing with broader ecosystem could be way to get it started
Kris: I was one of original developers for ad choices
… I know how it works
… but I will say ad choices does not work
… you can opt out of ads, topics
… was ability to control it
… but you have to dive into it
… no one in world goes in to click and find the functionality
… when we were developing the ad choices and text
… blindness of people seeing what they can do with that
… any program like this, with users opting in
… or giving more feedback
… it needs to be more 'confrontational', in their face
… we actually did campaigns and that did not work
… Google also asks for feedback on topics
… not seeing much usage
… I think browsers have to offer up a screen to make users know they have it
… these subtle options just don't work for people
Ben: I totally agree
… Facebook, when you see an ad, you can say, 'hide the ad'
… click a drop down menu
… We do see people hiding ads, but rate is quite low
… compared to the "like" button
… you get a lot of interaction with that
… I am lazy and won't load a menu with another click
… it needs to be inline and do a single click
… Like TikTok
… videos, system seems to know what I like
… how cool it would be to do that with ads
Isaac: question in line with others
… wondered feedback on ads versus user generated content
… I am curious
Ben: entirely a function of the user experience
… you can make it higher or lower
<kris_chapman> The click through on AdChoices was originally .002%
Ben: could be clickable and small, get no clicks, bigger and...get higher clicks
Wendy: Thank you for a lively discussion on both presentations
… Ben, people should follow up on Privacy CG
… and bring back for future discussion
… please keep raising issues in Github or on list
… We will not meet next week for US Independence Day holidays, but return after that
… thank you very much
… and see you in two weeks; stay cool
<wseltzer> [adjourned]