[OpenID] New OpenID Customer Research Activity - Google research on federated login

[Eric Sachs]

>>
We also tested with different ways and are positive on the process
where we just ask the RP's username in the first step. (So only leave
the password box from your proposal). This could be e-mail or any
other unique identifier.

We also found that worked great in the lab setting, especially since a bunch
of banks now use that approach.  Google also uses that approach in some
unusual corner cases for our AppsForYourDomain product.  However it had a
significant negative impact on the number of users signing in on a daily
basis because it causes problems with the browser auto-fill of login details
for those user's who still have a LEGACY account and have not switched to
federated login.  Even the addition of the extra step caused some dropoff
for users without autofill enabled.
However, from an IDP perspective it had one significant advantage which was
that for user's who had enabled federated login on their account, the second
time they visited the RPs website they did not have the problem of
mistakenly entering the IDP E-mail+password into the login box of the RP.
 So as an IDP, I prefer this option.  But until RPs reduce the % of their
userbase that have legacy account without federated login, I doubt we will
see very many of them move to this model.

>>
Looking at the authentication market another development we see is the
discussion about a 4 corner scheme within the identity space.
I had not seen your Youtube video, but I have a PowerPoint presentation I
have used which is quite similar.  I spend a lot of time educating different
industries about the OAuth protocol, and one of the common topics is the
potential to use it in exactly the model you describe.  I have a general
overview of OAuth for non-technical people
at<http://sites.google.com/site/oauthgoog/oauth-practices>

http://sites.google.com/site/oauthgoog/oauth-practices

and there is a short reference to this e-commerce topic in the "Financial
Transactions" section.

I have done limited usability testing of this model, and ran into some
problems that are similar to the federated login scenario.  In particular,
the e-commerce merchant has to provide a way to support both the "legacy
payment" mechanisms as well as the "federated payment" mechanism.  That
unfortunately adds a bunch of complexity, but until a large % of banks
and/or credit-card companies support this type of technique, the merchants
will have to support both options.  The next usability problem is helping
the user to identify their "payment IDP" whether that is a bank or just a
credit-card company like AmericanExpress.  The best UI I have found so far
is the one that Quicken uses to help a user identify the name of their bank
or credit-card company.  However, I've explored some UI improvements that
related to the comments I just made in my previous mail to Allen.  The first
optimization is to have the user's federated login IDP save the name of the
user's payment provider, and send that to the RP at the time of login.  The
second optimization is to try to get the end-user to use their payment
provider as their IDP, instead of their E-mail provider.  As I have noted,
that adds additional usability challenges, but the banks may be willing to
take more time to train their users on how to use the bank as their IDP as a
way to increase the user's loyalty to their bank.

>> I
think the main goal for the OpenID foundation should be to work on
such a scheme definition.I
actually think this is a better task for the OAuth community initially, but
I agree that longer term OpenID could be used to further optimize the
process.

On Wed, Sep 24, 2008 at 4:29 AM, Kick Willemse <K.Willemse at diginotar.nl>wrote:

>  Eric,
>
>
>
> First of all I would like to complement on the work and really addressing
> this issue.
>
>
>
> I would like to share the issues we have been working on and wonder if this
> passed your discussions as well:
>
>
>
> We also tested with different ways and are positive on the process where we
> just ask the RP's username in the first step. (So only leave the password
> box from your proposal). This could be e-mail or any other unique
> identifier.
>
>
>
> In the end the RP always need to make a link to the user and his federated
> ID within his authorization DB. This link is made during the "help me login
> step (during registration)".
>
>
>
> Therefore the RP will always know what second screen needs to be shown
> (Redirect to IDP, Show legacy password/ sms screen)  By doing so the
> end-user will never type in his IDP password in the legacy password box. The
> login process can be a 2 step process.
>
>
>
> Looking at the authentication market another development we see is the
> discussion about a 4 corner scheme within the identity space.
>
>
>
> This model is taken from the banking industry but should be used in every
> network oriented service (Like payments, postal service, fax, telephone etc)
> The most important thing of this model is that besides the end-user and the
> RP. There will be an IDP and Acquirer role. You can find a link to a
> presentation on this here: http://nl.youtube.com/watch?v=Nu3W3ECxlM8
>
>
>
> I think the main goal for the OpenID foundation should be to work on such a
> scheme definition.
>
>
>
> Kick
>
>
>
>
> -------------------------------------------------------------------------------------
>
> Kick Willemse
>
> Manager Product Management & Developemt
>
> e-mail: k.willemse at diginotar.nl
>
> weblog: http://www.papierloos.nl
>
>
>
> DigiNotar B.V.
>
> Vondellaan 8
>
> 1942LJ Beverwijk
>
> telefoon: 0251-268888
>
>
>
>
>
> *Van:* general-bounces at openid.net [mailto:general-bounces at openid.net] *Namens
> *Eric Sachs
> *Verzonden:* woensdag 24 september 2008 1:59
> *Aan:* max engel
> *CC:* OpenID List
> *Onderwerp:* Re: [OpenID] New OpenID Customer Research Activity - Google
> research on federated login
>
>
>
> >> Do you have any information on how many steps a user was usually willing
> to put up with?
>
> In our research document (
> http://sites.google.com/site/oauthgoog/UXFedLogin) there is a section
> titled "Comparison Summary for Sign Up."  It compares that traditional
> approach with the one we are suggesting, but admits that our approach adds a
> third page.  We had some research where there was a fourth step, and in each
> case we got audible sighs of annoyance from the users, and specific
> complaints about the "# of steps."  So 3 pages seems to be the maximum
> before really hurting usability.  The research that Allen Tom from Yahoo
> showed last week also seemed to fit with that conclusion.
>
>
>
> However, if the site was one that required E-mail validation, then that
> theoretically removes some additional work the user has to do.  But users
> seemed to be so trained on that process, that they just accepted it as a
> requirement.  So they did not give us "points" for removing that step which
> made them willing to do more work in other places.  In general though,
> websites that require E-mail validation report that it causes an additional
> 15% dropoff, especially if their E-mail messages end up in spam folders.  So
> it may be that adding a 4th page will lose the RP fewer users then that 15%
> dropoff, and thus overall it would be worth it.  That type of statistic
> would need to be measured on a live site though, as opposed to using
> usability research.
>
>
>
> However, even if that is true, RPs would still prefer to both avoid that
> 15% dropoff AND avoid any additional dropoff from a 4th (or additional)
> pages.  So if we want to convince them to support advanced features which
> require more steps, such as login with a vanity URL, then we probably are
> going to need some live sites who can share trustworthy statistics that show
> this feature is worth it.
>
>
>
>
>
> On Tue, Sep 23, 2008 at 3:07 PM, max engel <max at 8bitkid.com> wrote:
>
> Do you have any information on how many steps a user was usually willing to
> put up with?  Also, I definitely agree on the TOS injection, that is
> feedback we are are getting from RP's as well.  It would be great to come up
> with some standard behaviors for that piece.
>
>
>
> Also, I wonder if an RP would trust e-mail more if it was coming from a
> Portable Contacts endpoint being declared in the XRDS as opposed to just
> being a data field passed by the OP...
>
>
>
> Cheers, Max
>
>
>
> On Sep 23, 2008, at 2:56 PM, Eric Sachs wrote:
>
>
>
>   >> So, in essence, this solution is geared towards RP's that mandate
> needing an e-mail address and don't trust an OP's assertion of ownership,
> thereby forcing them to only accept e-mail-based OP's.
>
>
>
> A better way to put it might be that it is geared towards RPs who measure
> the # of user actions in the signup flow as the definition of success.
>  Until we have some other good statistics to show them from live RPs, that
> seems to be the primary measure that they use to grade the different
> options.  That is why we added such a long section to our UI research which
> describes the specific # of steps in different scenarios.  And along those
> lines, that is one of the reasons many of them are interested in the idea of
> the IDP showing the terms of service of the RP just to reduce the # of steps
> further.
>
>
>
> >> I agree that many sites do want e-mail, but this is a data field that
> can be gathered after authenticating against a non-email-based OP.
>  Also, sites that don't require e-mail address validation after initial
> sign-up probably can trust an OP's claim about a user's e-mail address.
>
>
>
> We thought so as well.  But what we then learned from talking to existing
> websites that have experimented with federated login is that without the
> immediate E-mail validation by the IDP, the RP has no way to determine
> whether the user has a pre-existing legacy account.  So they need to both
> ask the user if they do (and to enter the legacy E-mail/password), as well
> as to send the background E-mail validation.  If they don't add this step, a
> lot of users end up with multiple accounts and get really confused.  If they
> do add these steps, then our "grade" goes down.
>
>
>
> I am hoping to find an RP for a mainstream website that is willing to try
> this new login box style, but to accept not only E-mail address inputs, but
> also OpenID domains/vanity-URLs.  Then hopefully they could report the % of
> accounts that use each of these 3 types, as well as the % success rate for
> the signup flow of each step.  My personal guess is that anyone who is
> technical enough to type in a vanity URL will be willing to perform the
> extra steps for legacy account matching & E-mail address validation.
>  However that is only a guess.
>
>
>
> On Tue, Sep 23, 2008 at 2:41 PM, max engel <max at 8bitkid.com> wrote:
>
> So, in essence, this solution is geared towards RP's that mandate needing
> an e-mail address and don't trust an OP's assertion of ownership, thereby
> forcing them to only accept e-mail-based OP's.  I agree that many sites do
> want e-mail, but this is a data field that can be gathered after
> authenticating against a non-email-based OP.  Also, sites that don't require
> e-mail address validation after initial sign-up probably can trust an OP's
> claim about a user's e-mail address.
>
>
>
> I do have concerns about pushing RP's to adopt this sort of federated
> model, because it creates a hierarchy of OP's.  That worries me, as does the
> potential fragmented user experiences between user's who use a URL-based OP
> and those coming from an e-mail provider...
>
>
>
> _max
>
>
>
>
>
>
>
> On Sep 23, 2008, at 1:22 PM, Eric Sachs wrote:
>
>
>
>   >> Now, with my "MySpace" hat on, I am biased towards URL-based
> identity, since our users will leverage their vanity URL as their OpenID
>
> Roughly half of Google Accounts are not gmail users, so we actually share
> that same bias :-)  The proposal in this research basically means that we
> would only be able to help half our users with federated login, and that is
> really unfortunate.
>
>
>
>
>
> >> I'd love to see RP's move towards similar design patterns like this to
> help users get acclimated to Federated Login, but do want to make sure that
> it would be extensible to non-email based OP's.
>
>
>
> There are certainly RPs who are not as worried about E-mail, and for those
> our best (but early) results have been to change this "next generation"
> login box to replace "Enter your E-mail address" with "Enter your E-mail
> address or OpenID domain."  We specifically left out the visual logo of
> OpenID and appended the term domain.  Those changes seem to reduce the % of
> regular users who were confused, but did catch the attention of highly tech
> savvy users who were aware of OpenID.  However it still requires them to
> know the domain of their OpenID IDP, and so that restricts further the % of
> savvy users who can make use of that advanced feature.  We tried a few
> options for leading the user to a drop down of IDPs, but they all had
> surprisingly high negative impact on regular users.  Another problem the
> live RP sites today who act as OpenID IDPs have found that after a user
> signs up with their site using OpenID, they then have to prompt them to find
> out if they have a legacy account on that site which should be migrated.
>  Advanced users might not mind that, but for average users every additional
> step in the account signup process causes a significant drop off.
>
>
>
> However even beyond the UI problem, the bigger problem is that most RPs are
> unwilling to give up on having an E-mail address for their customers, and
> they don't trust one OpenID IDP to assert that a user owns an E-mail address
> at a different provider.  If an RP of that type wants to support IDPs that
> are not E-mail providers, then their account signup process will require the
> user to both prove ownership of an OpenID URL, and then prove ownership of
> an E-mail address via a second flow.    If we wanted to reduce that
> usability problem, we could try to get E-mail providers to allow their users
> to publicly specify a different OpenID IDP which was their identity
> provider.  That might be enough for an RP like an online magazine to allow
> an IDP like MySpace or Google to assert the E-mail address of a school
> alumni E-mail address.  But this is a pretty advanced use case, so while we
> hope it will possible to do this some day (so that we can help the other
> half of our user base), we have been more focused in the near term on the
> simpler model around E-mail address.
>
>  Interesting,
>
>
>
>
>
>
>
>
>
> On Tue, Sep 23, 2008 at 12:48 PM, max engel <max at 8bitkid.com> wrote:
>
> My main concern is that the federated model doesn't support IDP's who use
> URL's for users.  Now, with my "MySpace" hat on, I am biased towards
> URL-based identity, since our users will leverage their vanity URL as their
> OpenID, but I imagine that blogs, etc. are all in a similar situation where
> we want to acclimate our users to thinking of themselves as URL's.
>
>
>
> While EAUT is definitely a great service for IDP's that are e-mail based,
> designing a federated login system around "Enter your eMail Address" does
> worry me.
>
>
>
> I'd love to see RP's move towards similar design patterns like this to help
> users get acclimated to Federated Login, but do want to make sure that it
> would be extensible to non-email based OP's.
>
>
>
> _max
>
>
>
> On Sep 23, 2008, at 8:55 AM, Eric Sachs wrote:
>
>
>
>   >> I do have a security concern with this approach in that most likely
> the AOL user will enter their AOL password because of the past experience.
> This causes a security leak for the user even if buy.com is not just
> throwing away the value.
>
>
>
> Yes, we did see that in user's who came back the "second time."  However
> the RP can detect that case, and warn the user of the mistake they are
> making which should also help train them in the future both on this RP, and
> others.  The IDP can also try to warn the user on the first identity
> verification step to avoid making that mistake, but that is not as a good a
> "trainable moment."  Along these same lines, we saw that by adding icons for
> IDPs to a login box, the pretty sizeable % of users immediately tried to
> enter their IDP E-mail/password directly into the login box.  Allen Tom from
> Yahoo shared some data last week that showed they saw the same thing.  I
> don't think there is a 100% perfect solution here, but the worst case is
> that RPs don't support federated login at all and end users just choose to
> use the same login/password as their E-mail provider across lots of other
> sites (and our stats indicate that most sadly do).
>
>
>
>
> >> Would it not be possible to use AJAX to check the user's entered email
> address against the buy.com data base to see if they've registered and if
> so, hide all the options and just show the user the login button? Or maybe
> replace the "Help me login" and "I have a password" options with text that
> says, "you are already a member of buy.com via your AOL identity. All you
> have to do is click the login button?"  I suppose that might scare some
> users because they would think their account doesn't have any password at
> all.
>
>
>
> This was an idea we considered and is on our list to evaluate, but we don't
> have any usability data on it yet.  Technically there were some concerns
> about how well this would interact with browser auto-fill of login box
> information.  It would be great if a live RP tried out a model like this and
> reported back the results.
>
>
>
> On Tue, Sep 23, 2008 at 8:02 AM, George Fletcher <gffletch at aol.com> wrote:
>
> Some thoughts after reading through the summary (
> http://sites.google.com/site/oauthgoog/UXFedLogin) page...
>
> Fortunately, even though they are confused, nearly all users did enter
> their E-mail address and clicked the login button.  As long as they do that,
> it does not matter whether they chose Yes or No in the UI, nor does it
> matter whether they typed a password.  Buy.com just needs to know that their
> domain is aol.com, and can then redirect them to AOL to verify their
> identity.
>
> I do have a security concern with this approach in that most likely the AOL
> user will enter their AOL password because of the past experience. This
> causes a security leak for the user even if buy.com is not just throwing
> away the value.
>
> Would it not be possible to use AJAX to check the user's entered email
> address against the buy.com data base to see if they've registered and if
> so, hide all the options and just show the user the login button? Or maybe
> replace the "Help me login" and "I have a password" options with text that
> says, "you are already a member of buy.com via your AOL identity. All you
> have to do is click the login button?"  I suppose that might scare some
> users because they would think their account doesn't have any password at
> all.
>
> Great research. It really helps to identify the problematic cases and where
> we need to focus UI efforts.
>
> Thanks,
> George
>
>
> Eric Sachs wrote:
>
> Last Week the OpenID Foundation held the first meeting of their Content
> Provider Advisory Committee to gather feedback on how to evolve the best
> practices for using OpenID so that it might be used by websites in a larger
> number of market segments. The meeting included representatives from many
> mainstream content websites including The New York Times, BBC, AARP, Time
> Inc., and NPR.  I attended from Google, and thought the team who pulled
> together the meeting did a great job arranging it.
>
> Google has been researching federated login techniques, and at the meeting
> we showed how a traditional login box might evolve (see below) to a new
> style of login box that better supports federated login.
>
> <http://sites.google.com/site/oauthgoog/UXFedLogin>
>
> We also shared a summary <
> http://sites.google.com/site/oauthgoog/UXFedLogin> of our usability
> research that explains how this helps a website add support for federated
> login for some users without hurting usability for the rest of the website's
> user base.  This research is not yet finalized, and we are still working
> with a bunch of companies to gather more feedback to tune this research.  If
> you have any feedback, feel free to get in touch with me.  However more
> generally we hope people will continue to contribute to the user experience
> discussions that are happening regarding many different use cases for
> OpenID, and not just the one covered in this research document.
>
>
>
> p.s. For Google's original blog post on this research, please refer to
> http://google-code-updates.blogspot.com/2008/09/usability-research-on-federated-login.html
>
> Eric Sachs
> Product Manager, Google Security
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> general mailing list
> general at openid.net
> http://openid.net/mailman/listinfo/general
>
>
>
>
> _______________________________________________
> general mailing list
> general at openid.net
> http://openid.net/mailman/listinfo/general
>
>
>
>
>
>
>
>
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-general/attachments/20080924/287da1fd/attachment-0002.htm>