Most people who volunteered through Dashboard, the Obama campaign's online organizing platform, went on to volunteer through a field office, Obama for America Director of Digital Organizing Betsy Hoover said today.

Speaking with reporters at a lunch event organized by New Organizing Institute, Hoover explained — as has been previously reported but not quite put in such clear terms — that Dashboard was meant to be a place for field organizers to identify people who might be persuaded to take action offline as well as online. Her remarks come the morning after an email to supporters from Jeremy Bird, OfA's organizing director, that explained a majority of volunteers on the campaign chose to do so from a field office, while many used Dashboard or other online tools instead.

The remarks contribute to an emerging understanding of the type of "ground game" that community-organizing adherents believe is the model for campaigns in cycles to come. In this model, supporters' behavior is analyzed at every step and volunteers who take on successively more difficult tasks are groomed to become high-power campaign assets. Campaigns also try to get as close as possible to establishing a one-on-one connection with each supporter — because experiments have shown that the closer a campaign can get to sending their message to a supporter though a personal contact from a friend, the more effect that contact will have.

That logic extended to affinity groups, Hoover said in response to a question about the potential pitfalls of a campaign making assumptions about who a supporter is based on what they know about that supporter.

"Our Women for Obama group was in many cases more active than any other segment of people," she said. "When we addressed them as such, they responded to that. However, it is really tricky. How do you balance asking people to join a group without making assumptions that you can't make? We tried to walk that line in a number of ways."

At this NOI event and in earlier interviews, staffers from Obama's 2012 campaign have said that models attempting to match people with messages are best used in subtle ways. Ethan Roeder, the Obama campaign's director of data, cited as an example an anecdote from the article Charles Duhigg wrote for the New York Times about Target's data practices. To avoid creeping out someone Target suspects is pregnant, promotion of a baby carriage might be wrapped with other unrelated promotions to make it seem more innocuous. Similarly, campaigns don't have to be blunt — Yale's Eitan Hersh would say as "pandering" — in the way they match messages to people.

Most of the data campaigns are using to do this work is information that's already public, Roeder said. One could argue that because of this, models that result from using this data do not fall into some ethically gray area — while people may not realize that quite so much information about them is public record, it's still public record. The observed behaviors that went into the model are behaviors a well-informed citizen would expect to be watched or actually disclosed to the campaign, like registering to vote or living in a certain area. That's true now, Roeder said, because campaigns aren't always as sophisticated when it comes to modeling behavior as we think they are. But that might not always be the case. Take marriage licenses between same-sex couples, for example. A campaign could use that and other information to build a model to predict whether or not an individual voter is gay — but is a conversation about gay marriage the conversation that voter wants to have? And if the campaign successfully guesses the orientation of a voter who has not disclosed an orientation, but who is gay, has it engaged in a violation of privacy? As far as I know, campaigns have some time to figure out the answers to these questions because they simply haven't gotten that far yet. But that time isn't unlimited: in Washington, Washington United for Marriage did build a model to predict which voters would be likely to support same-sex marriage, and used that model to decide who to contact during get-out-the-vote operations. This is a level removed from lists of who signed petitions in support or opposition to controversial issues — something that, thanks to case law originating in the state of Washington, remains public record.

Some of the logic behind what campaigns are doing is easy to understand, like moving Dashboard volunteers up what is known in organizer parlance as the "ladder of engagement" to turn them into field volunteers. Some of it, like the algorithms that decide which message to send you based on what the campaign knows about you, is much more obtuse.

In fact, said Daniel Mintz of MoveOn.org, the algorithms that decide which affinity groups a supporter might belong to are sometimes so complex as to be "not explainable in human terms."

That's where the future of this type of data-driven organizing gets ethically and tactically murky. Besides the question of possible differences between what a model was meant to do and what it actually did, there's the potential to start a conversation with a supporter based on data a campaign thinks is complete but isn't.

"There's no conversation more uncomfortable" than one begun on that premise, Mintz said.

This post has been updated.
An earlier version of this post incorrectly referred to a Washington same-sex marriage advocacy group. Its proper name is Washington United for Marriage.