Relational Data: Your volunteers' phones as the file

In brief

A campaign that takes your phone book keeps a matched copy of it as its own data. The names are lined up against the voter file using whatever is on the contact card, often a first name and a mobile number, and much of it lands nowhere. Academics who checked one vendor's tool found that two independent matching attempts agreed on about 30% of the same people. The volunteer consents on your behalf, and the campaign ends up owning the list. What it gains beyond the voter file is the record of who knows whom, and no platform publishes how much of that survives.

How to use this

Before you let a campaign app read your contacts, ask what it will keep: the matched list goes into the campaign's data, not yours. If you run a party or a field programme, measure your own imports before spending on texting: count how many contacts arrive and how many match, rather than repeating a vendor's description of the feature. Treat an imported list as a thin sample built from whichever volunteers downloaded the app, and expect much of it to be unusable. If you are reviewing the practice from outside, ask for import and match figures, and ask what happens to the contacts of people who never agreed to anything.

What the story is about

A volunteer connects a phone book to a campaign app. The step that follows is a join: the names and numbers in the book are lined up against the campaign's voter file. Vendors sell the feature that way. Reach, a texting platform, tells campaigns to match their contacts to their campaign's data for easy and accurate relational organising and to find them directly in the voter file (Reach, no date). Numinar describes the same step in its own app: its software syncs a volunteer's address book with the voter file, matches targets, and records each conversation so campaigns can measure results (Numinar, no date). Both pages are the companies' own marketing, so they show what is claimed rather than what works.

That join is assembled from whatever scraps a phone contact card happens to hold. In the study that documents this funnel, Outvote matched people using only the details users shared and never asked for anything missing (Schein et al., 2021). A contact saved with a first name and no surname was matched on that first name and a mobile number. To check the result, the researchers ran a second, independent matching of the same numbers and kept the entries both attempts agreed on. About 30% of subjects were matched to the same entry by both (Schein et al., 2021). That is an agreement rate between two attempts at the same job, and it is the only figure of its kind in the published record. No platform publishes its own match or import rate, so the number comes from academics checking a vendor's tool rather than from the tool itself. The paper is a peer-reviewed conference account of one programme.

What arrives on the phone is mostly the platform's text rather than the friend's. The same study analysed the wording of the messages sent and estimated that 98% of them were the default message or a minor variant of it (Schein et al., 2021). The volunteer picks who to text; the software supplies the words. A message arriving from a friend's number is therefore closer to a form letter than to a note. The voice on the other end is largely the platform's.

Consent here runs through the volunteer. Reach's privacy policy carries a section headed "Data You Provide About Others", and it describes the core features as "gathering data from and about other people and providing that data to the campaign or organization you're Reaching for" (Reach, 2026). The policy then says, "By entering it into Reach, you are giving your consent for it to be shared with your campaign or organization", and on ownership it states that under its agreement with the campaign, "they will remain the owner of this data" (Reach, 2026). The policy is the company's own document, so it shows the terms a volunteer agrees to. The person whose number was uploaded was never asked. So the campaign ends up holding a matched list of its volunteer's contacts, and along with it the record of who knows whom. No state collects that relationship, and it sits in no public voter file. That last point is an argument rather than a measured finding.

Whether a campaign reaches anyone the voter file misses is unmeasured. The one published clue points the other way. In that Outvote study, 97% of the subjects were queued by a single user, which points to a list built from one volunteer's phone book rather than a broad network (Schein et al., 2021). Contact lists gathered this way look like many small, personal address books that barely overlap. That shape matters, because the promise of relational organising is reach: a message to someone the file cannot see, carried by a person who knows them. If the books overlap little, the reach stays thin and scattered. Whether any of it goes beyond the file is the part still waiting on a measurement.

So the phone book goes in, and a campaign-owned list comes out. The app takes the volunteer's address book, matches what it can to the voter file, and writes the result into the campaign's own data. A rough match means many numbers land nowhere. A scripted message means the text travelling along a link is not really the volunteer's. What the campaign keeps, either way, is the record that one of its supporters knows a particular person. Few imports are large, and each one covers a small circle. Put together, they give a campaign a picture of who knows whom, assembled one phone book at a time.

So what

An upload is the supply step in a longer chain, and the genuinely new thing it supplies is the relationship. The voter file is the spine of campaigning, and what it cannot show is which supporter knows which neighbour. That knowledge is what the volunteer hands over. The matching then decides how much of it survives, and a low agreement rate tells you that many of the recorded links are shaky. Size alone says little. What matters is how many of those links are real, and that is the number a campaign should want before it spends anything on texting.

For political parties

Treat an imported contact list as a thin, volunteer-shaped sample, and say that inside the party before anyone builds a field programme on it. The people in it are the friends of whichever volunteers downloaded the app, so it reflects that group rather than the electorate. Own the obligations that come with holding it: the number was given by the volunteer, and the party is now the one holding data about people who never signed up for anything. Then expect the question. Regulators, journalists and your own compliance team will ask how many contacts arrived and how many matched. Nobody publishes those figures today, so a party that wants a defensible answer has to measure its own imports rather than repeat a vendor's description. The rigorous version and the responsible version come to the same thing: if the match rate is poor, you would rather learn it before you spend money texting strangers.

For government

The practice a government would be regulating has no published accuracy and no published scale, and its consent comes from the volunteer rather than from the person whose number was uploaded. That is the awkward shape of it. The relationship between volunteer and contact is the piece a voter file has never held, so a review that looks only at names and addresses will miss what the upload actually added. The practical first move is to ask for the numbers nobody publishes: how many contacts are imported, how many match, and what happens to the contacts of people who never agreed to anything. Whatever rules follow should rest on those measurements rather than on a vendor's description of a feature. The asymmetry is the substance: one person consents, and another person's name goes into the campaign's data.

Case studies

The one programme with a full published account is Outvote's work in the 2018 US midterms. Roughly 5,000 users queued 500,000 phone contacts and sent about 132,000 messages. Of the people queued, 195,118 were eligible subjects, and the refined analysis population was 27,464 (Schein et al., 2021). Queue position could not be reconstructed for 40% of subjects, often because a user had pressed a button that queued every phone contact at once. The work reads as a study of a funnel: a large intake of address books, a heavy loss at matching, and a much smaller group that could actually be analysed.

References

Numinar (no date) Relational organizing in political campaigns: what it is, why it works, and how to get started. Arlington, VA: Numinar. Available at: https://www.numinar.com/blog/relational-organizing-in-political-campaigns-what-it-is-why-it-works-and-how-to-get-started (Accessed: 10 September 2026).

Reach (2026) Privacy policy. New York: Reach Progress PBC. Available at: https://reach.vote/privacy/ (Accessed: 10 September 2026).

Reach (no date) How it works. New York: Reach Progress PBC. Available at: https://reach.vote/how-it-works/ (Accessed: 10 September 2026).

Schein, A. et al. (2021) 'Assessing the effects of friend-to-friend texting on turnout in the 2018 US midterm elections', in Proceedings of the Web Conference 2021. New York: ACM, pp. 2025–2036. Available at: https://doi.org/10.1145/3442381.3449800 (Accessed: 10 September 2026).