Real voter conversations, now at scale
Human phone banks finish about 1 in 4 calls they try. Every unfinished call is a lost chance to persuade.
Field operations have known the hierarchy for fifty years. A conversation persuades more than a leaflet, a leaflet persuades more than an advertisement, and an advertisement persuades more than nothing. The hierarchy has never been in dispute. What has been in dispute is whether the top of it can be reached at national scale.
It could not, for a structural reason: conversations require people, people require recruitment and training, and recruitment collapses in exactly the weeks when contact rates matter most.
Completion rate, conversational AI
Completion rate, human calling
Calls per hour, achievable capacity
Why latency decides everything
The difference between a conversation and a recorded message is not the script. It is the rhythm. Human conversational turn-taking operates in a narrow timing window, and a system that responds outside it is identified as a machine within one exchange — after which the call is over regardless of what the script says.
At 250 milliseconds, response latency sits inside the window. The conversation holds. That single engineering constraint is what separates a deployment that completes 89% of its conversations from one that completes almost none.
The multilingual problem disappears
In multilingual markets the staffing constraint compounds: a campaign needs not just callers but callers in every language its electorate actually speaks, and it needs them in proportion to where those speakers live.
A system operating across 117 languages, capable of switching language mid-conversation, removes the constraint rather than easing it. The question stops being which communities the campaign can afford to reach in their own language, and becomes which communities it has decided to.
What this does not remove
Legal obligation. Deployment remains subject to each country's tele-solicitation laws, do-not-call registers and consent requirements. Those have to be enforced by the system itself rather than by policy documents describing what the system should do.
Built in, they are invisible. Bolted on, they are the reason a programme gets suspended in week three.
