CORRESPONDENCE CENTRAL

The original note, as written at the time

CORRESPONDENCE CENTRAL. How about this for trapping those people who refuse to answer emails and think they can simply reply by phone call. When the phone call comes through, a voice-to-text AI record transcriber picks up with you and, as this fellow speaks, the content is set straight into a central bucket that links email, text and phone calls. That way, the record is never lost.

Ongoing development

A way of capturing the people who refuse to answer emails and believe they can simply reply by phone call.

When the call comes through, a voice-to-text transcriber picks it up, and as the person speaks, the content is set straight into a central bucket that links email, text and calls. The record is never lost.

There are two ideas in this and they are different sizes.

The small one is a grievance: some people will not write, and the record suffers. That is real, and it is worth solving, but solving it by transcribing their calls is solving it in a way that would upset them if they knew, which usually means the solution is not as good as it looks.

The large one is that correspondence has no home. A decision is spread across an email thread, a WhatsApp group, a phone call and a meeting, and the four never meet. Six months later nobody can reconstruct what was agreed, and the reconstruction attempt is itself a week of work.

A bucket that catches all four, in one place, in time order, is the thing every organisation actually needs and almost none has. The phone call is simply the hardest channel to capture, so it is the one that exposes the gap.

The consent question has to be answered honestly. A recorded call is a recorded person, and in many jurisdictions it needs to be declared at the start. That declaration is also, conveniently, a filter: the person who objects to being recorded is usually the person whose calls are the reason for the system.

And the transcription must be attributed. A line in a record that says this is what somebody said needs to be marked as a machine’s hearing of it, not as their words. That distinction will matter in exactly one conversation, and it will matter enormously.