How we use AI, and how we don't
A clear, no-mystique note on the role data analysis plays in the watch.
We use it for triage and reading.
The AI we run does two things. First, it triages the firehose of raw signals, about 10,000 daily across the client base, into the small subset that needs a human. Second, it reads source documents in source language and surfaces the parts relevant to a given member, so the team isn't reading every regulator filing in full to find the one paragraph that matters.
We don't use it to make decisions.
Decisions about what reaches you, how it's framed, and what we recommend are human. The AI shows the team the candidates. The team picks. We don't auto-send alerts. We don't auto-write briefs that aren't reviewed.
Why this balance.
Pure-AI is too noisy and the failure modes are bad. Pure-human can't read at the volume the system needs to work. The team-of-four model only works because the AI is doing its job. We are not afraid of saying that. The point is the work, not the marketing of who did the typing.