Case notes · AI systems
An assistant that gathers the brief over the client’s sources, follows clear rules and hands off to a person.
Anonymous case note. No client names, no logos, no figures and no reserved terms — only publishable method.
An organization handled every briefing request by hand: whoever received the question had to reconstruct which sources to consult, where to find them and who to involve. The result was slow, dependent on whoever was on duty, and prone to dropping details when the context was fragmented.
The useful information already existed, but it was scattered across documents, archives and earlier conversations. Without a single point of collection, every briefing started almost from scratch and the answers were not comparable to one another.
Work out, case by case, which materials are actually needed before answering, instead of asking for everything or settling for too little.
Every answer had to respect precise limits on what can be said, to whom and in what form, without improvising.
Recognize the cases that exceed its own scope and pass them to a person, without pretending a competence it does not have.
Together with the client, we described which sources matter for each type of briefing and how to recognize them.
We translated the internal policies into rules the assistant can read, so that every answer was born within a clear scope.
We defined the signals that trigger the handover to a person, before the assistant steps outside its remit.
We tested the assistant on real cases and refined rules and sources until the behavior became predictable.
In successive phases, with client review at each step.
An assistant that, once it receives a request, gathers the information about the client's sources, organizes it and prepares a coherent briefing, following the agreed rules.
When the request leaves the scope or touches a delicate point, the assistant does not insist: it passes the case to a person with the context already gathered, so the contact does not start from scratch.
For the same request, the answer follows the same path: same sources consulted, same rules applied, same point at which to ask for human help.
A useful assistant is not the one that always answers, but the one that knows where its limits are. The handoff is not a failure: it is part of the project, designed with the same care as the rest.
The value is not in collecting everything, but in knowing what a given briefing needs. Selecting the sources is the real work, and it has to be built together with those who know the domain.
Does the assistant decide on its own what to say?
No. It works within agreed rules and, when a request exceeds them, it passes the case to a person instead of forcing an answer.
What happens if information is missing?
The assistant gathers the useful sources and, if gaps remain or the topic is delicate, it flags the case for human handoff with the context already prepared.
Can this case be traced to a specific client?
No. The description is deliberately anonymous: it tells the approach and the structure, without details that would allow anyone to identify the commissioner.