Most descriptions of AI in professional services are quantitative. A percentage of the work is automated, and the number is expected to rise. That framing is useless for deciding anything, because it says nothing about which work.
The rule here is a boundary. AI automates production. It does not automate decision-making, or execution that requires judgment. Everything follows from where that line falls.
What sits on the production side
Research, content execution, lead qualification, appointment booking and CRM routing. What these have in common is that each finishes at a defined completion state. A record is routed or it is not. An appointment exists or it does not. "Ready" means something specific that can be checked by someone who was not there.
Work of that shape is where a machine genuinely substitutes for headcount, at volume, without anyone having to trust it about anything.
What sits on the other side
Positioning. Offer design. The conversation where a client decides what the engagement should actually become. These do not end at a completion state. They end when a person with enough context decides they have ended, and that decision is the thing being bought.
The allocation has a second half that is usually left out. The entry sale is deliberately not given to the most capable people, because it does not require them. Once it closes, those people take over, and expanding the work into whatever solves the real problem is where scarce judgment gets spent. A firm that inverts this puts its best people on qualification calls and its juniors on the decisions that matter.
Why the line is published
A client buying an engagement that includes automated components is entitled to know which parts of their account are handled by a system, before rather than after. Where a firm will not answer that plainly, the honest assumption is that the line moves depending on who is asking.