A chatbot skin is not the same as knowing the business
Most AI sales tooling on the market has a script behind it and a hand-off point built in for the moment a real conversation starts, which is exactly the moment a script stops being useful. That shape is common because it is easy to sell and easy to demo, and it is also why most of it does not actually know the business it is supposedly representing. We wanted something that draws on what the firm genuinely knows, not a layer sitting on top of the business pretending to.
Run it through the record, not beside it
The decision was to train the agents directly on the firm's own relationship database, the real people, companies and interaction history already in the system, rather than build a separate knowledge layer for the agents to reference. That is a deliberate choice about where intelligence should live: inside the same data the team already works from every day, so a conversation the agent has is grounded in the same facts a partner would use. It is also part of a broader pattern in how the firm builds: most of what gets built here is infrastructure meant to run under someone else's process, an operator's model for a business built to support other businesses rather than to be the front-facing brand itself. The sales agents are one instance of that.
How we came at this one
The test held to from the start was whether the agent makes the call a person would actually make, given what it knows, instead of falling back to a generic script the moment the conversation gets specific. That test fit because a script that sounds fine in a demo and falls apart on the third real question is not a small bug, it is the whole failure mode this system exists to avoid.