Our voice agents open every call the same way: 'Hi, this is an automated assistant calling from Retrievant on behalf of a patient.' We did not always do this. In our earliest tests the agent introduced itself by a first name and let the receptionist assume.
We changed the policy for three reasons. First, it is the law in a growing number of places, and will be in more. Second, the people on the other end are healthcare workers handling protected information; deceiving them, even by omission, is not a relationship we want. Third, we ran the experiment.
Over four weeks we split calls between disclosed and undisclosed openings. Undisclosed calls had a 4-point higher completion rate in the first week. By week four the gap had reversed: receptionists at practices we called repeatedly had figured out the undisclosed agent and were noticeably less cooperative, while the disclosed agent had become 'the polite one that waits'. Disclosed completion ended 6 points higher.
There is a broader point. Agents that operate in the world on people's behalf will be tolerated exactly as long as they are honest about what they are. The first time a healthcare worker feels tricked by one, every agent pays for it. We would rather lose four points in week one.