It's a little more than that; it also incorporates silence detection so that it waits for the caller to finish talking before 'Lenny' plays its next line.

But otherwise, yeah, it's just recordings.

It does seem natural that it is a recording with silence detection. What about the process of choosing these specific set of dialogue? It is definitely not chosen arbitrarily as lenny's response blend pretty well with marketer's question. I was imagining some analysis on marketer's most used conversation were done.

The lines are exactly the same for each caller, played in the same order, on a loop. The recordings are obviously tailored for telemarketer scripts, and to string them along and make them think a conversation is going, but there's no dynamic analysis going on.

If that's what you're saying, I apologise for misinterpreting!