Erlang B and Poisson (or Molina, as it is known in Europe) are similar, with one major exception.
Erlang B assumes that blocked calls leave the system immediately and do not return. Poisson assumes that blocked calls remain in the system for the length of their call duration, had they been answered.
Poisson’s assumption is usually wrong, but in practice, Molina can be used to take retries into account; that is, calls which return immediately if they are blocked (people redialling if they hear busy).
For low blocking levels, the difference is probably negligible. I prefer to use Extended Erlang B to model retries though, because it allows you to set the percentage of calls which will retry if they encounter busy.
Cheers,
Tom.