I suggest HTML::Parser and a state machine.

You want three states:

• idle
• find telephone number item
• extract telephone number

Start in idle state and transition to "find telephone number item" when you get a start event for a P tag with class="title", then transition from that to "extract telephone number" when you get a text event containing "Telephone", otherwise return to idle state at the next text event. In "extract telephone number" state, store away the phone number at the first text event matching m/[[:digit:]]/ and return to idle state. If you only have one telephone number per page, you can also abort the parse at that point.

See the documentation for HTML::Parser for more details about that module and any good computer science text for more details about using finite state machines as parsers.