We systematically overestimate how bad social risks will feel. Commuters told to talk to a stranger predicted a worse ride, then reported the happiest one instead. The dread is real, but it is a faulty forecast, not a verdict on how it will actually go.
The fear that stops you is a prediction: this will be awkward, they will not want to talk, I will regret it. When researchers actually tested those predictions, they were wrong. People enjoyed the conversations they were sure they would hate.
You cannot argue yourself out of a bad forecast, but you can collect counter-evidence. Each rep where the dreaded conversation goes fine quietly corrects the prediction, until the dread loses its grip and starting feels less like a gamble.