Interview question

An A/B test randomises logged-in users, but the analyst computes the standard error of revenue per session with every session treated as an independent row: 10,000 users, about 5 sessions each, a z-score of 2.58 (p = 0.01). Within a user, sessions are strongly correlated (intra-user correlation about 0.3, hypothetical). A colleague says the sample is 50,000 rows, so the result is solid. What do you say, and how do you analyse it properly?