Mock Interview Guide: Practising Out Loud the Right Way

Most candidates prepare for interviews by reading: question lists, model answers, solutions. Then, in the room, they discover that knowing an answer and saying it clearly, in order, under time pressure and with someone watching, are different skills. A mock interview trains the second skill. Done well, it is the most effective preparation there is; done casually, it is a pleasant chat that changes nothing. This guide explains the difference.

Make the session realistic

Copy the real format, including the time limits. MIT's career office runs employer mock interviews as 45-minute sessions: 30 minutes of interview, 10 minutes of feedback and a 5-minute break, in standard, behavioural, technical or case formats (MIT CAPD: mock interviews). That split is a good default. Thirty minutes is long enough for two technical questions or four behavioural ones; ten minutes of feedback forces the interviewer to prioritise.

Agree the round type in advance and stick to it. A SQL round should use a real schema and a shared editor, not a conversation about SQL. A case round needs a dataset or a business prompt and a clock. A behavioural round needs questions you have not rehearsed word for word.

Score against a written rubric

Real interviews are increasingly structured: the same questions for every candidate, planned follow-ups, and answers rated against benchmarks, which makes ratings more consistent and more valid (OPM: structured interviews). Give your mock interviewer the same tool. A simple rubric for data roles, scored 1 to 4 on each line:

• Clarifying the question: did the candidate ask about the goal, the data and the definitions before starting?
• Correctness: is the query, code or reasoning right, including edge cases such as NULLs, duplicates and ties?
• Thinking aloud: could the interviewer follow the approach while it was being built?
• Checking the work: row counts, a sanity check, a second method?
• Communication of results: a clear answer first, then the evidence, then the limits.

For behavioural questions, score structure separately: did the answer follow Situation, Task, Action and Result, with most of the time on the actions and a concrete result (MIT CAPD: STAR method)?

Think aloud in technical rounds

In SQL, coding and case rounds, interviewers score the process as well as the result, and they can only score what they hear. Practise narrating at three points. Before writing: "I'll assume one row per order line; I'll check that first." While writing: "This LEFT JOIN keeps customers with no orders; I'll filter in the ON clause so it stays a LEFT JOIN." After writing: "Row count went from 25,900 invoices to 25,900, so the join didn't duplicate anything." Silent solving, even when correct, scores lower than visible reasoning, and it gives the interviewer nothing to work with when you go wrong.

The topic guides give you material to practise on: scenario-based SQL interview questions for live queries, and statistics interview questions for reasoning aloud about evidence.

Record, review, repeat

Record every session, with permission, and watch it within a day. You will notice what nobody tells you: filler words, answers that start in the middle, long silences while typing, results buried after the method. Pick the single weakest moment of each session and turn it into the next practice. If it was a muddled answer to "tell me about a mistake", rewrite that story and redo just that question tomorrow. Improvement comes from repeating the weak part, not from repeating whole sessions.

Who should interview you

A peer preparing for the same role is good for volume. Someone who has interviewed for real, or works in the role, is better for realism: they ask the follow-ups that real interviewers ask. Rotate between both, and ask each interviewer to play a slightly sceptical stakeholder, because polite mock interviewers rarely push hard enough on weak reasoning. Practising alone, by recording yourself answering a question from a list and scoring the recording against the rubric, is far better than no practice, and it costs nothing.

A four-session plan

1. Session 1: behavioural. Four questions from your story list; score structure and length.
2. Session 2: SQL or coding. Two problems on a real dataset; score thinking aloud and checking.
3. Session 3: case or take-home presentation. Ten minutes presenting, twenty minutes of challenge.
4. Session 4: a full mixed interview at real length, ideally with someone you do not know well.

Mistakes that waste mock interviews

• Using questions you have already memorised.
• No clock, so answers sprawl.
• Feedback without a rubric, which drifts into general encouragement.
• Never reviewing the recording.
• Repeating whole sessions instead of drilling the weakest part.

For the questions themselves, start from the role guides for data analysts, data scientists and data engineers, and prepare stories with behavioural interview questions for data roles.

Interview preparation is part of the Data Analytics course.

Sources

• Offer mock interviews (MIT Career Advising and Professional Development) — MIT CAPD
• Structured Interviews (U.S. Office of Personnel Management) — U.S. Office of Personnel Management
• The STAR method for behavioral interviews (MIT CAPD) — MIT CAPD