Numerical Reasoning
18 questions · 20 minutes (~66 seconds per question)
What it tests. Quickly identifying, isolating and manipulating numerical data to make business decisions. Mental maths agility, chart interpretation, currency translation, multi-stage ratio calculations.
Worked example. A regional performance table shows EMEA revenue of 620M euros and operating costs of 440M euros, with the exchange rate 1 euro = 0.85 GBP. The question asks for total operating profit in pounds: 620 minus 440 equals 180M euros, multiplied by 0.85 equals 153M pounds.
Common traps. Distraction trap: tables present 5-6 columns of detail but the question stem only needs two data points. Unit scale traps: missing "in thousands" or "in millions" labels causes orders-of-magnitude errors in multiple choice.
How to handle it. Read the question stem before looking at the table or chart. Identify the data points you need so you can ignore the irrelevant background. Set up a structured scratchpad with formulas written cleanly so you can verify steps if the final answer does not match the multiple-choice options.
Verbal Reasoning
30 questions · 20 minutes (~40 seconds per question)
What it tests. Evaluating written arguments and drawing conclusions from text data. Distinguishing verified fact, logical inference, and unproven assumption against a passage.
Worked example. A passage discusses automated trade execution algorithms reducing block sizes and mid-tier firms scaling back operations. A statement claims "automated trading algorithms have reduced the average block size of sovereign debt transactions". The passage states this directly, so the answer is True. A second statement claims the firms scaled back due to "lack of development capital", which is not in the passage even though it may be true in reality, so the answer is Cannot Say.
Common traps. The single biggest failure mode is bringing outside knowledge in. Smart candidates who read widely often evaluate statements against real-world facts rather than the strict text. Missing qualifying words (always, never, sometimes, specifically) leads to mis-scoping logical statements.
How to handle it. Strict text-only evaluation. If a statement seems highly likely in the real world but the passage does not explicitly say it, select Cannot Say. Underline qualifying language; a statement using "always" is much harder to prove True than one using "sometimes".
Inductive / Logical Reasoning
18 questions · 24 minutes (~80 seconds per question)
What it tests. Abstract logical thinking and non-verbal pattern recognition. Identifying rules across geometric shapes, matrix shifts, and spatial transformations.
Worked example. A 3x3 matrix where each box combines three elements: outer frame alternates square, circle, triangle across rows; internal shading moves clockwise across corners; central lines increase by one per column. To solve the blank tile, apply each rule independently rather than trying to read the full shape at once.
Common traps. Trying to analyse the entire complex shape simultaneously causes cognitive overload. Getting stuck on a single difficult puzzle early burns through the time budget and leaves you rushing the back half.
How to handle it. Isolate one attribute at a time, just the background colour, just the line count, just the rotation. Use that single rule to eliminate incorrect options before moving to the next attribute. If you cannot identify the pattern within 60 seconds, make an educated guess, eliminate obvious wrong options, and move on.
OPQ32 Personality Questionnaire
104 question blocks · Untimed, typically 30-45 minutes
What it tests. Behavioural and work-style profile. Goldman screens for alignment with the firm's competencies: relationship management, analytical thinking, resilience and drive.
Worked example. Each block contains four behavioural statements. You select exactly one as "Most like me" and one as "Least like me". The system shuffles similar combinations throughout the test to check internal consistency.
Common traps. Trying to game the test by creating a "perfect" persona (marking every aggressive leadership statement as Most, every team-focused statement as Least) creates contradiction flags in the consistency checks and can trigger automatic rejection.
How to handle it. Evaluate options as comparisons within the block, not in isolation. Reflect actual workplace or academic behaviour. Keep the target division's skills in mind but avoid extreme patterns; balance initiative with collaboration.
HackerRank Technical (Engineering only)
2-3 coding tasks + 2 maths/logic questions · 90-120 minutes
What it tests. Software engineering fundamentals, algorithmic efficiency, optimisation, and mathematical reasoning under time pressure. Coding difficulty aligns with LeetCode Medium, focused on array manipulation, string processing and search algorithms.
Worked example. A circular distribution problem: N variables represent children in a circle indexed 1 to N. Starting from position D, items are distributed clockwise for T turns. Return the final 1-based index. A naive O(N*T) loop fails large hidden test cases; the optimised O(1) solution uses modular arithmetic: (D + T - 2) mod N + 1.
Common traps. Writing brute-force O(N squared) solutions that pass visible tests but fail hidden performance cases. Failing to account for empty inputs, negative integers or integer overflow boundaries.
How to handle it. Write a clean working solution first to secure baseline points before optimising. Analyse time and space complexity upfront, aim for O(N) or O(N log N) using HashMaps, two-pointer tracking or sliding windows. Test against zero values, single-element arrays and maximum integer limits before final submission.