Numerical reasoning tests eliminate more Big 4 applicants than almost any other online assessment stage. Not because they require advanced maths — they don't. But because candidates consistently misjudge what's being tested, skip meaningful practice, and then discover mid-test that 70 seconds per question feels a lot shorter when the data table has six columns and the question asks about Q3 variance.
Here's what you're actually walking into, broken down by firm, with a practice strategy that builds real speed rather than false confidence.
What the test actually looks like
Every Big 4 numerical reasoning test follows the same core format: you get a data set — a table, a graph, or both — and a multiple-choice question that requires you to extract and calculate. The data is always on screen. There's nothing to memorise. The difficulty is reading efficiently under a ticking clock.
Question types that come up repeatedly:
Percentage change — "Revenue fell from £4.2m to £3.7m. What was the percentage decrease?" Roughly 30% of questions across all firms follow this pattern.
Ratios and proportions — "If the ratio of managers to analysts is 1:6 and there are 84 analysts, how many managers are there?"
Currency conversion — "Convert €18,400 to USD at a rate of 1.09. What is the answer to the nearest dollar?"
Graph reading — Identifying the right data point before calculating. Usually where mistakes begin.
Multi-step calculations — Combining two or three of the above in a single question, under a single time budget.
A calculator is available on screen for all four firms. What remains is your ability to read the data correctly and choose the right operation — fast.
How the four firms compare
The format differs enough across firms that it's worth knowing exactly what you're walking into before your test window opens.
The firm-by-firm breakdown:
Deloitte — Numerical reasoning sits within the Immersive Online Assessment. Approximately 20 questions in a 10–30 minute window, putting you at roughly one minute per question. Multiple-choice, data presented as tables and graphs, with a calculator on screen. The tightest time pressure of the four. (Full Deloitte assessment guide →)
KPMG — Delivered via SHL or Cubiks platform. Around 18–25 questions in roughly 25 minutes (~70 seconds per question). Bundled with verbal and inductive reasoning in a single session. The SHL numerical format is well-documented — firm-specific practice packs exist for it. (Full KPMG assessment guide →)
EY — SHL-style assessment, similar structure to KPMG. Delivered within a 48-hour window, which removes scheduling pressure but adds the risk of leaving it too late. The benchmark for progression sits at approximately the 70th percentile. (Full EY assessment guide →)
PwC — SHL cognitive assessment with 12–18 questions in 17–25 minutes (~84 seconds per question). Marginally more time per question than Deloitte, but question complexity compensates for that. Note: this is separate from PwC's Arctic Shores gamified assessment, which tests different cognitive traits altogether. (Full PwC assessment guide →)
If you're targeting multiple firms, your numerical prep transfers across all four — the underlying maths is identical. The main variable is time pressure, and Deloitte is noticeably tighter.
Five maths skills to get sharp on
None of the following requires anything beyond GCSE-level arithmetic. But knowing a formula and applying it correctly in an unfamiliar data set with a clock running are two genuinely different things.
1. Percentage change — Formula: (New − Old) ÷ Old × 100. Comes up in roughly a third of questions across all firms. Know whether the question asks for an increase, a decrease, or the resulting value after the change. Practise until setting this up is automatic.
2. Finding a percentage of a value — Convert to a decimal first (17% = 0.17), then multiply. Faster than building fractions from scratch under time pressure.
3. Ratios — Know how to scale proportionally and how to find a part versus the whole. Ratio questions are short to read but easy to rush through incorrectly.
4. Chart and graph reading — Before touching the numbers, identify the axis labels, units, and which data series the question is actually about. Misreading the legend is where most wrong answers start. Treat data reading as a separate skill from calculating — it takes the same deliberate practice.
5. Unit awareness — The most mechanical and most avoidable source of wrong answers on the test. If a column header reads "Revenue (£000s)" and you treat the figures as full pounds, every calculation will be wrong. One habit prevents this: read every unit label before writing a digit.
How to build real test speed
Most candidates do a handful of practice questions, decide they understand the concept, and stop. That's familiarisation, not preparation. The skill you actually need — consistent speed under pressure — only comes from deliberate repetition with progressively tighter time constraints.
A practical three-week approach:
Week 1 — Build accuracy: 10 untimed questions per day. Check every answer and understand every mistake before moving on. Don't add time pressure until accuracy is consistently above 85%.
Week 2 — Add the clock: Set a timer for 75 seconds per question. The goal is to replicate test conditions, not to rush. Track which question types slow you down — those are the weak spots to address in week three.
Week 3 — Firm-specific practice: Switch to materials that mirror your target firm's exact platform. SHL, Cubiks, and Arctic Shores have distinct formats. Practising on the wrong one in the final week is genuinely wasted time.
Free resources worth using: Practice Aptitude Tests offers 20 free questions with full worked answers — enough to build an accuracy baseline. JobTestPrep has a free taster and paid firm-specific packs that closely simulate the SHL format you'll face at KPMG and EY.
What actually costs candidates a place
The numerical test doesn't fail the weakest candidates — it fails the ones who prepared for the wrong thing.
Four patterns come up consistently among candidates who don't progress past the numerical stage:
Rushing past the data. Spending ten seconds reading the question and thirty seconds calculating is backwards. Spend time reading the table first — what it's measuring, the units, the time periods, which column the question actually references. The arithmetic is usually simple once the data is correctly understood.
Getting stuck on one question. If you're past 80 seconds, flag it and move on. Most platforms allow you to return. A skipped question scores zero — but spending three minutes on a single hard question while leaving two easier ones unanswered costs you more overall.
Ignoring units. Already covered, but worth repeating because it's the most avoidable mistake in the test. Read every column header before starting a calculation. That's it.
Underestimating the benchmark. You're competing against a broadly numerate, prepared applicant pool. Scoring at the average isn't enough at firms targeting the 70th percentile. Consistency matters more than occasional brilliance.
The numerical test doesn't get you the job — failing it ends the application before you've had a chance to demonstrate anything else. Treat it accordingly.
For the bigger picture of what comes before and after, the Big 4 recruitment timeline maps every stage from first application through to offer — so you can see exactly where the online assessments sit and what to prepare for next.
