SHL numerical reasoning tests whether you can pull the right figure out of a table or chart and act on it quickly, not whether you are good at maths. You are scored against a norm group of other candidates rather than on a pass mark, so speed matters as much as accuracy, and the reliable way to improve is to practise the specific question shapes under timed conditions until the arithmetic stops being the slow part.
- It is a data-interpretation test, not a maths test. The arithmetic is GCSE level; the difficulty is the clock and the distractors.
- Your result is a percentile against a norm group, so "how many did I get right" is the wrong question.
- A calculator is normally allowed. Check your own invitation, because the rules differ by employer and by test version.
- Most lost marks come from answering a question the data does not ask, not from getting the sums wrong.
What the SHL numerical reasoning test actually measures
The name is misleading. An SHL numerical test is not asking whether you can do arithmetic; it is asking whether you can find the one relevant number in a cluttered table, decide what to do with it, and move on before the clock takes the decision away from you.
Every item follows the same shape. You are shown a data set (a table of financial results, a bar chart of regional sales, a line graph over several years), then asked a question that requires one or two operations on figures drawn from it. Percentage change, ratio, share of total, and simple projection account for the overwhelming majority of questions you will see.
The arithmetic itself sits at GCSE level. The difficulty is entirely in the combination of time pressure and deliberately plausible wrong answers.
How it is scored, and why that changes your strategy
This is the part candidates most often get wrong, and it has direct consequences for how you should sit the test.
Your result is not a mark out of twenty. It is a percentile against a norm group: a comparison between your performance and that of a reference population of similar candidates, typically graduates applying for comparable roles. An employer then sets a threshold on that percentile.
Two things follow:
- Speed is scored, not just accuracy. Unanswered questions do not help you. If everyone in the norm group finishes and you answer three-quarters of the paper perfectly, you are still below average.
- Comparing raw scores with friends is meaningless. Different sittings draw different items from the bank, and different roles are scored against different norm groups.
Treat every question as costing you the same fixed slice of time. The one you refuse to abandon is not one hard question, it is the three easy questions at the end that you never reach.
Where the marks actually go
The distractors are not random. They are built from the specific errors candidates make, which means a wrong answer usually feels right.
| Trap | What it looks like | The fix |
|---|---|---|
| Unit mismatch | The table is in thousands, the answer options are in millions | Read the column header and the axis label before the question |
| Wrong base | Percentage change calculated from the new figure instead of the old | Say "change divided by the ORIGINAL" out loud before dividing |
| Right sum, wrong row | Correct method applied to the adjacent year or region | Put a finger (or cursor) on the cell you are reading |
| Over-reading | Question asks for the total, you compute the average | Re-read the final clause of the question before selecting |
| Sunk cost | Two minutes spent on one item because you have already invested one | Fixed time per question, then guess and move |
Only the first two are arithmetic problems. The rest are reading problems, and they are the ones practice fixes fastest.
A method that survives the clock
- Read the question before the data. You will read the table more efficiently when you already know what you are looking for.
- Find the units. Header row, axis label, and any footnote. This takes three seconds and prevents the single most common error.
- Estimate before you calculate. A rough answer tells you which options are impossible, and often only one survives.
- Do the arithmetic once. Trust the calculator; do not re-key to check unless the result contradicts your estimate.
- Re-read the final clause. "Which region grew fastest" and "which region grew most" are different questions with different answers in the same table.
- Commit. Select, move on, do not revisit unless you finish with time spare.
How to practise so it counts
Untimed practice builds false confidence, because the thing being tested is performance under time pressure and untimed practice removes it. Three rules:
- Always run the clock. Even when you only have ten minutes to spare, sit ten minutes of timed questions rather than thirty unpressured ones.
- Review every wrong answer against a worked solution. The score tells you nothing; the reason you picked the distractor tells you everything.
- Repeat the format, not the questions. You are training recognition of question shapes, so a bank that regenerates is worth more than one you can memorise.
If you want the whole thing in one place, CorPrep's SHL replicas run to the same timing and question structure, and every item comes with a full worked solution and a per-topic breakdown of where the marks went.