Updated April 2026 · 9 min read · Saville Assessment · Adaptive combined cognitive
| Provider | Saville Assessment |
|---|---|
| Test family | Swift Assessments (combined adaptive) |
| Format | ~20-25 minutes · adaptive · numerical + verbal + diagrammatic |
| Variants | Swift Analysis Aptitude, Swift Comprehension Aptitude, Swift Apprentice, Swift Executive |
| Used by | Aon (some pipelines), Centrica, Capita, Babcock, Network Rail, Lloyds Banking Group |
Saville's Swift Assessments are adaptive cognitive batteries that combine multiple reasoning domains into one efficient test. Where Saville's Standalone tests dedicate a full session to a single domain (numerical or verbal alone), Swift tests bundle the domains and use adaptive item delivery to triangulate ability faster.
| Variant | Combines | Used for |
|---|---|---|
| Swift Analysis Aptitude | Numerical analysis + verbal analysis + diagrammatic analysis | Senior analyst, manager, leadership roles |
| Swift Comprehension Aptitude | Numerical comprehension + verbal comprehension + error checking | Operational, administrative, customer-facing roles |
| Swift Apprentice | Numerical + verbal + checking · school-leaver level | Apprenticeship and school-leaver schemes |
| Swift Executive | Higher-order numerical, verbal, diagrammatic | Director / partner-level selection |
Swift uses an item-response-theory engine: difficulty calibrates to your responses. Initial questions are medium-difficulty; correct answers escalate difficulty, wrong answers de-escalate. Score is your demonstrated ability level, not raw item count. This means two candidates can answer the same number of questions correctly but receive very different percentile scores.
Sten scores (1-10) and percentiles relative to graduate or professional norm groups. Most graduate cutoffs sit at sten 7+ (top third), with senior roles requiring sten 8+.
Centrica, Capita, Babcock International, Network Rail, Lloyds Banking Group (selected pipelines), Severn Trent, Anglian Water, and several UK public sector recruitments use Saville Swift across graduate and operational tiers.
Practice across all three modes. Swift's combined nature means you can't rely on one strong area to compensate for a weak one. The adaptive engine evaluates each domain semi-independently.
Early questions matter more. Adaptive engines weight early items because they determine the entry difficulty for subsequent items. Take 5-10 extra seconds on questions 1-3.
Don't gamble. On adaptive tests, gambling that you'll guess right and unlock harder questions doesn't work — wrong answers de-escalate difficulty and your score drops. Answer only when reasonably confident.
TestSolve handles all three Swift cognitive modes: numerical, verbal, diagrammatic. Press F8 and the answer arrives in 4-6 seconds. The adaptive nature is invisible to TestSolve — each question still has one correct answer regardless of how the engine selected it. Current accuracy: numerical 94%, verbal 96%, diagrammatic 76%. Try free with 3 captures.
Related: Saville hub, Saville Standalone tests.
TestSolve delivers AI-powered answers to your phone in seconds. Invisible to all test platforms.
Try a free solve Buy question packagesNumerical reasoning on Swift tests is almost always table-based: two or three small tables of financial, sales, or operational data, followed by a question that requires a multi-step calculation and a unit conversion.
Q. A retail chain sells three product lines. Units sold last quarter were 660 (Line A), 1,140 (Line B) and 310 (Line C). Average selling price was £1.00, £1.00 and £1.00 respectively. Total revenue to the nearest £ was:
A) £1,780 B) £1,950 C) £2,048 D) £2,110
A. Sum the units: 660 + 1,140 + 310 = 2,110. Answer: D.
The actual Swift question adds distractors: prices in pence rather than pounds, mixed currencies, unit ambiguity (per pack vs per item). Candidates who rush the unit check pick C or B despite nailing the arithmetic.
Standard Swift Verify numerical assessments give 18 questions in 18 minutes — about 60 seconds per question. That sounds generous but each question has 3–5 numbers to read, a calculation (often multi-step), and a unit conversion.
If you're past 75 seconds and still unsure, flag and move on — you can't recover four lost minutes from one stubborn question.