Assessing work skills: Word, Excel, Python and AI

CVs list tools; they rarely tell you whether a candidate can use them. Here is how to assess the skill areas that now appear in almost every job description — and how our work skills suite measures each one.

Measure the work, not the vocabulary

"Proficient in Excel" is the most common and least informative line on a CV. Self-ratings correlate weakly with measured performance, and interviews reward confidence over competence. A well-constructed skills test gives you the same evidence for every candidate: performance on the tasks the job actually requires.

That is why none of the four tests in our suite is a multiple-choice quiz about menu locations. Each one puts the candidate inside a working application — a spreadsheet, a word processor, a Python editor, an AI assistant — and sets real tasks: fix this document, summarise this data, debug this script, catch the errors in this AI-written report. Scoring inspects what the candidate actually did, not just the answer they arrived at: a lookup built with a formula scores differently from the same number typed in by hand, and Python tasks check the approach in the code, not only its output.

The tests are timed, with sufficient time to complete. Efficiency with these tools is itself part of demonstrating that you know how to use them — and where a candidate needs an accommodation, the timer can be extended without affecting scoring.

The four tests

All four are built around one coherent business scenario — the quarterly figures of a fictional supply company — so a candidate who takes several tests works with the same familiar data throughout, and skills can be compared across tests.

01 Spreadsheet skills

A full working spreadsheet — formulas, lookups, pivot tables, charts, sorting, filtering and conditional formatting — with tasks like cleaning order data, building summaries and fixing broken worksheets. Scoring checks both the result and the method: a correct value typed in by hand earns less than one computed with a formula.

What a strong score means: the candidate can take raw data to a defensible summary — and won't quietly break a shared workbook.

02 Word processing skills

A working word processor where candidates edit a real business report: styles and headings, tables, find and replace, page setup, tables of contents, comments and tracked changes. Tasks are fix-this-document, not name-this-menu.

What a strong score means: the candidate can produce clean, consistent documents without supervision and collaborate through comments and tracked changes.

03 Python skills

A code editor with a live Python runtime. Candidates analyse the company's order data in real code — reading files, aggregating, writing functions, debugging a broken script. Checks run the candidate's actual code and inspect how it was written: hard-coding the answer instead of computing it is detected and scored accordingly.

What a strong score means: the candidate can automate a repetitive task or analyse a dataset in code, and understand code written by others.

04 AI skills

Realistic AI-at-work scenarios: drafting a report with an assistant, choosing and writing effective prompts, ordering a sensible workflow, fact-checking AI output against the source data, and judgment calls including data privacy. Flagging a correct statement as an error costs points — verification means precision, not suspicion of everything.

What a strong score means: the candidate treats AI as a productivity multiplier — directing it well and catching its mistakes rather than pasting output unchecked.

Two tests per skill, pitched to the role

Each skill is assessed at three levels — basic, intermediate, advanced — delivered as two separate tests. The Core test covers basic and intermediate levels in one sitting; the Advanced test is offered separately, typically to candidates who pass Core. This keeps tests pitched at the right level and no longer than they need to be: if basic or intermediate proficiency is all the role requires, that is all you test. Reserve the Advanced test for roles where the tool is central to the job, such as an analyst living in spreadsheets or a developer writing Python daily.

A level is passed at a fixed threshold across a dozen independent tasks, so a verdict never hangs on one lucky or unlucky question, and partial credit distinguishes a right answer reached the wrong way from full command of the tool.

A fair test experience

Every test opens with an optional unscored practice task so candidates learn the interface before the clock starts. During the test, candidates can skip and return to tasks, re-attempt freely within the time limit, and check their work as they go — they see whether a task passed, never running scores. Results go to the assessor only, as a per-task breakdown showing what passed, how the candidate went about it, and how long each task took.

The AI skills test deserves a special note on fairness: its in-test assistant is fully scripted, not a live model. Every candidate sees identical content, so scores are deterministic and comparable — no candidate is helped or hindered by the mood of a live AI.

Reading the results together

Few roles need all four skills at the same depth. An analyst role might weight the spreadsheet and Python tests heavily and treat word processing as a threshold check; an operations role might reverse that. Set a minimum level per skill rather than averaging scores — a brilliant Python result does not compensate for being unable to produce a readable report.

Where a candidate takes several tests, the combined report also shows where skills overlap: fact accuracy appears in both document proofing and AI verification; data handling appears in both the spreadsheet and Python tests. The AI test additionally breaks its result into four subscores — prompting, verification, judgment and workflow — because "good with AI" is really several distinct skills, and roles differ in which one matters.

Finally, keep tests job-relevant. Work skills tests belong alongside — not instead of — ability and situational judgement measures: the software test tells you what a candidate can do today; the ability test tells you how quickly they will pick up what they can't do yet.