Why Word, Excel, Python and AI skills matter at work: what the research shows

Four skill areas keep surfacing in labour-market research: word processing, spreadsheets, programming, and — most recently — working with AI. The evidence shows each matters across a far wider range of roles than commonly assumed, though to different degrees. This piece reviews what the published research actually says.

Baseline digital skills are the price of entry

Large-scale analyses of job postings consistently find that productivity software is a near-universal requirement rather than a specialist one. Burning Glass Technologies' analysis of millions of US postings found that nearly eight in ten middle-skill jobs — office administration, retail supervision, sales — require digital skills, with spreadsheet and word-processing proficiency a baseline requirement for the majority of them.^[1] Digitally intensive middle-skill jobs also grew as fast as high-skill positions in the post-recession recovery, while non-digital middle-skill jobs grew slowest of all.^[1] UK evidence tells the same story: research analysing 10.2 million online job postings distinguishes "baseline" digital skills — word processing and spreadsheets — from advanced ones, and treats the baseline set as a general employability requirement across occupations.^[2]

Requirement and supply do not match. The UK's Essential Digital Skills research found that 23% of employees have trouble using spreadsheets to interpret data, and that nearly four in ten workers lack essential digital workplace skills altogether.^[3] The gap persists even among incoming graduates: a study of 440 first-year students found that while participants rated spreadsheets the most crucial skill to develop, most classified themselves as beginners.^[4]

Spreadsheet skill is a quality issue, not just a speed issue

The case for spreadsheet competence is not only that jobs demand it, but that weak skills carry measurable costs. Field audits of operational spreadsheets in real organisations, summarised by Panko, found errors in 94% of the 88 spreadsheets examined, with an average of 5.2% of formula cells containing an error.^[5] Human-error research explains why: people make errors at a few percent of all complex cognitive actions and are only moderately good at detecting them afterwards — and spreadsheet developers are consistently overconfident about the accuracy of their own work.^[6]

The practical implication for hiring: the difference between a basic and a skilled spreadsheet user is not typing speed but structure — knowing how to build a worksheet that computes rather than hard-codes, that can be checked, and that survives someone else's edits. That is a testable skill, and one CVs do not reveal.

Programming has spread beyond programmers

Python's workplace relevance is no longer confined to software engineering. Developer surveys place it among the most-used languages worldwide, with roughly 18 million users, driven substantially by data analysis, automation and AI work rather than application development.^[7] Labour-market research increasingly frames data skills as a general career asset: the Burning Glass Institute documents skills in data collection, analysis and visualisation becoming critical to workers in jobs and sectors "not historically thought to be data intense".^[8]

For most roles the relevant standard is working proficiency — reading a script, automating a repetitive task, aggregating a dataset — not professional software development. That distinction matters for assessment: testing an analyst against a software-engineering bar produces false negatives, while testing only vocabulary produces false positives.

AI skills: the newest premium, the widest spread

The evidence on AI skills is recent but unusually consistent across independent datasets. Lightcast's analysis of over 1.3 billion job postings found that postings requiring AI skills advertised salaries 28% higher — nearly $18,000 more per year — and that by 2024, 51% of postings requiring AI skills were outside IT and computer-science occupations, with generative-AI mentions in non-tech roles up 800% since 2022.^[9] PwC's AI Jobs Barometer, drawing on close to a billion job ads, put the wage premium for AI-skilled workers at 56% in its 2025 analysis, up from 25% the year before.^[10] The Oxford Internet Institute, studying more than 10 million UK vacancies, found a 23% premium for AI skills — larger than the premium for a master's degree.^[11]

Causal evidence on productivity supports the premium. In the first major field study of generative AI at work, Brynjolfsson, Li and Raymond tracked 5,179 customer-support agents given an AI assistant: productivity rose 14% on average, and 34% for novice and lower-skilled workers, with the tool effectively spreading the practices of the best performers to everyone else.^[12] Notably, the most-demanded AI skills in postings are general ones — effective use of assistant tools — rather than machine-learning engineering,^[9] which is precisely the skill set that is hardest to read from a CV: it is defined by judgment (what to delegate, what to verify, what not to paste in) rather than credentials.

Two cautions belong in any honest reading. First, enterprise-level results are more uneven than individual-level ones — a widely cited MIT report found most enterprise generative-AI pilots delivered no measurable profit impact.^[13] Second, premiums measured in postings partly reflect scarcity; as AI fluency becomes common it may behave like spreadsheet skill — less a bonus, more a requirement. Neither caution weakens the case for assessing the skill; both strengthen it: organisations capture AI's individual-level gains only when the people using it can direct and verify it.

Different roles, different depths — same four skills

Put together, the research describes a labour market in which the same four skills recur at different intensities. Administrative and coordination roles need dependable word processing and basic spreadsheets; analytical roles need deep spreadsheets and increasingly some Python; technical roles need real code; and a growing majority of roles — 51% of AI-skill demand now sits outside tech^[9] — need the judgment to work with AI. The skills are also linked: verifying an AI's output on a data question is a spreadsheet skill and an AI skill at once.

The consistent finding across every strand of this research is a gap between what employers require and what self-report reveals — workers overrate spreadsheet accuracy,^[6] graduates rate themselves beginners at the skill they consider most crucial,^[4] and AI fluency varies sharply between otherwise similar candidates.^[12] That gap is the argument for direct assessment: measuring the skills in realistic tasks, at the depth the role actually requires.