AI has made data outputs cheap, the real skill lies in interpretation
GenAI has democratised data work; even non-analysts can now generate summaries, forecasts and dashboards in seconds. What arrives is finished work, fluent and formatted and confident in tone, carrying no signal of how sound the thinking behind it was. Because AI produces plausible output regardless of the quality of that reasoning, organisations can act on analyses that look convincing but may be wrong.
The mechanical part of analysis, the part that once took training and time, has become close to free, while the difficulty of knowing whether an answer is any good has not moved at all.
Recent research led by Dr Grace Lordan, Founding Director of The Inclusion Initiative at LSE, suggests AI could save workers roughly one working day per week, though unevenly: those who can tell a good output from a plausible one stand to benefit most.
The saving is real enough, but it accrues to those already equipped to make that distinction, and judgement of that kind has always been a slower capability to acquire.
Where the gap has moved
The data skills gap predates GenAI: the UK Parliament estimates that there are around 178,000 unfilled data specialist roles nationally, and Robert Walters reports that over 80% of UK technology vacancies are data roles. Both figures measure the same thing: a shortage of people capable of producing analysis.
There is now a second gap, and it’s the harder one to close. Professionals across every function work with AI-generated analyses without the literacy to assess whether those analyses are sound.
Deloitte found that 60% of executives regularly use AI in decision-making, yet only 34% are reimagining how they work with it. An organisation can recruit analysts to fill the first gap. It cannot recruit its way out of the second because the people who need that judgement are already doing the jobs where decisions are made.
What that judgement consists of is worth setting out.
What data literacy looks like in the AI era
From tool proficiency to question fluency
Traditionally, data literacy meant tool skill: could someone build a spreadsheet model, create a dashboard or write a SQL query? Those capabilities still count, but as AI automates more of the mechanics, the value shifts to the questions asked of the output.
If an AI system produces a customer segmentation model, can you determine whether the assumptions are valid, identify any missing data, or assess whether the analysis addresses the business problem? None of those questions can be answered by the system that produced the model.
From consuming outputs to interrogating them
Dr James Abdey, Associate Professorial Lecturer in Statistics at LSE, calls the problem the "illusion of understanding": tutorials create familiarity, but genuine understanding comes only from working with ambiguity and messy real-world data.
That distinction explains why this capability is slow to build. Familiarity can be acquired quickly, and AI provides a great deal of it, since a fluent explanation of a method resembles understanding it.
Judgement comes from encountering data that does not behave as the tutorial described. Data literacy now means holding an output up against that experience: what data informed this, what assumptions shaped it, what would change if one variable shifted?
From technical skill to business translation
One of the clearest markers of value in the AI era is the ability to translate analysis into action. Luis Cantu, an alumnus of the LSE Data Analytics Career Accelerator, used the programme to move from business consulting into data, and now works as a data analyst at Leeds United Football Club.
During a Career Accelerator Q&A, he described the structured thinking he now applies to his work:
""This is the problem that you asked me to solve. This is how I structured my solution. These are the tools I used to achieve that.""
Rather than describing a technical process, he frames the business problem and how his solution addressed it. During a recent LSE Data Analytics Career Accelerator webinar, Course Facilitator Harry Kelleher argued that analysts will spend less time writing code and more time on strategic thinking, which is the same shift seen from the other direction.
From data specialist to data-fluent professional
Perhaps the most significant change in the AI era is who needs these translation skills. Data literacy is no longer confined to analysts and data scientists: marketing teams work with predictive customer models, finance professionals rely on AI-assisted forecasting, HR leaders use workforce analytics, and operations teams interact with optimisation systems.
Karen Munro, a Career Coach, noted during the webinar that employers increasingly seek professionals who can "partner with AI to get the answer faster". Every function, not just specialists, is now expected to operate with a degree of data fluency.
Why this is difficult to learn
AI has redistributed demand for data skills, but demand has not translated into capability. Online learning has expanded to meet it, and the format suits the part of the problem that was already solved: a tutorial can teach a method, and AI can now apply that method without help.
What a tutorial cannot supply is the experience Abdey describes. Judgement about data comes from working with data that is incomplete, badly structured or answering a question slightly different from the one asked, and from having to defend a conclusion drawn from it. That requires real datasets, feedback on the reasoning rather than the output, and the discipline of explaining the result to someone who will act on it.
This is where university-industry collaboration earns its place: employers supply the ambiguity of genuine business problems, while universities provide the analytical and ethical grounding to work through them.
The LSE Data Analytics Career Accelerator is built on that combination: a six-month online, part-time programme pairing academic learning with a six-week Employer Project based on a real business challenge.
Bridging the gap will define the next decade
GenAI has raised the baseline for data skills, and the advantage now lies with those who can judge what it produces. That judgement compounds: better questions produce better analysis, better analysis produces better decisions, and each decision tested against reality sharpens the judgement that follows.
The organisations that recognise this are investing in it now, while the capability is still scarce enough to be a differentiator.