What Is Data Literacy? Definition, Examples and Why It Matters

Part of our topic guides on Data Literacy and AI Skills for Business.

By James Cotton, Founder, iO-Sphere

Read, question, act

Reading comes first and is easy to skip. It means knowing what a figure says before deciding what it means: the measure, the unit, the period and the base. A percentage of what, counted when, compared with which earlier figure? Many arguments in meetings start with two people reading the same line differently.

Questioning is asking what the number can carry. Every figure was collected by someone, for some purpose, and will bear some conclusions and not others. A data-literate reader asks where it came from and what it leaves out, without rebuilding it.

Acting is where the skill shows. The reading ends in something done in the reader's own role: a decision, a request for one more breakdown, or a choice to wait a month. A team leader, a buyer and a ward manager will act on different figures, but the habit is the same.

Ninety-six per cent, delivered on time

Suppose the monthly operations pack opens with one line: 96% of orders delivered on time in September, up from 93% in August. Nobody in the room built that figure. Several people will make decisions on it before the meeting ends.

Where it came from. "On time" is measured against a date. Is it the date promised at checkout, or a later date the system set after a delay? Is "delivered" the courier's scan at the door, or the customer's confirmation? Each choice gives a different September, and the pack may not say which one it used.

What it leaves out. Cancelled orders may sit outside the base, in which case a customer who gave up after a delay counts as neither late nor on time. Split orders may count once or twice. And 96% of how many? In a quiet month, a three-point rise can be a few dozen parcels.

What it can and cannot support. On its own definition, the figure can support the claim that dispatch held up in September. It cannot, by itself, show that customers are happier, that the new courier is better, or that October will look the same. Each of those needs another piece of data.

What to do next. The data-literate manager accepts the number for what it is and asks for two things: the late 4% split by region and by courier, and the count of cancelled orders. With those, the question of whether to renegotiate the courier contract can go on next month's agenda or come off it.

A manager can put questions like these back to an analyst without knowing anything about how the warehouse system built the figure.

The same habit at an HR desk

A second illustration, from a different job. An HR adviser preparing the quarterly people report sees sickness absence at 2.1% of working days, down from 3.4% a year earlier. The board will read it as good news about staff health.

The adviser starts with how the number is made. Here, absence is whatever line managers record in the HR system, so a team that stopped logging single days would show a fall with nobody any healthier. Long-term cases moved onto a separate category would drop out too. Someone who took a day of annual leave while ill never appears at all.

So the figure supports one claim, that recorded sickness absence fell, and no more. Before it goes to the board as a wellbeing result, the adviser compares one team's recorded absence with its rota for the same weeks. If the two agree, the fall can be reported with confidence. If not, the report says what was recorded and what was not.

What it is not

It is not statistics. The reading of the 96% used nothing harder than a percentage. The UK government's Digital and Data Profession Capability Framework lists "Applying statistical and analytical tools and techniques" as a skill of its own, alongside a separate one for "Communicating analysis and insight".

It is not a tool skill either. Knowing where the filter sits in a dashboard does not tell you whether cancelled orders are in the base, and no setting in the software will.

Nor is it the same as being an analyst. The analyst builds the delivery figure and needs data literacy as well. Most people who need it will never build a report, and they make most of the decisions.

The recognised definition

The recognised reference is the Open Data Institute's working definition. In a post on the ODI's blog by Dave Tarrant, first published in January 2021 and updated in November 2023, it reads: "Data literacy is the ability to think critically about data in different contexts and examine the impact of different approaches when collecting, using and sharing data and information."

A government report used a longer one. The 2021 DCMS data skills report adopted a definition from a 2016 IFLA Journal article: "a specific skill set and knowledge base, which empowers individuals to transform data into information and actionable knowledge by enabling them to access, interpret, critically assess, manage and ethically use data."

The definition on this page is narrower than the ODI's, which also reaches into how data is collected and shared. Ours is written for the person at work who receives a figure and has to do something with it.

Why it matters

A figure leaves the person who built it and keeps travelling. The analyst knew that cancelled orders were excluded and that the absence data depended on managers logging it; by the time the number reaches a slide, those conditions may have fallen away. The people who decide on it were not there when it was made.

That is why data literacy sits with the reader. An analyst can add footnotes, but cannot be in every meeting where the 96% or the 2.1% is quoted. If the person holding the figure cannot ask where it came from, nobody in the room will, and the decision rests on a number that may mean something other than it appears to.

In the 2021 DCMS data skills report, whose survey was conducted during the pandemic, half (51%) of the workers surveyed had not undertaken any data skills related training in the previous two years. The wider data and AI skills gap has its own guide.

Building it where people already work

Need varies by role. Everyone who receives figures needs to read them and question them. Managers who decide on them also need the confidence to ask for another breakdown and to say "not yet"; for leaders weighing data and AI training together, the same split applies one level up.

A smaller group builds the reports and needs the technical work as well, and for some of them a funded apprenticeship route fits.

The quickest material is the team's own. A delivery team learns faster on its own delivery pack than on a generic example, because it knows what a cancelled order looks like in its own system; the HR team learns on its own absence report.

In the DSIT employer survey by Ipsos (2024 fieldwork), being unsure what training was relevant was the main barrier to AI training, cited by 50% of employers. Starting from the reports a team already uses settles that question for data.

For non-technical teams, start with the reading habit.

At iO-Sphere, people learn it by doing the work on real problems with a coach who has done the job; for a whole team, we run team training in data and AI fluency.

Someone also has to own it. The government's capability framework has a skill called "Data literacy improvement": at working level it includes supporting training "by enabling data consumers to understand what they can and cannot do with data", and at expert level it includes taking "responsibility for improving the organisation's data literacy".

Frequently asked questions

What are data literacy skills?

They are the small, repeatable checks a reader makes: finding the definition behind a measure, checking the base a percentage is taken from, spotting a missing comparison, reading a chart's axes before its shape, knowing when a change is too small to mean much, and explaining a figure to a colleague in a sentence. None of them needs specialist software.

Is data literacy the same as digital literacy?

No. Skills England's June 2026 digital and technologies assessment used the core skills of the UK Standard Skills Classification and defined "digital literacy" as "using digital tools and technologies effectively (including AI)", with numeracy as a separate core skill. Data literacy was not one of those core skills. Someone can use every feature of a dashboard tool and still misread what it shows.

How do you measure data literacy?

Give someone a real report from their own job and ask what it can support, what it leaves out and what they would do next; repeat it after training and compare the answers. A quiz on terms is quicker to mark, but it checks recall, which is a different thing from the reading the job asks for.