Is datafication always digital?

Quantification has a longer history, but contemporary discussions usually emphasize digital recording and the wider systems built around it.

Is it the same as digitization?

They overlap, but datafication emphasizes making phenomena analyzable through representations. Digitization can simply convert an existing artifact into digital form.

Does collecting more data make a measure more meaningful?

Not automatically. More observations improve some analyses, but cannot alone fix a poorly chosen definition or an excluded population.

AI Glossary / Data engineering and architecture / Beginner

Datafication

Definition

Datafication is the process of turning activities, relationships or phenomena into data that can be recorded, compared and analyzed. In discussions of digital society, it particularly concerns representing aspects of human life through digital measures and categories. The process can enable useful services while changing what becomes visible, valued or acted upon.

What it means

A conversation, journey or learning experience contains more than a single number. Datafication selects aspects of it, such as duration, route, clicks or completion status, and makes them available for systematic processing. The resulting records are representations of the activity, not the whole activity itself.

This differs from merely digitizing an existing artifact. Scanning a paper attendance sheet produces a digital image; extracting attendance events and calculating participation measures makes selected aspects analyzable. The terms overlap in practice, but the distinction helps explain why creating more digital files does not automatically create meaningful evidence.

How it works

Someone chooses a phenomenon, a purpose and a way to represent it. A system then captures events or observations, assigns identifiers and categories, and stores the resulting records. Analysis can summarize them or use them to train a predictive model.

Every step includes choices. A learning platform may count a page opening as participation, while a transport service may define arrival using a location threshold. Those operational definitions influence what the resulting measure means and which cases it misses.

The records can also shape later behavior. If a team is evaluated on a measured indicator, people may adapt their work to that indicator. Reviewing a datafication project therefore involves both technical quality and the consequences of turning a selected representation into a basis for decisions. More detailed recording does not remove the need to ask whether the representation fits the purpose.

Three practical examples

1. A delivery service records dispatch, arrival and completion events. It can study delays, provided those timestamps have consistent definitions and missing events are investigated.

2. A language-learning app records practice sessions and answers. These data can support feedback, but time spent in the app does not by itself establish how well someone can communicate outside it.

3. A community center records room bookings to understand demand. The team also considers people who wanted a room but could not find availability, because completed bookings alone omit unmet demand.

Think of it this way

Think of drawing a map of a neighborhood. The map makes selected features easier to compare and navigate, but its symbols omit many lived details. The analogy is limited because datafication can operate continuously and feed automated decisions that influence the people being represented. A measure may become an intervention, not merely a passive description.

Worked example

Suppose a fictional learning platform records 100 enrolled learners. Eighty open at least one lesson, while 50 submit the final exercise. If the team defines engagement as any lesson opening, its measured engagement rate is 80%.

If it instead uses final-exercise submission, the rate is 50%. Neither calculation is arithmetically wrong; they describe different events. Calling both measures learning success would hide the distinction, because an opening or submission does not directly establish understanding.

The team tests a separate comprehension activity and asks learners about obstacles. It also records that some people study downloaded material offline, so platform events may miss part of their activity. The example shows that the choice of representation precedes the percentage. An AI model trained on one indicator inherits that indicator's scope rather than automatically discovering the broader concept the team had in mind.

Illustrative case study

Fictional case: MeasureMeadow manages a shared workspace and begins counting desk reservations to estimate occupancy. The dashboard initially treats every reservation as a person physically present, even though some people cancel informally or never arrive.

The team distinguishes reservations from attendance and states the uncertainty in each measure. It chooses a proportionate way to validate aggregate occupancy and asks what information is actually needed for space planning. It avoids treating the resulting occupancy measure as a complete account of staff contribution.

Digital traces need boundaries: a useful representation for one question may poorly answer another.

Common misconceptions and mistakes

Do not assume that data are neutral simply because they are numerical. Categories, collection conditions and missing observations reflect choices. Avoid equating a proxy, such as clicks, with a broader quality, such as satisfaction.

Do not assume that all unrecorded activity is absent activity. Offline behavior and excluded participants can create important gaps. Finally, avoid treating a model's prediction as proof that its input representation is appropriate. A model can predict a recorded label well while the label remains poorly matched to the decision people want to make.

Business use

Datafication can support operational understanding, product improvement and research when the collected representation serves a clear purpose. Begin with the decision to improve and define what evidence would help, rather than collecting every available signal first.

Document measurement rules, missingness and intended uses. Consider who gains visibility, who is represented poorly and who can challenge an interpretation. For AI projects, assess the target label and collection process alongside model performance. A technically accurate pipeline can still produce an incomplete or misleading account of human activity.

Check your understanding

A platform counts lesson openings and labels the result learner understanding. What should be checked first?

Answer: Whether lesson openings are a valid representation of understanding. The record directly measures an interaction, and additional evidence is needed before using it as the broader learning outcome.

Related terms

Digitization
Data collection
Data Quality
Dataveillance
Missing data
measurement
proxy variable
digital trace
data literacy
algorithmic decision-making

Learn next

Learn measurement validity and proxy variables next. Choose one dashboard measure and trace it back to the event that creates a record. Write down what the measure captures, what it misses and how people might respond if it becomes a target. Compare it with another source of evidence about the same activity.

Sources

Ulises A. Mejias and Nick Couldry, Datafication, offers a primary scholarly account of quantification and its social consequences: https://policyreview.info/concepts/datafication

Jose van Dijck, Datafication, dataism and dataveillance, provides a related critical analysis; the publication abstract was consulted: https://ojs.library.queensu.ca/index.php/surveillance-and-society/article/view/datafication

The first article was read for its conceptual framing. The calculations, measurement examples and fictional case are original illustrations and do not attribute particular outcomes to real organizations.

Previous: Database virtualization Next: Datalog