A family receives a message from school: a new app will help with maths practice, adapt activities to each pupil's level, and make sure nobody falls behind. It is a hard offer to refuse. It does not ask the school to redesign its timetable, it does not require smaller classes, and it arrives with progress charts that feel reassuringly precise. Yet the important question is not whether the app looks current, or even whether it is engaging for a week. The question is what genuinely changes in the experience of learning, and whose experience changes.
Educational apps are not one thing. Some support vocabulary practice, some sequence exercises, some record response times, and some provide conversational tutoring. Treating them as one category encourages both automatic enthusiasm and automatic dismissal. Useful evidence begins by separating a sales claim from a testable proposition: in which subject, with which learners, for how long, against which alternative, and at what cost to the people using it? Without those questions, the discussion remains marketing, even when it is conducted by a public institution.
Learning is larger than a metrics dashboard#
It is easy to confuse activity with learning. A platform may record one hundred completed exercises, several earned badges, and an unbroken streak of logins. None of that, by itself, shows that a pupil understands a fraction, can explain a method, or is willing to tackle an unfamiliar problem without prompts. Usage metrics describe what happened inside the product. Learning also becomes visible away from it, when the setting changes and there is no arrow indicating the next move.
It is easy to confuse activity with learning. A platform may record one hundred completed exercises, several earned badges, and an unbroken streak of logins.
Consider a learner who races through a set of calculations because they have learned to recognise the question format. The next day, in class, they freeze when asked to decide which calculation belongs in a word problem. The app has reinforced an answer, but perhaps not the understanding that was needed. That does not make the tool useless. It points to something more specific: its effect depends on how it fits into a teaching sequence, on the conversation that follows, and on chances to carry the learning into another situation.
That is why a sensible evaluation decides in advance what signal it is looking for. It might be better performance on a shared task, a stronger ability to revise an error, a narrower gap between learners with different starting points, or more sustained participation from those who used to withdraw. It is better to select a few measures and inspect them honestly. When an organisation tracks twenty indicators until it finds a flattering one, it is not learning from a trial; it is manufacturing a justification.
The effects that do not appear on a dashboard matter too. How much time do teachers need to set up groups, resolve incidents, and make sense of reports? Which classroom conversations have been shortened to make room for individual practice? Who ends up doing technical support at home? A system can claim efficiency because it reduces minutes of explanation while shifting the work of recovering a password, interpreting an instruction, or maintaining attention to families with less spare time.
Compare with a real alternative, not with nothing#
Many technology promises look persuasive because they are compared with a caricature: a motionless classroom, a teacher repeating the same thing to thirty people, and workbooks with no opportunity for correction. But that is not the only available scenario. An honest comparison might be peer tutoring, a better staffed library, an hour for teacher coordination, accessible materials, or simply time for pupils to receive specific feedback on what they have attempted.
Money and attention are finite. When a school buys licences, installs devices, and arranges training, it is choosing not to spend some of those resources on other needs. This does not force a choice between people and technology. It does require us to stop pretending that technology arrives free of trade-offs. An app can be valuable when it removes a repetitive task and gives a teacher more time for a difficult conversation. If it occupies the hour in which that conversation could have happened, its educational cost does not vanish because it has been labelled innovation.
The comparison also needs to include maintenance. The first term is usually the kindest: there is curiosity, vendor support, and people especially determined to make the project succeed. Then come expired accounts, new school years, broken devices, learners without home connectivity, and content that grows old. Evidence that only observes the launch confuses a demonstration with a durable practice. An educational project deserves to be judged by what can continue once the presentation has ended, not by the exceptional energy of its debut.
Personalisation needs someone able to challenge it#
The word personalisation has an almost magical pull. It suggests that every learner will receive exactly what they need. Yet an automated recommendation always starts from partial data: previous answers, speed, repeated errors, perhaps an assigned level. It does not know on its own that a family is grieving, that a child cares for a younger sibling, that an undetected reading difficulty is present, or that the question itself is poorly written.
When a platform lowers a pupil's level after several mistakes, it may be offering useful help. It may also be trapping them on an easier route at precisely the moment they need a different explanation, more time, or reassurance that mistakes do not define their ability. A slogan about artificial intelligence cannot settle that difference. It takes a person who can see the history, ask what happened, and alter the recommendation without turning every exception into an administrative battle.
This calls for a practical right to disagree. Learners should know why they have received an activity, how to say that it is not working for them, and who can correct the route. Teachers need to be able to depart from a recommendation without being penalised by the reporting system. Families deserve a clear explanation of what information is used and what cannot be inferred from it. This is not about giving every household a programming lesson. It is about stopping an opaque label from replacing professional judgement.
Good personalisation does not imitate a human relationship or aspire to replace one. It can offer practice at different paces, notice a recurring difficulty, or remind someone of a goal. Its limit appears when it decides that a person is only the pattern of their clicks. In education, a data point can open a conversation; it should not close one.
A small trial can be an act of care#
Before extending an app across a whole district, it is worth running a bounded trial that does not treat participants as silent statistics. The trial should begin with a concrete question: for example, whether a reading tool helps pupils in a particular year understand informational texts and discuss their sources with more confidence. It needs a realistic period, a varied group, preparation time, and an alternative for anyone who cannot or does not want to use it.
During that trial, qualitative voices are as necessary as numerical results. A short interview may reveal that the exercises were appropriate, but that logging in embarrassed pupils who had forgotten their credentials. A classroom observation may show that the tool supported practice while reducing exchanges among peers. A teacher may point out that the report looks clear until they must justify an exception. These are findings that a completion rate rarely supplies on its own.
It is wise to state in advance what would cause a change of course. If the tool raises overall participation but greatly widens the gap between those with home connectivity and those without, will it be paused? If its recommendations create recurring errors that staff cannot correct quickly, who is responsible and by when? Putting those conditions on the table is not pessimism. It is how a purchase becomes a decision that can be revisited.
A carefully run trial also protects trust. Families can tolerate uncertainty more readily when they know there is a review date, a named contact, and a genuine possibility of withdrawing a measure. The alternative is to present a novelty as inevitable and discover later that nobody knows how to leave it. Educational innovation needs ways in, but it also needs clear ways out.
Privacy is not an institutional silence#
Digital classrooms generate many traces: answers, timings, messages, errors, uploaded files, and sometimes inferences about attention or level. Asking for consent through a lengthy notice does not automatically solve the problem. A family may agree because they fear that refusal will leave their child outside the activity. A teenager may press “agree” without understanding that some records will persist long after the exercise has ended.
The ethical question is not only whether a datum can be collected, but whether it is needed for a particular educational purpose. If a spelling-practice platform needs to retain every keystroke for years, it carries a high burden of explanation. If a report attributes low motivation to a usage pattern, the need for human review is greater still. Information should be proportionate, time-limited, and protected from uses unrelated to learning.
Useful transparency resembles a clear conversation. It should explain what the tool does, which data it uses, which decisions it does not make alone, who can access the records, and how an error can be corrected. It should also be available when it matters, rather than hidden behind a link at the end of a document. Saying “we comply with the rules” does not help a family understand whether a photograph, a voice recording, or a learning difficulty will travel beyond the school.
A test for tomorrow morning#
We do not need to wait for a major international report to make a better everyday decision. A school team can meet and ask: what problem have we observed precisely? What would we do if this app did not exist? What evidence would persuade us that it helps? Who might it harm or exclude? Who can stop it? The answers do not need to be perfect. They need to be visible and open to revision.
The hardest test comes the next morning, when a pupil arrives tired, a connection fails, and a teacher must choose between following a recommendation and hearing an explanation. A valuable educational tool does not demand that we pretend these conditions are noise. It helps us notice them earlier and respond to them better. Its promise deserves trust when it improves the capacity to teach, learn, and correct, not when it turns the complexity of a classroom into a collection of comforting numbers.
The next step can be modest: choose one digital practice already in use and write, on a single page, its purpose, its alternative, its risks, and the date on which it will be reviewed. Then ask two pupils and two families to point out what that page fails to explain. That small discomfort offers more honest evidence than any slogan. It does not guarantee that an app will work, but it returns the decision to the people who will have to live with its consequences.
