A procedure metric can look neutral and still mislead a public decision. Average resolution time, the percentage of completed files, or the number of digital applications describes one part of a service, not the whole experience or the security of a right. For a figure to be useful, its purpose, represented group, excluded cases, and consequence of use must be known. Analysis must bring context, accessibility, equity, and human evaluation to the table before an indicator becomes a reward, penalty, or automated decision.
A figure is not the same as a guaranteed right#
Measurement is needed to manage resources, find delays, and explain results. The problem begins when a measure replaces the question it was meant to answer. A short deadline may result from good organisation, but it may also result from closing incomplete files, diverting difficult enquiries, or discouraging appeals. High digital use may mean convenience for part of the population while also indicating silent abandonment among people who need support, a screen reader, interpretation, stable connection, or time outside care work. None of those possibilities is solved by assuming that an average speaks for everyone.
Measurement is needed to manage resources, find delays, and explain results. The problem begins when a measure replaces the question it was meant to answer.
Definition must come before the dashboard. Before publishing a metric, an institution should state the right or duty connected to it, the decision it is meant to improve, how each case is counted, which dates start and finish the count, and what cannot be inferred. The journey should also be retained: application received, information requested, support offered, decision, appeal, and remedy. An administration that counts only the final box can miss the moment when a person became stuck. Aggregated data must not erase the duty to hear the individual file.
Comparison requires honest denominators. Reporting resolved files over started files is not the same as reporting them over potentially eligible people; nor is comparing offices without considering volume, complexity, language, distance, technical interruption, or regulatory change. A useful figure separates what is known from what is assumed and shows missing data, method changes, and limits. Institutional language should not present correlation as cause or operational improvement as an improvement in the right. If the indicator cannot bear that explanation, it is not ready to guide policy.
Context: what the average hides#
Context is not a footnote. A housing, dependency, migration, or grant procedure may require documents that are not equally available, journeys that do not cost the same, and deadlines that weigh differently according to urgency. Two people with the same number of waiting days can face very different consequences if one retains housing and income while the other faces eviction, medication needs, or loss of schooling. Monitoring should therefore combine time with seriousness, blockage points, sent communications, and material outcome, without turning vulnerability into a public label.
Disaggregation can reveal inequality, but it requires care. Relevant, justified, and protected categories should be used; small groups must not be exposed or treated as defective. It is better to compare results with the population the service must reach and to review interpretation with support organisations. The absence of a variable does not prove the absence of a barrier. It may mean nobody could safely ask, or that the system did not record an abandonment. Testimony and sample review complement, rather than decorate, the numerical record.
An indicator should trigger a question, not end a discussion. When a value changes sharply, the responsible response is to investigate what happened: an information campaign, a portal outage, a new instruction, a supplier change, or a classification practice. Human evaluation must be able to open files, consult staff and service users, and correct an interpretation before a figure determines budgets or priorities. Transparency is not displaying many numbers; it is making their reach and their limits understandable.
Accessibility and equity along the journey#
Accessibility cannot be measured merely by counting website visits. It is necessary to observe whether instructions are understandable, whether a form works with assistive technology, whether an equivalent telephone or in-person channel exists, whether documents can be submitted without a home scanner, and whether communications arrive in a suitable format and language. Effort must also be measured: repeated calls, journeys, money spent, informal help requested, and waiting time. A system that reduces internal time while shifting work to a person has not removed the cost; it has hidden it.
Equity requires looking at distribution, not only at the mean. If nine simple applications are resolved in one day and one complex application takes months, the average can look acceptable although the serious case remains unanswered. Medians, high percentiles, abandonments, successful appeals, corrections, and outcome differences between channels, areas, and groups should be reviewed. Such comparisons do not authorise decisions about individuals because of their category. They locate a process deserving redesign, extra support, or supervision, and then help test whether repair had an effect.
Accessible publication of results is part of democratic control. Tables need clear titles, plain-language explanations, downloadable data where appropriate, and non-digital ways to request clarification. A metric that can reidentify a person or disclose a sensitive situation should not be released. Data protection and access to information are not opposing aims: both require deciding what is published, at what aggregation level, and with what context. The institution should explain who owns the indicator and how an incorrect reading can be challenged.
Three cases that separate efficiency from justice#
Case 1: the appointment that disappears. A service reports that 92 percent of appointments are obtained in fewer than three days. The dashboard does not count people who abandon after failing digital identification, nor people who can see only distant appointments. The figure improves when the portal reduces visible options, but access worsens. Review should add failed attempts, availability by channel, distance, support, and equivalent outcomes. Listening to calls and testing with service users can reveal the barrier before the shorter deadline is celebrated.
Case 2: the rapidly closed application. A unit lowers its average time by classifying as complete files that do not provide a document within ten days. Many people did not receive a comprehensible notice or could not obtain the document in time. The productivity indicator rises while new applications and appeals increase. Evaluation should distinguish favourable resolution, reasoned refusal, informed withdrawal, and closure for no response; it should also inspect communication samples and permit recovery without penalty when the channel fails.
Case 3: the prioritising algorithm. A model orders reviews to reduce backlog. It appears successful because it processes more files, yet it learns from records in which certain neighbourhoods received more requirements because of a past practice. Using global performance as the only evidence would preserve an unequal burden. Errors and times should be assessed for relevant groups, variables and sources reviewed, human intervention documented, and an effective appeal offered. No score should prevent a person from explaining urgency, supplying context, or seeking review by someone with real authority.
Human evaluation, sources, and correction#
Human evaluation does not mean an automatic signature at the end. It requires the capacity to understand an indicator, question a source, pause a decision, and repair harm. Responsibilities, review deadlines, and resources for exceptions should be assigned. Supervisors need enough information, accessibility and bias training, and freedom to disagree with an automated recommendation. Affected people should receive an understandable explanation of the applied criterion and a simple route to request review. Learning happens when complaints change the process, not when they are filed as inconveniences.
Four institutional sources help place this work. The OECD provides people-centred digital government principles. The European Commission brings together the web accessibility framework. The United Nations publishes the Convention on the Rights of Persons with Disabilities. The Spanish Data Protection Agency provides information on data protection and accountability. These references do not replace local analysis, but they recall that public technology must be accessible, explainable, and reviewable.
A mature practice starts with a short record: purpose, included and excluded population, formula, source, period, interpretation risks, owner, and appeal mechanism. It then combines quantitative analysis with file review, accessibility testing, and conversation with people facing the procedure. Finally it publishes what was corrected and what remains uncertain. In that way, the metric ceases to be an appearance of objectivity and becomes a limited but valuable tool for securing rights. The final question is not whether the number improves, but whether a person can access, understand, be heard, and obtain a fair response.
The evaluation cycle should be regular and public. In every period, it is worth checking whether the need that justified the indicator still exists, whether rules changed, whether the channel changed, and whether the served population recognises the stated outcome. An audit should not merely reproduce a formula: it should test source data, change logs, exclusion criteria, and decisions made from the dashboard. Where automation exists, its version, date, owner, tests, and reversal rule should be retained. Where data are incomplete, uncertainty should be shown rather than dressed up as precision. Review meetings should include frontline service, accessibility, data protection, and social representation, with a record that explains agreement and disagreement. A sample of complex cases and people who abandoned should also be kept, because these provide the strongest contrast to apparent success. If review finds unequal harm, the response is not to wait for an annual report: it is to suspend harmful use, explain clearly, offer remedy, and measure whether that remedy arrived. The indicator serves the institution; the institution remains obliged to serve the person.
A further practical safeguard is needed: no team should be rewarded only for raising this metric. Shared objectives should include explanation quality, appeal resolution, tested accessibility, and reduction of unfair differences. Before changing a rule, an institution should run a limited trial with alternative channels, publish its plan, and establish who can stop it. Afterwards, affected people should be able to confirm whether the change actually made the procedure easier. That conversation is not a communications appendix; it is necessary evidence. When a number and an experience conflict, the number should not automatically win: the difference should be investigated. Only then does monitoring retain public meaning, proportionality, and the ability to correct.
