The quarterly report from a consumer office contains apparent good news: reports of digital fraud have fallen by thirty percent. At a meeting, the number is presented as proof that an advice campaign worked. Yet, after the meeting, a staff member remembers the queue of people who started a form and never returned, the calls that ended after callers heard a forty-minute wait, and the conversations in which someone said, “Never mind, it is too late anyway.” None of those exits appear on the chart. The figure may describe less fraud; it may also describe fewer open doors for telling someone about it.
Records are necessary. Without them, services cannot assign staff, detect patterns, or be held accountable. The problem begins when a single number is treated as though it were a transparent window onto what is happening. A closed case is the end of an administrative journey, not a complete measure of deception or of a community's ability to seek help. Before reaching a report, many experiences disappear through shame, exhaustion, lack of time, language barriers, or fear that sharing information will cause another harm.
Looking at those silences does not require guessing private stories or surveilling people. It requires respectful ways to learn where a service stops being accessible. The goal is not to inflate statistics. It is to prevent a cosmetic improvement from hiding a silent retreat.
A fall can have three explanations#
Suppose an organisation records fewer cases in May than in April. Attempts may have fallen because prevention worked. Attempts may be unchanged, while the help channel has become so difficult that fewer people complete it. Or cases may have moved to other services: local police, a neighbourhood group, a bank, or simply a family conversation that is never recorded.
Suppose an organisation records fewer cases in May than in April. Attempts may have fallen because prevention worked.
Those explanations require different responses. The first suggests maintaining and evaluating prevention. The second requires repairing access. The third calls for coordination to understand routes that are now scattered. Declaring success without distinguishing them is risky, because it may reward the very obstacle that makes people abandon help. For example, a form that looks shorter can reduce cases if it now asks for a document many people do not have nearby.
Context matters. A scam campaign can change language, platform, or timing. A community may have more information circulating after a public leak. An office may have reduced in-person support. None of this reads itself from the final number of reports. Data do not lie, but they may answer a question that is too narrow.
Count the journey, not only arrival#
One alternative is to observe the help journey through minimal milestones. How many people find an explanation of how to report? How many start a request? At which step do they stop? How many receive an answer within the promised time? How many return because they did not understand what to do next? These data must be collected carefully. A service does not need to store the content of a fraudulent message or track someone across the web to know that a step produces abandonment.
A useful dashboard can show aggregate trends: inquiries begun, calls answered, forms completed, appointments requested, and cases given a recovery plan. If the gap between two steps grows suddenly, it does not prove a problem on its own, but it offers a question worth investigating. Perhaps wording changed. Perhaps a button fails on mobile phones. Perhaps someone must attach too many files. Perhaps calls cluster at the precise time working people cannot answer.
The key is not to turn every abandonment into suspicion about the person who leaves. Someone may stop because a bank resolved the matter or because they decided to wait. The service should leave a dignified exit: “If you do not want to continue now, here is how to return, what information will be useful, and which urgent steps you can take.” Measuring the journey is for improving a door, not pursuing the person who does not walk through it.
Ask little and listen better#
Counts need qualitative questions as well. A two-minute survey, voluntary and separate from a case file, can reveal what the system cannot see: “Which part was hardest?”, “Could you find a safe channel?”, “What would have helped you ask for assistance sooner?” Open responses are helpful, but nobody should be required to retell the scam or identify another person. Information that already created vulnerability must not become the price of improving a service.
Conversations with community intermediaries are valuable too. Librarians, older people's groups, educators, and front-line staff can notice recurring doubts before they become reports. They must not turn into informal collectors of personal data. Their role is to point out access patterns: many people do not know the official number, the website language feels accusatory, or the timetable clashes with caring duties. That knowledge can guide small, concrete tests.
A municipal office might try three changes for one month: a plain-language guide, a no-appointment period late in the afternoon, and a call-back channel. It should then compare not only case files, but the share of people receiving a first answer and the reasons given by those who leave. If the number of cases rises, that should not automatically be treated as failure. It may mean the service has become visible to harms that previously remained alone.
Publishing uncertainty strengthens trust#
Public reports often seek clean conclusions. That is understandable: teams need to communicate and residents want clarity. But good communication can name limits without losing usefulness. Rather than stating, “Fraud has fallen,” a report can say, “Registered cases have fallen; we have also seen fewer completed forms since the platform change and are investigating that effect.” The sentence does not weaken commitment. It shows that an institution understands the difference between its records and social experience.
Transparency should include definitions. What counts as an attempt? When does an inquiry become a case? How are repeated incidents grouped? Which data are not collected to protect privacy? Without those clarifications, comparing months or neighbourhoods can produce unfair conclusions. A place with more reports may be less protected, but it may also have a more accessible support network and a population more willing to use it.
It also helps to publish which decisions will follow the signals. If inquiries that never become cases increase, will the form be reviewed? If missed calls accumulate, will opening hours expand? If a community points to language barriers, will interpretation be offered? Connecting observation to action prevents listening from becoming a ritual without consequences.
Prevent measurement becoming another barrier#
There is a real risk that the wish to know more produces more forms, more confusing consents, and more distrust. That is why every data point should pass a simple test: which concrete improvement will it enable, and for how long will it be kept? If there is no answer, it probably does not need collecting. Minimisation is not a brake on evaluation. It is the condition for people to participate without feeling that a report will demand another exposure.
Accessible design matters just as much. A survey that is only digital excludes someone sharing a phone or needing support. An automated call can exclude someone who cannot speak privately. Offering equivalent options, explaining that refusal does not affect help, and removing identifiers when no longer needed make the promise of care believable.
The fraud that vanishes before the report is written is not a mysterious category of invisible people. It is the foreseeable result of services that look only at their own final screen. This month, choose one point on a help journey and check, using minimal data and a voluntary conversation, who can use it and who is left out. If the answer reveals a silent exit, the most honest figure will not be the one celebrating the fall. It will be the one that prompts the door to be opened again.
