How the Platform Turns Access into a Relationship
Overstapbuddy.nl did not begin with a conversation. It began with a supply problem, an eligibility filter, and a series of choices about who could see whom. This chapter follows the platform from recruitment to contact so that “access to a buddy” does not sound like a single event. The useful judgment is operational: a service can have a compelling mechanism and still produce little help if the right people do not enter, the match is not relevant, or the relationship has no workable way to continue.
A supply-and-eligibility funnel
The platform needed two populations before it could make a match. Buddies were recruited from successful job finders in an earlier experiment and through LinkedIn. The recruitment sought people who had recently found work in a different occupation from their former one. They were told about the purpose of the platform and the research, and information sessions offered guidance on what participation involved and how to be a good buddy. The researchers did not pay them.
That choice was not an incidental detail. The authors wanted intrinsically motivated helpers and wanted to test whether a public employment service could operate the model without a large recurring compensation bill. A person who volunteers because they want to help may bring a different tone to the relationship from someone completing a paid counseling shift. But unpaid supply is not free supply. Recruitment, information sessions, screening, safeguarding, coordination, and the effort of making suitable matches all still consume time. The study’s low-cost claim concerns implementation after the platform is set up; it is not a full cost-effectiveness analysis.
Job seekers entered through the Dutch public employment service. The service first sent an information message to people who were eligible: they were registered job seekers primarily looking in one of 31 occupations with poor job-finding prospects, had completed an online résumé, had sufficient Dutch-language skills, and had at least 90 days of unemployment-benefit rights remaining. They could express interest, but interest did not guarantee access. The researchers told them that places were limited because the number of buddies was limited.
This two-stage entry mattered for both logistics and inference. The first message identified people willing to consider the opportunity. The later random allocation meant that half of those interested were invited to create an account, while the other half were told they were not selected or placed on a waitlist. The control group therefore did not represent people who had never heard of career change. It represented interested job seekers who did not receive the platform offer during the study period.
Once the invitation existed, four states had to be distinguished:
- Eligible: the person met the employment-service and study requirements.
- Interested: the person responded to the initial invitation.
- Invited: the person was randomly selected for access.
- Connected: the person created an account, requested contact, and was accepted by a buddy.
Only the last state is a completed relationship, and not every treated person reached it. In the main sample, 53% of treated job seekers created an account and 19% found a buddy; 49 matches formed overall. Those figures are not evidence that the platform failed. A voluntary relationship may be valuable precisely because it is not forced. But they do change the interpretation of the outcome: the experiment tested an offer of access that included several chances to stop, not a fixed amount of mentoring delivered to everyone.
The funnel also explains why the control group is so important. Comparing people who formed a buddy relationship with people who did not would mix the relationship with willingness to sign up, perceived suitability of available buddies, confidence using the platform, and other unobserved differences. Randomizing the invitation avoids some of that selection problem. It does not make a matched relationship random. The cleanest estimate is the effect of being offered access among the defined population of interested job seekers.
Design choices at the point of contact
After creating an account, a job seeker saw a homepage with a buddy-search function and a video of former job seekers describing their occupational transitions. Profiles included first names, age, gender, province, former or current occupations, and a short account of labor-market experience. The purpose was not to expose a directory of every volunteer. A job seeker saw up to seven potential buddies and could send a request to up to three. A buddy could accept at most five job seekers.
These limits performed several jobs at once. A short list reduced congestion: the job seeker had a manageable choice, and the buddy was less likely to receive an unworkable number of requests. The limits also made occupational proximity potentially measurable because the available set could be compared. In principle, a researcher could ask whether seeing a closer or more distant set of occupations affected outcomes. In practice, the sample was too small for that analysis to provide a reliable answer.
The same design creates a trade-off between navigability and choice. A small menu can make action easier, but it can also hide the best possible match if the supply of buddies is thin. A cap on requests protects volunteers and reduces noise, but it gives a job seeker fewer attempts when the first choices do not respond. Matching quality is therefore a system property, not a decorative feature of the interface. It depends on who volunteers, what information profiles contain, how occupations are related, and whether people can recognize relevance from a short description.
The profile is the first interface between two people who do not yet know whether a conversation will be useful. Occupation history and a short labor-market description can make a transition legible enough to prompt a request. At the same time, the platform displayed personal details such as first name, age, gender, and province. Those details may help a job seeker recognize a possible point of connection, but they also make privacy part of matching rather than a separate technical issue. The study tells us how the profiles were used, not whether particular profile fields improved match quality. A later service would need to decide which information is necessary for relevance and which merely creates exposure.
The interface also leaves an important decision with the people involved. The platform can present a candidate and provide a request mechanism, but it cannot know from a short profile whether the buddy is a good listener, whether the job seeker is ready for a particular transition, or whether the timing is workable. The buddy’s acceptance is a second filter. The result is less like assigning a case to a professional and more like creating a bounded opportunity for mutual recognition. That can make the exchange more human, but it makes the conversion path harder to control.
A successful request led to a relationship that was intentionally flexible. Each pair had a chat window and could schedule a meeting by choosing a time and a method of communication. They could also exchange phone numbers and agree informally to talk. The platform supplied a bridge to contact; it did not prescribe the destination. That flexibility matters when the problem is personal fit. A question about an unfamiliar occupation may be answered more naturally in a conversation or phone call than in a standardized form.
It also creates a measurement boundary. For privacy reasons, the researchers did not record the contents of messages. They collected only limited activity information, such as the number of characters sent, and used surveys to ask some job seekers what they had discussed. This protects the authenticity of the exchange and reduces the feeling that a public service is monitoring a private conversation. It also means the study cannot tell us whether the decisive interaction was advice about work, a discussion of setbacks, a useful introduction, or something less directly related to occupational switching.
The percentages in this lifecycle must be read with their denominators intact. The 53% figure is the share of treated job seekers who created an account; the 19% figure is the share of the treated group who found a buddy, not the share of accounts that necessarily converted. That distinction prevents a common implementation error: treating every invitation as if it delivered the same service dose. It also clarifies why the platform can have a meaningful effect at the population level even when many invited people never reach the matched state.
The lifecycle shows why “the platform” is not one indivisible treatment. A person may receive an email and never sign up. They may sign up, inspect profiles, and decide that no available buddy fits. They may send a request that is not accepted. They may match, exchange contact details, and leave the platform for a phone call. Each transition changes what the intervention means, yet the randomized estimate combines them because it begins with the invitation.
The human layer and the scaling claim
The platform also included a professionally edited video featuring nine people who had changed occupations. This was a lower-intensity form of exposure: a job seeker could hear transition experiences without waiting for a match. It extended the reach of the role-model idea, but it cannot be treated as equivalent to a conversation. A video can make a possibility visible; a buddy can potentially respond to the details of one person’s situation. The study did not isolate the contribution of the video, profiles, matching limits, chat, or calls.
Engagement was front-loaded. Activity spiked after each experimental wave, with roughly 50 to 75 log-ins per week at the beginning, then fell close to zero after about three months. Invitations and chat messages also concentrated early. A decline in platform activity does not prove that every contact ended: pairs could exchange details and continue elsewhere. It does show that a digital service cannot assume that an initial burst of curiosity becomes a durable relationship. A scale-up would need to decide whether short, purposeful contact is sufficient or whether longer-term support is part of the promise.
The no-payment model makes this question more important. The authors argue that public employment services have access to a continuing stream of former job seekers who have successfully moved into new work, and that each might serve for a short period. That is a plausible resource model. It depends, however, on a recurring supply of people who are willing and able to help, a process for preparing them, and a way to protect job seekers when a volunteer is unavailable, unsuitable, or overconfident.
Volunteer experience can also be uneven. A person who successfully changed occupations may remember the transition accurately but still be a poor explainer. Another may give advice that worked in a different region, industry, or stage of life. The platform’s profile and matching design can make relevance easier to judge, but they cannot guarantee it. Professional guidance may be needed when the question involves benefits, training eligibility, disability, discrimination, or a complex skills assessment. A buddy layer is most defensible as a complement to those services, not as a reason to withdraw them.
The public-service setting helps the platform and narrows the claim at the same time. The employment service could identify people in poor-prospect occupations, send the invitation, and connect the study to administrative records on employment, earnings, hours, and benefits. A standalone community platform might not have that reach or outcome data. Conversely, a public agency must account for language access, digital skills, privacy, and the possibility that people who most need guidance will be least able to use an online service. The eligibility rules themselves required a résumé, Dutch-language ability, and remaining benefit rights; they were appropriate to the study but not a universal definition of who deserves support.
The funnel’s main lesson is therefore about conversion and responsibility. A promising mechanism becomes useful only when someone can enter, recognize a relevant person, make contact, and continue the exchange in a safe and workable channel. Each step belongs partly to a different actor: the employment service reaches and selects eligible people, the platform organizes discovery, and the job seeker and buddy decide whether a relationship forms. The study did not monitor the conversation, so it preserved autonomy by accepting uncertainty. The next task is to interpret what the randomized invitation changed despite that imperfect chain—and to keep separate the effects supported by the design from the effects that remain only plausible.