What You Lose When You Scale a Connection Programme

A pilot that works for 40 people rarely works the same way for 400. This is not a staffing problem you can solve by hiring more link workers. It is a measurement problem: most pilots never identified which part of the design was doing the work, so when funders ask for scale, programmes scale the wrong parts.

The mechanism, not the activity

A 2022 qualitative meta-synthesis of social prescribing found that participants described benefit going beyond social contact itself — what mattered was restored meaningful participation and purpose. Structured, purposeful group activity outperformed contact alone. That is a specific claim about mechanism, and it has a specific implication: if your walking group, cooking class, or peer support circle works, it is probably not working because people are in a room together. It is working because of the role structure gives each person, the recurrence, and the sense that showing up matters to someone.

Scaling almost always preserves the room and drops the structure. You can add a second cohort, a third venue, a rotating facilitator — the room replicates easily. The role each participant played, the informal leadership that emerged in the first group, the facilitator who happened to remember everyone’s name — none of that replicates by default. It has to be redesigned, deliberately, for each new instance.

This is why programmes so often report strong pilot outcomes and weak scaled outcomes without anyone being able to say why. Systematic review evidence on social prescribing and loneliness already flags this as a structural weakness in the field: a 2021 review found all nine included studies reported positive individual impacts, but the evidence base is thin on what specifically drove them and thinner still on what happens at volume. A separate systematic review the same year noted the same problem from the other direction — self-esteem and confidence gains show up consistently, but the studies are heterogeneous enough that you cannot isolate the active ingredient with any confidence.

The named failure mode: replication without the third place

Ray Oldenburg’s concept of the third place — informal, low-stakes gathering spots distinct from home and work — is useful here because it names what programmes accidentally rely on and then lose. Eric Klinenberg’s later work on social infrastructure extends the point: physical space itself has causal weight in whether contact happens and endures, not just a backdrop to it. A pilot often has an accidental third place: a particular café, a particular corner of a community centre, a bus stop where people started talking before the session began. Call this replication without the third place — the failure mode where a programme scales the curriculum and the referral pathway but not the physical and social conditions that made the original site sticky. The new site has the same worksheet and a worse room, and attendance drops by week three without anyone naming why.

You cannot always give a second site the same café. But you can audit for what the original site’s informality actually provided — a space to arrive early, a reason to stay late, low pressure to perform — and build an equivalent into the new site rather than assuming the curriculum alone carries it.

What to measure before you scale, not after

Most evaluation happens at the end of a pilot, when it is too late to isolate mechanism. Build these into the pilot itself:

1. Track attendance patterns within sessions, not just across them. Who arrives early, who stays after, who sits near whom. This tells you whether informal structure is doing work the curriculum takes credit for.
2. Ask participants what role they played, not just how connected they felt. “Did anyone rely on you for anything?” surfaces the purpose-and-participation mechanism the 2022 synthesis identified, which a satisfaction survey will not.
3. Separate the facilitator’s personal effect from the design’s effect. If outcomes collapse when the original facilitator is unavailable, you have not built a replicable programme — you have built a talented person’s side project.
4. Log the physical site’s informal features — the loitering space, the food, the seating arrangement — before you assume the next venue only needs to match capacity.

What this means in practice: before adding a second cohort or site, write down the three things you believe made the pilot work, in order of confidence. If you cannot rank them, you are not ready to scale — you are ready to replicate blindly and find out which one mattered by watching it disappear.

Evidence status

Claim	Evidence status
Purposeful, structured activity outperforms unstructured contact	Reasonably supported — consistent qualitative finding across syntheses
Physical/informal site features causally affect connection outcomes	Plausible and theoretically grounded, but evidence is largely descriptive rather than causal
Specific staffing or facilitation style drives retention	Anecdotal in the literature; not rigorously isolated in the sources reviewed here
Scaled programmes reliably underperform pilots	Inferred from heterogeneity across studies, not directly measured by any single source

The clinician-facing commentary on the National Academies’ 2020 report makes a related point about routine assessment: doing something correctly once, in a research setting, does not tell you how it performs when embedded in ordinary practice at volume. The gap between demonstration and system is exactly the gap most connection programmes fall into when they scale.

What this does not solve

None of this tells you how many sites you can run before quality collapses entirely, because the literature has not measured that threshold. It also does not solve the reach problem underneath all of this: the pilot cohort that made your programme work was very likely people who sought it out, had transport, and had enough baseline social capacity to benefit from a nudge. Scaling to reach people who did not opt in — the isolated, the housebound, the people social prescribing is nominally aimed at — is a different and harder problem than replicating a room that already worked for the willing.