Scaling a Connection Programme Without Losing What Made It Work

A pilot works. Twenty people, one site, a coordinator who knows every name. The funder likes the results and asks for the same thing in ten more towns. Eighteen months later, attendance is down, the coordinator has quit, and the evaluation shows nothing like the original effect. This is not a rare outcome. It is the default outcome, because the thing that gets scaled is usually the wrong thing.

Most connection programmes are scaled as a format: same session structure, same materials, same referral pathway, rolled out to more sites. But the format was never what produced the result. The result came from a mechanism — usually a specific person’s relational skill, a small enough group that people were missed if absent, or a physical space that supported unplanned repeat contact. Ray Oldenburg’s account of third places describes exactly this: it is not the coffee or the room that matters, but the low-stakes repeated contact the space makes possible. Scale the room without protecting the repeated contact, and you have replicated the furniture, not the effect.

This piece sets out a sequence for scaling that keeps the mechanism intact, and names the point at which most scaling plans quietly abandon it.

Step 1: Write down the mechanism before you write the scaling plan

Before drafting a bid or a rollout timeline, force a one-paragraph answer to: what specifically, in this pilot, produced the change? Not the programme description — the causal claim. Options tend to fall into a small set:

• A single staff member’s ability to build trust with people who don’t trust services.
• A group small enough that absence gets noticed and followed up.
• A physical space with genuine drop-in access, not bookable slots.
• A referral relationship (a GP, a housing officer) who trusted the programme enough to make a real recommendation, not a generic leaflet.

Each of these has a different scaling constraint. If it’s the staff member’s skill, you are constrained by hiring and training capacity, not budget. If it’s group size, you are constrained by a ratio, and adding participants without adding coordinators breaks it immediately. If it’s the referral relationship, you are constrained by how many individual professionals can plausibly build that trust — and that number is nowhere near the number of GP practices in a region.

Funders should ask for this paragraph as a condition of a scaling grant. A rollout plan without it is a format rollout by default.

Step 2: Cost the ratio, not the activity

Scaling proposals routinely keep the per-participant unit cost flat by adding participants faster than staff. This is the easiest place for a mechanism to die, because it looks like efficiency on a spreadsheet.

If the pilot mechanism depended on a coordinator noticing when someone stopped coming, that noticing has a ceiling — somewhere between 20 and 40 relationships, depending on complexity. Beyond that ceiling, the coordinator switches from relational work to administrative triage, and the thing that made the pilot work is gone even though attendance figures look fine for another two quarters.

The AARP Foundation’s 2018 survey of adults 45 and older is a useful anchor here: the strongest predictors of loneliness in that population were the size and diversity of a person’s social network and physical isolation — not whether a service existed nearby. A programme that keeps someone on an attendance list without anyone noticing their network hasn’t changed is optimising for the wrong variable.

Step 3: Budget for local adaptation, not local compliance

A programme built around one community centre’s culture will not transplant unchanged into a different neighbourhood, workplace, or clinical setting. Eric Klinenberg’s account of social infrastructure makes the point that the same physical resource — a library branch, a park — produces different outcomes depending on the surrounding social fabric it plugs into. Scaling plans that ban local adaptation, in the name of fidelity, often produce programmes that are procedurally identical and practically inert.

The alternative is not “let every site do whatever it wants.” It is: fix the mechanism as a non-negotiable, and leave the format as the variable. Fund each new site with an explicit adaptation budget and a short window — four to eight weeks — to adjust delivery before locking it in.

Step 4: Stage the evaluation so scaling failure is visible early

Evaluate the first two or three new sites properly before committing to the full rollout, using outcome measures close to the mechanism, not just attendance or satisfaction. Systematic reviews of social prescribing consistently find self-reported gains in confidence and self-esteem but also flag thin, heterogeneous trial evidence across programmes — which means a single pilot’s success tells you less than you’d like about transferability, and the burden of proof sits with the scaling phase, not the pilot.

If early scaled sites show attendance holding but the relational outcome (loneliness reduction, sustained contact, reduced isolation) falling away, stop and diagnose before adding more sites. This is the point at which most programmes instead accelerate, because contracts and funding cycles reward speed over diagnosis.

The toolkit problem

Call this failure mode the toolkit problem: the pilot gets written up as a manual — session plans, forms, a logo — and the manual becomes the thing that gets funded to scale. Manuals travel easily. Trust, ratio, and adaptation time do not. A programme can follow its own toolkit precisely at every new site and still fail to reproduce the original effect, because the toolkit never described the mechanism in the first place. Watch for this specifically at the point a pilot is asked to become a “model” — that’s usually where the mechanism gets quietly swapped for the format.

What this means in practice: before signing a scaling grant, ask the programme to name its mechanism in one paragraph, cost the staffing ratio that protects it, and set aside an adaptation budget for the first cohort of new sites. If the proposal can’t do the first of these, the rest of the plan is a format rollout wearing a mechanism’s evidence.

Evidence status

Claim	Evidence status
Third places produce contact through informal, repeated, low-stakes access	Long-standing conceptual account (Oldenburg), not experimentally tested
Civic and associational participation has declined structurally over decades	Well-documented descriptive trend (Putnam)
Social prescribing improves confidence and self-esteem	Reported across multiple studies, but trial evidence is thin and heterogeneous
Social prescribing reduces loneliness specifically	Positive in all nine studies in one systematic review, but small evidence base
Network size and diversity predict loneliness in older adults	Direct survey finding, not causal
Scaling by fixed staffing ratio preserves programme effect	Practical inference from programme design literature, not directly tested at scale

What this does not solve

None of this addresses the deeper reach problem: pilots and their scaled successors both tend to serve people who were already willing to walk through the door — those referred by a trusted GP, those already near a community centre, those with enough stability to show up twice. The UK’s 2018 loneliness strategy pushed measurement and social prescribing funding into the system, but funding a delivery mechanism does not by itself fix who gets referred into it. Scaling carefully protects a programme’s effectiveness for the people it already reaches. It does nothing to widen who is offered it.