What to Measure Before You Scale a Connection Programme

A programme that reduces loneliness for 40 people in one neighbourhood does not automatically reduce loneliness for 4,000 people across a region. Something in the small version was doing the work — a facilitator who knew everyone’s name, a room that held twelve chairs comfortably, a waiting list that created just enough demand to matter. Scale usually removes exactly that thing, then wonders why the outcomes went with it.

This is a measurement problem before it is an operations problem. If you cannot say which mechanism was producing the effect, you cannot tell, once you scale, whether you have kept it or replaced it with something that merely looks similar on a dashboard.

The mechanism, not the activity

Funders and boards ask for numbers on reach: sessions delivered, people enrolled, postcodes covered. None of that tells you why the programme worked. A qualitative meta-synthesis of social prescribing found that participants described benefit coming not from contact itself but from restored participation and purpose — feeling useful, having a role, being expected somewhere. Structured, purposeful activity outperformed unstructured contact. That is a mechanism claim, and it has a testable implication: if your scaled version drops the “role” element to move more people through faster, you should expect the outcome to weaken even while attendance numbers climb.

Before scaling, write down, in one sentence, what you believe is actually producing the effect. Not “peer support” — what specifically. “Being assigned a recurring task within the group that others rely on.” “Meeting the same six people weekly rather than a rotating group.” “A facilitator who follows up by phone when someone misses twice.” Then treat that sentence as the thing you protect, and everything else — venue, hours, group size — as negotiable.

Three questions to answer before you grow

What ratio made this work, and can it survive? A facilitator who knows twelve names by week two cannot know sixty. Social prescribing evidence is consistently limited by small, heterogeneous trials, which makes it hard to say what ratio is required — but the qualitative pattern points toward relationship density, not headcount, as the active ingredient. If your scaling plan increases caseload per staff member, you are running a different intervention, whatever the name on the door.
Does the outcome hold in a second setting before you replicate it in twenty? Test in one new site with the same measurement approach you used originally, not a lighter version of it. If the effect halves in the second site, that is information about the mechanism, not a rounding error to absorb before national rollout.
What is your leading indicator of drift? Waiting for six-month loneliness scores to tell you the scaled version failed is too slow and too expensive. A leading indicator — attendance regularity, staff-reported “do I know this person’s situation” checks, dropout by week four — should move first and should be checked monthly, not annually.

What this means in practice: write your mechanism sentence before you write your scaling plan. If the new sites can’t deliver that sentence, they can deliver the activity, but they will not deliver the outcome — and your evaluation should be designed to catch that difference, not paper over it with attendance counts.

The named failure mode: fidelity drift

Fidelity drift is what happens when a programme scales its footprint while quietly abandoning the parts that made it work, because those parts don’t scale cheaply. It rarely looks like failure from the inside. Sessions still run. Rooms still fill. The people running site fourteen believe they are doing what site one did, because the manual says so. What has actually changed is facilitator-to-participant ratio, or the consistency of who shows up week to week, or the presence of a role for each person versus generic attendance — and those are exactly the variables the qualitative evidence points to as doing the work.

Fidelity drift is hard to see because the metrics organisations default to — attendance, referrals, session counts — are activity measures, not mechanism measures. They will keep rising even as the mechanism degrades. This is why an evaluation built only around throughput will tell you the programme is succeeding right up until the outcome data, arriving much later, says otherwise.

Evidence status for common scaling claims

Claim	Evidence status
Purposeful, structured activity outperforms unstructured contact	Reasonably consistent in qualitative synthesis; not confirmed by controlled trials
Smaller facilitator-to-participant ratios improve outcomes	Plausible and widely assumed in practice; not directly quantified in the available evidence
Volunteering-type prosocial engagement reduces loneliness in older adults	Supported by one randomised trial (Hong Kong, older adults); single trial, so treat as promising rather than settled
Social prescribing reduces downstream service use	Reported in a minority of included studies in one systematic review; not the default finding
Physical isolation and small, undiverse networks predict loneliness	Fairly well supported by large survey data

Where the table shows “not confirmed” or “single trial,” resist the instinct to treat internal programme data as a substitute. An internal counting exercise showing attendance held steady across scaling tells you about attendance. It does not tell you about mechanism.

Build the audit before the growth, not after

A practical version of this: before signing off a scale-up, run a short fidelity audit against the mechanism sentence, at the new site, using the same instrument used originally — even if that instrument is informal. Eric Klinenberg’s account of social infrastructure is a useful frame here: the physical and organisational features of a place shape contact rates directly, so a change in venue, room layout, or group size is not a logistics decision, it is a programme design decision, and it belongs in the evaluation plan alongside outcome measures.

What this does not solve

None of this addresses whether the people who need the programme are the people finding it. Scaling debates tend to assume the referral pipeline is already reaching an appropriate population; in practice, most programmes reach people who were already connected enough to hear about them, seek a referral, and show up twice. A mechanism can be perfectly preserved across twenty sites and still only reach the fraction of the eligible population that was reachable in the first place. That is a different problem, and this note does not solve it.