How to Decide When to Stop Running a Connection Programme

Most programmes don’t get killed. They get renewed by default, year after year, because no one built a decision rule for stopping and the alternative — actively choosing to end something people say they like — feels harder than just running it again. That default is expensive. It also crowds out programmes that might work better, because the budget is already spent.

This is a procedure for deciding whether to stop, redesign, or continue a connection programme, using the evidence you can actually collect rather than the trial evidence you wish existed.

Step 1: Separate three different questions you’ve been treating as one

Programme managers tend to ask “is this working?” as a single question. It isn’t. Unpack it into three:

1. Is anyone showing up? — a delivery question.
2. Do the people who show up change on any measure that matters? — an outcome question.
3. Is the cost per person changed worth what else that money could do? — an allocation question.

A programme can pass the first, fail the second, and still be worth running if it’s cheap and the third question has no better alternative. It can also pass all three individually and still be the wrong thing to keep funding, because something else would do more per pound. Confusing these three is the single biggest reason weak programmes survive: a coordinator reports good attendance, and everyone treats that as evidence of impact.

Step 2: Set the stopping threshold before you look at the data

If you decide what would count as “not working” only after you’ve seen the results, you will find a reason to keep going. Set the threshold at the start, in writing, and hold three people accountable to it: whoever runs the programme, whoever funds it, and whoever evaluates it.

A workable threshold has two parts:

• A minimum reach figure — the number or proportion of the target group you must be engaging by a set date, below which the programme is not viable regardless of outcomes.
• A minimum outcome change — a specific, prespecified improvement on a named measure, by a named point in the programme, below which you stop or redesign.

Group befriending programmes give you a real anchor point for that second figure. A 2025 randomised trial of befriending in aged care found UCLA Loneliness Scale reductions of 2.39 points at eight weeks and 2.71 points at sixteen weeks against a control group. That’s a genuine effect, from one of the few controlled trials in this literature. It is also smaller than what a structured psychological intervention achieved in a comparable trial the same year — the Lancet Healthy Longevity trial of prosocial engagement and volunteering among lonely older adults in Hong Kong. If your befriending programme isn’t producing change in that range, on that instrument, by that point, you have grounds to ask whether a more structured model would do more with the same money.

What this means in practice: write the stopping rule on the same page as the funding approval, not in the evaluation plan six months later. If the funder signs off on “continue if X, stop if Y” at the start, no one has to have the argument in real time when the numbers are disappointing.

Step 3: Use the right comparator, not “better than nothing”

Almost every social prescribing evaluation finds “positive impacts” — a 2021 systematic review of nine studies found all nine reported some positive individual effect. That’s not a high bar. Compared with no intervention, most structured social contact helps a bit. The question that should drive a stop/continue decision isn’t “did anything improve” but “did this improve enough, for enough people, relative to what a plausible alternative would achieve.”

The stronger 2022 meta-synthesis on social prescribing found something more specific and more useful: participants described benefit less in terms of “feeling less lonely” and more in terms of restored purpose and meaningful participation. Unstructured contact — a coffee morning with no shared activity — reliably scores lower than structured, purposeful group activity. If your programme is the unstructured kind and it’s plateauing, the fix may not be “stop entirely” but “add structure,” which is a redesign decision, not a closure decision.

Named failure mode: the plateau you mistake for maturity

A programme launches, attendance climbs for two or three cohorts, then flattens. The natural read is “we’ve found our steady state — this is what success looks like at scale.” Often it isn’t. It’s the same core group re-enrolling and cycling through, while new referrals trickle in at a rate that barely offsets dropout. Nobody notices because the total headcount on the register looks stable.

Check for this specifically: track the proportion of participants in any reporting period who are new versus returning. If the new-participant share is falling quarter over quarter while total numbers hold steady, you are not looking at a mature programme. You are looking at a small group of repeat users propping up a topline number that is quietly failing at reach. That is a redesign or closure signal, not a “the programme has found its level” signal.

Step 4: Check whether you’re measuring loneliness or something adjacent

A programme that fails to move loneliness scores may still be succeeding at something else worth funding — social support, confidence, service use. The AARP’s tracking of adults 45 and older found that network size and diversity, not simple frequency of contact, were the strongest predictors of loneliness, and its 2025 follow-up found loneliness among this group rising further. If your programme increases attendance but doesn’t diversify anyone’s network — same six people, same room, every week — it may never move loneliness scores even if it’s genuinely valued by participants. Don’t let a null result on the wrong outcome kill a programme that’s doing something real. Measure what the design is actually built to change before you decide it has failed.

Evidence status of the claims underneath this decision

Claim	Evidence status
Structured, purposeful activity outperforms unstructured contact for loneliness outcomes	Moderate — consistent across qualitative meta-synthesis, not yet confirmed by head-to-head trials
Befriending reduces loneliness scores measurably	Reasonably strong — one 2025 RCT with a clear effect size
Befriending is outperformed by structured psychological/volunteering interventions	Suggestive — based on comparing two separate trials, not a direct comparison
Social prescribing reduces loneliness generally	Weak — reviews report near-universal “positive impact” findings, a sign of publication and design bias more than a strong effect
Effectiveness of social prescribing specifically for older adults	Genuinely unresolved — a 2025 review protocol notes only one peer-reviewed RCT exists in the field

What this does not solve

None of this tells you what to do about the people your programme never reached in the first place. A stop/continue decision, however rigorous, is being made about the population that already enrolled. It says nothing about whether the design was ever going to reach the isolated adult who never got a referral, never saw the flyer, or doesn’t trust the venue. Fixing measurement inside a programme doesn’t fix a referral pathway that only surfaces people who were already connected enough to ask for help.