The Capability Gap: Training Staff for Asks the Evidence Can't Support

A commissioner asks for a projected reduction in GP visits. A board asks whether the peer support group will “cut loneliness by half.” A well-meaning trustee asks for evidence the walking group prevents dementia. None of these numbers exist in the literature at the resolution being asked for, and no amount of dashboard design will manufacture them. This is not a measurement problem. It is a staff capability problem: someone in the room has to say what the evidence can and cannot bear, in real time, without losing the funding or the relationship.

Most training on this topic teaches people what the evidence says. Almost none of it teaches people what to do when someone in authority asks for more than the evidence gives.

Why this keeps happening

The connection field has strong association evidence and weak intervention evidence, and the two get conflated constantly. Julianne Holt-Lunstad’s 2015 meta-analysis is genuinely robust: social isolation, loneliness, and living alone each carry mortality risk on the order of 26 to 32% increased odds. The American Heart Association’s 2022 scientific statement extends this to cardiovascular outcomes specifically, and states plainly that the absence of intervention evidence is the central gap in the field — it documents that isolation is dangerous, not that any particular programme reverses it. The National Academies’ 2020 consensus report calls for routine screening in health care settings, which is an assessment recommendation, not a demonstrated treatment effect.

Systematic reviews of social prescribing report positive individual-level outcomes such as self-esteem and confidence, but flag limited trial evidence and heavy heterogeneity across programmes. A 2025 review protocol notes that only one peer-reviewed randomised controlled trial exists for social prescribing’s effect on isolation in older adults, despite widespread adoption. Where randomised trials do exist — the HEAL-HOA volunteering trial, a 2025 befriending trial in aged care, and the 2026 HEAL-HOA behavioural activation trial — they show real but modest effects, and they show that some widely used interventions (befriending) are outperformed by others (structured behavioural activation) in head-to-head comparison. That is a useful, humbling finding. It is not evidence that any single group activity will deliver a named percentage reduction in loneliness for a named population by a named date, which is what commissioners routinely ask for.

The framework: three tiers of claim

Train staff to sort every request into one of three tiers before answering.

Tier	What it asks for	What you can honestly say
Population-level association	“Is loneliness dangerous?”	Yes, with strong meta-analytic backing. Cite it freely.
Programme-level plausibility	“Could this kind of activity help?”	Plausible, supported by some trial and observational evidence, but effect sizes are modest and heterogeneous.
Programme-specific prediction	“Will this group cut loneliness by X% for these people?”	No source supports a number at this resolution. Say so.

The failure mode — call it borrowing the wrong tier — is answering a Tier 3 question with Tier 1 evidence. It sounds credible in the room and it collapses at renewal, because nobody can reproduce the number that was never real.

What to train people to say

Staff need three rehearsed moves, not a lecture on epidemiology.

1. Name the tier out loud. “The mortality link is well established. What this specific group will do to that risk for this cohort isn’t something anyone can measure reliably in a twelve-week pilot.” Say it before being asked to defend a number nobody can produce.
2. Offer the evidence that does exist, precisely. If asked for an outcome claim, cite the closest real trial and its actual effect size — the aged-care befriending trial’s 2.39-point reduction on the UCLA scale at eight weeks, for instance — rather than a rounder, more flattering number that isn’t anyone’s finding.
3. Redirect the ask toward what can be measured. Reach, attendance, retention past week four, referral completion. These are answerable now, with the data the programme is already generating, and they are what a funder can actually be shown next quarter.

What this means in practice: before a bid, evaluation plan, or board update goes out, someone should be assigned to check every outcome claim against this table and strike or rephrase anything in Tier 3 that isn’t sourced to a specific trial. This is a five-minute review, not a research exercise, and it is the single most common gap in connection-programme reporting.

Training this as a skill, not a caveat

Teaching staff to hedge everything is as useless as teaching them to oversell. The skill is precision under pressure: state the strong claim plainly, name the weak one honestly, and do it without apologising for the field’s evidence gaps as though they were a personal failing. The difference-in-differences evaluation of the UK’s national loneliness campaign is worth using in training precisely because it is one of the rare controlled evaluations of a population-level intervention in this space — it shows what rigorous testing actually looks like, and how rare it still is. Staff who have seen one real evaluation design tend to stop inventing precision they don’t have.

What this does not solve

This framework helps a staff member survive a conversation with a funder honestly. It does not close the underlying evidence gap — the field still has far more association data than intervention data, and that will not change because one team got better at saying so. It also does nothing for programme reach: the training makes existing staff more precise with existing participants, and says nothing about the people who never got referred, applied, or heard about the programme at all. That is a different problem, and no amount of honest language fixes it.