When a Funder Wants Outcomes the Evidence Cannot Promise

A commissioner asks for a 20% reduction in GP attendance among referred patients within six months. A funder wants you to show reduced loneliness scores by the next reporting period, with a comparison to a control group you don’t have the budget to recruit. A trustee reads a headline about social prescribing “cutting NHS costs” and wants that written into next year’s targets.

None of these asks are unreasonable on their face. They’re the wrong ask, and saying so badly costs programmes their funding. Saying so well changes what gets promised — and protects the programme when the promised number doesn’t show up.

Step 1: Separate what the evidence supports from what it doesn’t

Before you respond to any ask, sort the claim into one of three bins. Most disputes with funders happen because everyone is arguing about the same sentence while meaning different things by it.

Claim	Evidence status
Social prescribing improves self-esteem and self-confidence for participants	Supported. A 2021 systematic review reports these as consistent outcomes across included studies.
Social prescribing reduces loneliness for at least some participants	Supported, cautiously. A 2021 systematic review of nine studies found all reported positive individual impacts on loneliness.
Structured, purposeful group activity works better than casual social contact alone	Supported by qualitative synthesis. A 2022 meta-synthesis found participants describe benefit tied to restored participation and purpose, not contact volume.
Social prescribing reduces GP or A&E attendance at a predictable rate	Not supported at population level. Only three of nine studies in the 2021 review reported any reduction in service use, and none offer a rate you can commit to.
A specific programme will show a measurable drop in loneliness scores within six months	Not supported. The intervention literature is dominated by small, uncontrolled evaluations. The 2023 BMC review names inconsistent measurement across studies as a structural barrier to comparison.
Social isolation causes downstream cardiovascular harm that a referral pathway can prevent	The isolation-mortality link is well established; the American Heart Association’s 2022 statement explicitly names the absence of intervention evidence as the field’s central gap. Prevention plausibility is not the same as programme-level proof.

Notice the pattern: outcomes about how participants feel and what they report are reasonably well supported. Outcomes about downstream service use, cost savings, and fixed-timeline change are not — because almost nobody has run the randomised trial that would establish them. The 2024 HEAL-HOA trial testing volunteering against a control for lonely older adults in Hong Kong is one of the few exceptions, and it is one trial, in one population, testing one intervention.

Step 2: Rewrite the ask, don’t just refuse it

Refusing an outcome without offering a substitute reads as evasion. Offer the version the evidence can actually carry.

1. If the ask is “reduce GP attendance by X%,” counter with “track self-reported confidence, participation frequency, and (where consented) service contact over 12 months, reported as a pattern, not a guaranteed percentage.”
2. If the ask is “show reduced loneliness by six months,” counter with “measure loneliness at intake and at a defined interval using a validated scale, and report the distribution of change, including for participants who don’t improve.” A six-month promise implies a causal claim the design can’t support; a measured pattern doesn’t.
3. If the ask is “prove this saves the health system money,” counter with “report the plausibility case — citing the isolation-mortality literature — alongside participant-level outcomes, and be explicit that a cost-offset claim requires a controlled study this programme is not resourced to run.”

What this means in practice: never let a funder write “reduction in GP attendance” into a grant agreement as the headline metric unless the programme has the budget and design for a controlled comparison. Negotiate the target down to what a routine evaluation can actually show — self-reported wellbeing, participation, retention — and report service-use data as exploratory, not as the proof point.

The week-four problem, restated for evaluation

Programmes that overpromise on outcomes tend to also overpromise on timeline — the same failure mode that shows up in attendance drop-off shows up in reporting. A funder expecting a six-month loneliness reduction will read a flat or mixed result at month six as failure, when the honest read is that six months was never long enough, and the scale used may not be sensitive to the change that did happen. Fix the timeline expectation at the negotiation stage, not the reporting stage.

What to put in writing before the programme starts

• The specific outcome measure, the instrument used to capture it, and the reporting interval — agreed in the funding agreement, not assumed.
• An explicit statement that service-use reduction, where reported, is observational and not causally attributed to the programme.
• A named evidence source for every plausibility claim in the bid — Klinenberg’s third places, the National Academies’ 2020 call for routine isolation screening — kept distinct from claims about this programme’s own results.

What this does not solve

This approach makes the ask honest. It does not make the underlying evidence base stronger, and it does not solve the fact that the strongest data — the AHA’s cardiovascular findings, the National Academies’ isolation-in-older-adults report — describes risk, not intervention effect. It also does nothing for reach: an evaluation built this carefully still only tells you what happened to the people who showed up, referred by a system that mostly reaches people already in contact with it.