Evaluating a Workplace Connection Programme: What You Can Actually Measure

An HR team runs a “connection month,” puts on three lunch-and-learns and a step challenge, and closes it out with a survey asking employees to rate their satisfaction on a scale of one to five. The average comes back at 4.1. Someone puts that number in a slide for the executive team, and the programme gets funded again next year. Nothing in that sequence tells you whether anyone became less lonely, less isolated at work, or more likely to stay.

This is not a hypothetical failure specific to one organisation. It is close to the default design across employers who take workplace loneliness seriously enough to act on it but not seriously enough to measure it properly. Gallup’s 2024 State of the Global Workplace found that one in five employees worldwide report feeling lonely a lot of the previous day, rising to 25% among fully remote staff versus 16% among fully on-site employees. Cigna’s 2025 survey put loneliness among U.S. workers at 52%. Those are good numbers for justifying a programme. They are useless for evaluating one, because they describe prevalence, not change, and prevalence surveys are not built to detect the effect of an intervention.

The single-item trap

Call this the single-item trap: measuring loneliness with one satisfaction question asked once, after the intervention, with no baseline and no comparison group. It produces a number that always looks good, because people who show up to a voluntary programme and are asked how they feel about it tend to say something positive. It cannot distinguish a genuine reduction in loneliness from simple politeness, from the natural mood lift of free lunch, or from the fact that lonelier employees may not have shown up to the survey at all.

The single-item trap survives because the alternative looks expensive. It is not, if you scope it correctly. What is expensive is running a research-grade randomised controlled trial inside a company, which is precisely why so few employers attempt one, and why the intervention evidence for loneliness overall remains thin. The National Academies and the American Heart Association’s 2022 scientific statement on social isolation both flagged the shortage of controlled intervention studies as the field’s biggest gap, not the shortage of prevalence data. You are not going to close that gap from inside an HR department. But you can build something closer to it than a satisfaction score.

What a workable design actually needs

Three components separate a design you can defend from one you cannot: a validated measure, a baseline taken before the intervention starts, and something resembling a comparison.

A validated measure, used consistently. Drop the bespoke satisfaction question. Use a short, validated loneliness or connection instrument — a three-to-six item version of a scale such as the one behind the meta-analyses on social isolation and mortality — and use the same wording before and after. Holt-Lunstad’s 2015 meta-analysis in Perspectives on Psychological Science is built on studies using consistent instruments precisely because inconsistent measurement makes results incomparable across time and across organisations. If your survey wording changes between rounds, you have broken the one thing that made comparison possible.

A baseline, taken before anyone knows what is coming. Measure before you announce the programme, not after people have opted in. This matters because the people most likely to sign up for a connection programme are often not the most isolated employees in the building; they are the ones already inclined toward social contact. A baseline taken from the whole team, before self-selection happens, lets you see whether the programme reached anyone beyond its own natural audience.

A comparison, even an imperfect one. A true control group is rarely available inside a single company. A staggered rollout is the realistic substitute: run the programme in one department or site this quarter and another next quarter, and use the not-yet-enrolled group as a rough comparison. It is not a randomised trial. It is far better than a before-and-after on the same group with no comparison at all, because it lets you separate “people got less lonely” from “everyone got less lonely because it’s spring.”

What this means in practice: if you can only afford one methodological upgrade this year, make it the baseline. A validated post-only survey with no baseline cannot show change. A crude pre/post with a staggered rollout, even with a rough survey instrument, can.

What the intervention literature actually supports

Be specific about what you are trying to move, because the outside evidence on what works is stronger for some intervention types than others, and workplace practitioners tend to import the wrong one.

Claim	Evidence status
Loneliness is elevated among remote and hybrid workers relative to on-site staff	Reasonably solid — Gallup’s 2024 workplace data and a 2025 nationally representative study in the Journal of Affective Disorders both find this pattern, though effect sizes and causal direction vary
Workplace isolation and personal loneliness are distinct constructs, not the same problem	Solid — a 2024 study of healthcare workers found perceived social support moderates the relationship between remote work and wellbeing, implying the two need separate measures
Structured, purposeful group activity reduces loneliness more than unstructured social contact	Moderate — supported by qualitative synthesis in the social prescribing literature, but few controlled comparisons exist inside workplace settings specifically
One-to-one befriending reduces loneliness scores	Solid for the intervention itself, weak for workplace transfer — a 2025 randomised trial found befriending cut UCLA Loneliness Scale scores by 2.39 points at 8 weeks in aged care, a very different population from employees
Volunteering-based prosocial engagement reduces loneliness more effectively than generic social contact	Moderate — the HEAL-HOA randomised trial found structured psychological and prosocial approaches outperformed simple contact among lonely older adults, again outside a workplace context
A one-off event (lunch, mixer, step challenge) meaningfully reduces sustained loneliness	Unsupported — no cited evidence tests this directly, and the mechanism (brief, low-structure contact) is the type repeatedly outperformed by structured approaches in the studies that do exist

The pattern across the intervention studies that do have controls is consistent: structure and purpose beat unstructured contact, and one-off contact is the weakest form there is. That should worry any employer whose entire loneliness programme is a single “connection week.” It should also reset expectations about what a lunch-and-learn is realistically capable of moving, which is attendance and short-term mood, not a validated loneliness score three months later.

Building the business case without overclaiming

If you are taking this to a budget holder, resist pairing a soft measure with a hard cost claim. Cigna’s 2020 loneliness index found lonely workers miss work roughly twice as often due to illness and five times as often due to stress — a striking number, but it describes the general association between loneliness and absence in a national workforce survey, not the effect of any specific programme you might run. Use it to justify why the problem is worth addressing. Do not use it to project savings from your intervention, because no cited study connects a specific workplace connection programme to a specific absenteeism reduction with the rigor that number deserves.

What you can honestly report to a funder or executive team, if you run the design above, is a before-and-after change on a validated scale, compared against a group that has not yet had the programme, alongside a straightforward operational metric like attendance or retention within the participating group over the following two quarters. That is a modest claim. It is also one that survives someone asking a hard question in the room, which a satisfaction score of 4.1 does not.

What this does not solve

This design tells you whether people who took part changed. It does not tell you whether the programme reached the employees who needed it most — the evidence on remote and hybrid loneliness suggests they are disproportionately dispersed, less visible, and less likely to opt into a voluntary lunch event in the first place. A staggered rollout controls for time; it does not control for who signs up. And because almost none of the underlying intervention evidence was generated in workplace settings, every conclusion you draw from a company-level evaluation is an extrapolation from a different population, applied to yours. Treat it as a working estimate, not a verified result.