Short answer
Wellbeing ROI compares total programme and evaluation costs with benefits the evidence can reasonably attribute to the programme. Define the intervention and outcome, choose a credible counterfactual, record implementation and other changes, monetise only defensible benefits, avoid double counting, show uncertainty, and report non-monetised effects separately. A before-and-after improvement alone does not prove a return.
Start with the decision, not the formula
A finance or HR leader may need to decide whether to continue, expand, redesign or stop an investment. That decision requires more than programme uptake and a headline savings estimate.
Create an evaluation brief:
| Field | Decision to record |
|---|---|
| Investment | Exact service, work redesign or support being evaluated |
| Population | Who was eligible, reached and affected |
| Decision | Continue, change, expand or stop |
| Mechanism | How the investment is expected to affect a defined outcome |
| Implementation | What must be delivered before an effect is plausible |
| Outcome | Measurable change, period and data source |
| Counterfactual | Best estimate of what would have happened without the investment |
| Costs | Full resource use, including internal time and evaluation |
| Benefits | Observed, attributable, monetised and non-monetised kept separate |
| Distribution | Which groups bear costs or receive benefits |
| Uncertainty | Assumptions, range and evidence limits |
| Decision owner | Person accountable for the conclusion and next step |
The 2026 HM Treasury Magenta Book distinguishes process, impact and value-for-money evaluation. It is UK central-government guidance, but its sequence is useful for an employer: understand delivery, outcomes and attribution before making an economic claim.
Write the outcome chain
A wellbeing activity is not the same as a wellbeing outcome, and an outcome is not automatically a financial benefit.
| Stage | Example | Question |
|---|---|---|
| Input | Budget, staff time, systems and evaluation | What resources were used? |
| Activity | Work redesign, manager practice or support service | What was actually delivered? |
| Reach | Eligible people who could use or experience it | Who was included or missed? |
| Near-term result | Greater schedule predictability or service access | Did the proposed mechanism occur? |
| Outcome | Defined change in absence, work experience or another measure | What changed, for whom and when? |
| Attributable effect | Difference relative to a credible counterfactual | How much change can be linked to the investment? |
| Monetised benefit | Attributable effect multiplied by a defensible unit value | Which part can be expressed in money? |
If a meditation app was purchased but the actual problem was unsafe staffing, utilisation says little about whether the relevant work condition improved. If a rota redesign was not implemented as planned, the evaluation should show that before interpreting outcomes.
Build the full cost ledger
Record costs in the same period and scope as the benefits:
- supplier fees and licences;
- internal design, procurement and implementation time;
- employee participation time;
- manager and specialist time;
- communications, training and accessibility;
- data, legal, security and representative review;
- backfill, overtime or temporary cover;
- evaluation design, collection and analysis;
- disruption, transition and unintended costs;
- ongoing administration and exit costs.
Separate cash spending from staff-time or opportunity-cost estimates. State the unit rate and source for every conversion. Do not count the same hour as both absence avoided and overtime avoided unless those are genuinely separate resources.
Sunk costs may matter for accountability, while the forward decision may depend on avoidable future cost. Show both rather than changing the denominator to make continuation look attractive.
Choose a counterfactual that fits the rollout
The counterfactual is what would probably have happened without the investment. A pre-programme month is often a weak substitute because seasonality, staffing, policy, workload, labour markets and other initiatives also change.
Possible designs include:
| Situation | Evaluation option | Important limit |
|---|---|---|
| Rollout can be randomised fairly | Eligible units assigned to different start times or conditions | Check spillover, non-compliance and whether the comparison is ethical and practical |
| Similar untreated units exist | Matched comparison with pre-period checks | Matching cannot remove unobserved differences automatically |
| Many stable time points exist | Interrupted time-series design | Concurrent changes and trend assumptions still need review |
| Staggered operational rollout | Compare earlier and later groups under a suitable design | Rollout order may reflect risk or readiness |
| No credible comparison is feasible | Theory-based or contribution analysis | Supports a bounded contribution claim, not a precise causal ROI by itself |
The Magenta Book's current methods annex says costs and benefits should be assessed relative to the counterfactual. It also describes contribution analysis where an experimental counterfactual is not feasible. Method choice depends on the intervention, consequences, data and operating constraints.
Do not withhold support that people need solely to create a comparison. Evaluation, employment, health, safety, equality, privacy and representative requirements remain part of the design.
Separate observation, attribution and valuation
Use a claim ladder:
- Delivery claim: the planned activity reached the stated population.
- Outcome claim: the defined measure changed during the evaluation period.
- Contribution claim: evidence supports the programme as one plausible contributor and examines alternatives.
- Impact claim: a suitable design estimates the difference caused by the programme against a counterfactual.
- Financial return claim: attributable effects are converted with defensible values and compared with complete costs.
Each step needs stronger evidence. A positive employee account can help explain a mechanism. It does not establish prevalence, impact or monetary value. A correlation between participation and absence may reflect selection into the programme.
Keep the benefit ledger auditable
For every proposed benefit, record:
| Field | Entry |
|---|---|
| Outcome definition | Event, quantity, population and period |
| Observed change | What changed in the available data |
| Counterfactual estimate | What the selected method estimates without the programme |
| Attributable quantity | Difference supported by the evaluation |
| Unit value | Financial value and source |
| Monetised benefit | Attributable quantity multiplied by unit value |
| Uncertainty | Plausible range and key assumptions |
| Overlap | Potential double counting with another benefit |
| Distribution | People, teams or budgets affected |
Some benefits should remain non-monetised. A safer work design, fairer access or better employee experience may matter even when a defensible cash value is unavailable. Report the evidence and decision relevance instead of inventing a price.
Use the ROI formula precisely
The conventional calculation is:
ROI (%) = (attributable monetised benefits − total intervention and evaluation costs) ÷ total costs × 100
Also show:
Net benefit = attributable monetised benefits − total costs
If the evaluation supports only an observed association or contribution claim, label any numerical return as a scenario rather than a demonstrated ROI. Do not count outcomes that the evaluation cannot attribute.
A fictional, conditional calculation
Fairhaven Logistics is a fictional UK and US employer evaluating a shift-recovery programme. Its complete intervention and evaluation costs are £120,000. This includes supplier spend, internal design, paid employee time, manager time, backfill and evaluation.
A suitable impact design estimates a central attributable reduction of 3,000 paid absence hours during the defined period relative to its counterfactual. Finance uses a fictional, agreed fully loaded value of £40 per hour. The central monetised benefit is therefore £120,000. This is a resource and opportunity-cost proxy, not evidence that £120,000 of cash was released. A cash-saving claim would need separate evidence of lower overtime, cover or other expenditure, without counting the same hours twice.
Central ROI = (£120,000 − £120,000) ÷ £120,000 × 100 = 0%
The evaluation's sensitivity range is 1,000 to 5,000 attributable hours. It is an explicitly hypothetical range, not a published benchmark or a guaranteed confidence interval.
| Case | Attributable hours | Monetised benefit | Net benefit | ROI |
|---|---|---|---|---|
| Lower | 1,000 | £40,000 | −£80,000 | −66.7% |
| Central | 3,000 | £120,000 | £0 | 0% |
| Higher | 5,000 | £200,000 | £80,000 | 66.7% |
The calculation counts only paid absence hours. It does not add assumed turnover, productivity or healthcare savings, because those effects were not separately established and could overlap. Employee accounts and the wellbeing measure are reported as non-monetised evidence.
If the comparison design is later found unsuitable, Fairhaven cannot present the central 0% as an achieved return. It can show programme costs, observed changes and conditional scenarios while improving the evaluation.
Review uncertainty before the funding decision
Vary the assumptions that matter:
- attributable effect and its statistical uncertainty;
- unit value and whether it represents cash or opportunity cost;
- participation and implementation level;
- time horizon and persistence;
- displacement or spillover;
- missing data and workforce composition;
- cost overruns and recurring costs;
- outcomes omitted because they cannot be monetised credibly.
The 2025 UK government AQuA Book says analytical assurance should be proportionate to risk and cover validation, documentation and uncertainty across the lifecycle. For a large investment or consequential workforce change, independent analytical review may be appropriate.
Present the funding decision with the central estimate, range, evidence grade and non-monetised findings. Continue, redesign or stop because of the full case, not because one uncertain figure crossed zero.
Where Lontra fits
When an evaluation needs employee examples about a defined work change, explore Lontra's employee conversations and manager briefs. Managers receive a brief rather than raw employee conversations, and the evaluation team decides how that evidence can be used.
Lontra is not a wellbeing instrument, impact-evaluation design or ROI calculator. A conversation can support a mechanism or implementation review, but it does not establish a counterfactual, causal effect, attributable saving or financial return.
Frequently asked questions
How do you calculate wellbeing ROI?
Use ROI equals attributable monetised benefits minus total intervention and evaluation costs, divided by total costs, multiplied by 100. The calculation is credible only when costs are complete and the evaluation supports the attributed benefit against a suitable counterfactual.
Can lower absence after a wellbeing programme be counted as ROI?
Not automatically. The change is an observed outcome. Estimate what would probably have happened without the programme, examine other changes, avoid double counting, and state the uncertainty before treating any share as attributable benefit.
Does a negative financial ROI mean a wellbeing programme failed?
No. Financial ROI answers a narrow monetised question. Report delivery, outcomes, distribution, non-monetised benefits, duties and strategic value separately so decision makers can judge the programme on the relevant evidence.
