Topic 33 of 64· Social sector and philanthropy metrics
Cost per Outcome
Cost per outcome is total program spend divided by the number of people who achieve a defined, meaningful change in their circumstances — not the number who merely received a service. It is the sharpest efficiency metric a funder or delivery team can use, because it forces a prior question that most charities avoid: what, precisely, counts as success?
Why it matters
A food bank can report two very different numbers from the same year's accounts. Cost per food parcel distributed might be £15. Cost per household that goes on to achieve food security — no longer needing emergency food aid, verified at a follow-up point — might be £340. Both are true. Only one tells a funder whether the money is working. The gap between them is the gap between an output and an outcome: a parcel handed over is an output; a household no longer in crisis is an outcome. See outcomes vs outputs.
The UK third sector has spent two decades building infrastructure to force this distinction. New Philanthropy Capital's "four pillar approach" to charity effectiveness explicitly asks organizations to state their outcomes before their outputs, and Inspiring Impact — the UK funder-backed impact measurement collaborative — publishes an Outcomes Matrix that many grant applications now require charities to complete. The Trussell Trust's annual "State of Hunger" research program, run with Heriot-Watt University, exists precisely because parcel counts alone say nothing about whether people escape food insecurity.
Cost per outcome only means something once you have fixed the counterfactual: an outcome achieved "anyway" is not an outcome the program bought. See counterfactual analysis and displacement and attribution.
The math
Cost per outcome = Total program cost / Number of beneficiaries achieving the defined outcome
where:
Total program cost = direct delivery cost + fair share of overhead
Defined outcome = a pre-specified, measurable state change
(e.g. "food secure at 6-month follow-up",
not "received a food parcel")
Compare against unit cost databases (e.g. sector-specific unit cost benchmarks) to judge whether a given cost per outcome is good, average, or poor relative to comparable interventions.
Worked example
Food bank, one year:
- Total program cost: £450,000
- Parcels distributed: 30,000
- Cost per parcel (an output metric): £450,000 / 30,000 = £15
The charity also runs a six-month follow-up survey with a sample of households, finding that 35% of households who received three or more parcels report no longer needing emergency food aid and score above the food-security threshold on a standard food-security survey module. Of 1,800 households receiving three-plus parcels that year, 630 achieve that outcome.
Cost per outcome = £450,000 / 630 = £714 per household achieving food security
That £714 figure is the one a funder comparing this charity to a cash-transfer pilot or a debt-advice service should use — not £15. If a comparable cash-transfer program in the same region achieves food security at £500 per household, the food bank is not obviously the more efficient route to the same outcome, even though its per-parcel cost looks cheap.
Software engineering connection
Most case-management systems are built to log outputs, because outputs are what happens inside the transaction (a parcel is handed over, a form is submitted). Outcomes usually happen later, often outside the system's normal capture window, and require a deliberate design decision: build a follow-up mechanism (a survey trigger, a re-contact workflow, a data-linkage exercise) as a first-class feature, not an afterthought bolted on for an annual report. Engineers building grant management or case management platforms for the sector should treat "what is the outcome event, and how do we observe it" as a requirements question asked before the data model is fixed — it is far harder to retrofit an outcome field than an output counter. See outcomes vs outputs and logic model for how to structure that requirements conversation, and cost per beneficiary for the faster, cruder metric teams reach for when outcome tracking is not yet built.
Pitfalls
- Reporting outputs dressed as outcomes. "People reached" is not "people helped." If the metric can be produced by a system log with no follow-up contact, it is almost certainly an output.
- Denominator gaming. Narrowing the outcome population to "those who completed the program" quietly drops the people who dropped out — often the hardest cases — and inflates the apparent rate. State the denominator as everyone who started, not everyone who finished.
- No counterfactual. Counting anyone who achieved the outcome, including those who would have anyway, overstates what the program bought. See counterfactual analysis.
- Comparing across incompatible outcome definitions. "Food secure" measured by a validated survey module is not comparable to "food secure" self-reported in a satisfaction form; a cost-per-outcome league table is only honest when the outcome definitions match.
Sources
- New Philanthropy Capital (NPC), "Four Pillar Approach" to charity effectiveness. https://www.thinknpc.org/resource-hub/four-pillar-approach/
- Inspiring Impact, Outcomes Matrix and impact measurement resources. https://inspiringimpact.org/
- Trussell Trust and Heriot-Watt University, "State of Hunger" research program. https://www.trusselltrust.org/state-of-hunger/
- GiveWell, "Our criteria" (cost-effectiveness as the leading criterion for charity recommendation). https://www.givewell.org/how-we-work/our-criteria