Did the grant cause the result? Attribution methods from before-and-after to contribution analysis

An illustrative bakery receives a GBP 10,000 grant for a new oven. Its turnover is 8 per cent higher after a 12-month period. A similar bakery without funding records a 6 per cent rise. The gap is worth examining, but these figures alone do not establish that the grant caused it.

A bakery result, checked in five moves

Treat the example as a question about evidence, not a search for the largest number. The bakery can record a rise, compare it with another business and examine other possible causes before deciding how strongly to describe the grant's role.

Reading what the bakery found

On these illustrative figures, the bakery has evidence of a rise and a comparison with another business. It does not have a precise causal estimate. The useful next question is whether the businesses and their earlier trends were sufficiently comparable, and whether other influences remain credible.

Before-and-after monitoring is inexpensive and straightforward to explain. It can show change, but other factors may account for some or all of it. Use several periods before delivery where possible, and document concurrent changes. The result should be presented as observed movement rather than a caused impact.

The target comparison answers a different question: did the project meet the plan? A target may rest on weak assumptions, and beating it does not show that funding produced the result. Keep the initial target and the way it was set on record, without revising either in response to actual performance.

A matched comparison can improve an estimate only when the selection rules are settled before analysis. For a UK small grant, compare filed accounts at Companies House for the funded firm with a similar local firm, using ONS business data for sector context. Check whether baseline trends are comparable, and remember that hidden differences can still remain.

Choosing the comparison after results appear invites a favourable but misleading match. A local benchmark is useful only when its selection and starting conditions are defensible. In this bakery example, the 2 per cent difference may be connected with the grant, but demand, pricing and seasonality could also explain it.

How to interpret evidence beyond the comparison

A comparison group is not the only route to a careful account. The method needs to fit the decision, risk, available data, ethics, cost and scale. Where several influences shape delivery, evidence may support a reasoned explanation of the project's contribution without supporting a precise causal percentage.

Experimental designs allocate exposure, often at random. Quasi-experimental approaches build a comparison from non-random data and stated assumptions. Both need specialist design, adequate sample size and ethical consideration. The 2026 Magenta Book explains the use of an unaffected group or period as a counterfactual and stresses suitable data quality and comparability.

Such designs may be infeasible or out of proportion for one small grant project. Seek an evaluation specialist before promising a causal estimate. A complex design is not automatically a better choice if the project cannot carry it out faithfully or the available data cannot support the comparison.

Contribution analysis offers another option for complex projects shaped by several influences. Set out how activities are expected to lead to outcomes, gather evidence along that route and test alternative explanations. Include conflicting evidence and describe the conditions under which the outcome occurred; this supports a reasoned account of the project's role, not a causal percentage by default.

Interviews, observation and case records can explain how delivery worked, who benefited and why. For a grant supporting ten firms, select five for interviews using clear criteria, such as high and low sales growth. Look for themes, check them against the contribution account and report cases or accounts that do not support it.

Set the effort before making the promise

Choose the evaluation question first, then decide what evidence is feasible to collect. The method should match the project's scale and the strength of claim needed. If there are too few units, no dependable comparison or no timely outcome measure, monitor results and investigate contribution instead.

The government's evaluation registry guidance describes proportionate evaluation: lighter monitoring for low-risk, well-evidenced activity and greater scrutiny for high-risk or novel interventions. Registry duties apply to specified government evaluations, not automatically to small grant recipients.

Weigh feasible choices against cost and timing and other practical limits. Run a feasibility review before promising an impact result: check there are enough units, that a comparison can remain uncontaminated and that the outcome can be measured at the right time.

Rules carried forward from the case

The bakery example points to a few practical rules: separate a recorded change from an explanation of its cause, decide comparison criteria before seeing results and match the method to the claim. State limits plainly, especially when other causes could explain the difference.