Attributing startup outcomes to funding, from before-and-after review to experiments

Attribution asks whether funding caused an outcome. Revenue can rise after a round because of earlier sales work, market conditions, staff or chance. Choose a method that matches the decision, evidence, ethics and cost. State uncertainty rather than forcing a causal percentage.

What to take away

Method comparison

Effort and cost ranges are illustrative for a small English study and depend on data access, sample size and review.

Use a counterfactual

Write what would probably have happened without the funding: slower growth, smaller milestone, customer finance, debt or closure. Preserve the forecast made before the result. A counterfactual invented afterwards tends to make the chosen route look better.

HM Treasury's quality guidance for impact evaluation distinguishes observed change from causal impact and examines counterfactual quality. Its public-policy setting differs from a board review, but the reasoning is relevant.

England funding context

For English companies, common funding contexts include SEIS and EIS rounds, Innovate UK grants and local growth grants from Growth Hubs or combined authorities. Companies House filings show accounts, charges and ownership. Beauhurst tracks private company funding rounds. Innovate UK reporting lists grant awards. Use these sources to define funding exposure and possible comparison pools.

Worked example: funded growth against a comparison group

Illustrative figures, not a real company. A software startup in England raises £500,000 through an SEIS and EIS round. Monthly revenue is £30,000 before the round. Twelve months later it is £45,000, so the before-and-after change is £15,000 a month.

A matched group of similar unfunded English software companies moves from £28,000 to £34,000 a month over the same period, a change of £6,000. The difference-in-differences estimate is £9,000 a month. The matched comparison alone, taking post-round revenue, would show the funded company £11,000 ahead of the matched average. Both figures are illustrative.

Match method to claim

A company asking whether capital arrived on time needs process measurement, not causal inference. A grant programme claiming jobs created because of awards needs a stronger design than recipient testimonials. An adviser reporting funds raised can describe transactions without claiming sole credit.

Define exposure, outcome, period, population and confounders. Register the main questions and analysis before inspecting outcomes where the claim carries public or financial weight.

Contribution analysis

Build a chain from funding to activity, output and outcome. Test each link with documents, interviews and operating data. Search for alternative explanations and contradictory cases. This method can support a reasoned account without pretending all other influences disappeared.

Quasi-experimental checks

Matching and difference-in-differences need suitable data, documented inclusion and specialist review. Report balance, missing data, pre-trends and sensitivity. Do not call two convenient companies a control group.

Randomisation can be powerful but must fit the programme's authority, fairness and ethics. It is rarely a sensible tool for one startup choosing capital.

Publish limits

State data, method, assumptions, uncertainty and conflicts. Separate gross funding, net proceeds, activities and later outcomes. Archive code and definitions where possible without exposing confidential or personal data.

The government's Evaluation Registry guidance describes central-government registration duties and proportionality for relevant evaluations. A private startup is not automatically subject to that process.

Quality checklist

Record who was included, how funding exposure was defined, when outcomes were measured and how missing data were handled. Check whether a comparison group could have received other support and whether companies selected for funding already differed.

Use sensitivity checks for material assumptions. Have a reviewer challenge the causal language against the design. "Associated with", "contributed to" and "caused" are different claims and should not be selected for marketing effect.

Keep unsuccessful companies and null findings where the method includes them. Excluding them after results are known can turn an evaluation into a case-study collection.

Treat this comparison as a starting point and record the review each claim needs.

Before you act

Common questions

Why is a before-and-after review often insufficient for causal claims?

Because time and other changes are confounded. Revenue can rise after a round due to earlier sales work, market conditions, staff or chance. The method compares the same company across time, which is useful for a quick operating review but cannot isolate the effect of funding.

When is a forecast against actual comparison useful?

When the board recorded assumptions before funding. It compares approved unfunded and funded plans, but forecast error affects the comparison.

What should be reported when publishing attribution results?

Separate gross funding, net proceeds, activities and later outcomes. Archive code and definitions where possible without exposing confidential or personal data. Use sensitivity checks for material assumptions and have a reviewer challenge the causal language.