Did the accountant make the difference? Five methods for testing the causal claim

Monitoring can show that a result changed after an accounting service began. It does not automatically show how much the service caused. Choose an attribution method that matches the decision, scale, available comparison and consequences of being wrong.

What to take away

Establish the causal question

Specify the intervention, population, outcome, period and alternative explanation. "Did reporting improve?" is too vague. Ask whether introducing a monthly close process reduced comparable correction work for a defined group over six months, after accounting for staffing and software changes.

Record a theory of change: which activity should alter which behaviour, producing which outcome. This exposes results that are too remote to claim confidently.

Method 1: simple before and after

Compare the same measure before and after the change. This is inexpensive and useful for monitoring, but seasonality, growth, new staff and other interventions can explain the difference. Use several comparable periods and label the result as association unless stronger evidence exists.

Illustrative figures: a practice records mean correction hours per month. Before a new close process, the average is 18 hours. After, it is 12 hours. The six-hour fall is an association until a stronger design tests rival causes.

Method 2: interrupted time series

Use repeated measures before and after the change to examine whether level or trend shifted. It is stronger than two points, but needs consistent definitions, enough observations and assessment of events occurring at the same time. A common working minimum is eight observations before and eight after, though more is better.

Method 3: matched comparison

Compare similar clients, teams or tasks where one receives the change and another does not. Match important factors such as entity, volume, record quality and complexity. Remaining differences can still bias the estimate, and withholding a necessary control may be unacceptable.

Method 4: phased introduction

Introduce the service at different times across suitable groups, measuring each under the same definitions. This may create a practical comparison while supporting implementation. Allocation chosen for convenience or risk can make groups systematically different, so record the rationale.

Use at least two groups and two periods per group. More groups improve precision.

Method 5: contribution analysis

Gather a structured chain of evidence that the service plausibly contributed, examine rival explanations and update the claim. This helps with complex outcomes where controlled comparison is impractical, but it cannot manufacture a precise causal percentage from interviews and a plausible story.

HM Treasury's current Magenta Book and impact-evaluation quality guidance apply to government policy evaluation. They provide useful methodological distinctions between observing an outcome and attributing it to an intervention, with proportionate design choices. A private accounting practice should adapt the concepts rather than claim compliance with a government evaluation standard.

For UK firm-level comparison, Companies House filed accounts and Office for National Statistics business data provide public benchmarks.

Compare methods consistently

Assess the outcome measure before debating causal design. Missing records, changed clocks or inconsistent correction categories can invalidate every method. Freeze a specification, test completeness and record later amendments. Where groups know they are being compared, consider whether changed behaviour affects the result.

Look for displacement as well as benefit. A faster month end may move work into client preparation or reviewer rework.

The conclusion should state what changed, what comparison was used, which other causes remain and how confident the team is, without converting uncertainty into a marketing promise.

Before you act

Common questions

How do I choose between the five attribution methods?

See the takeaway list and the method notes above.

What are the limits of contribution analysis?

See Method 5 above. It supports a plausible contribution claim, not a precise effect size.

Why should I measure the whole pathway rather than one improved step?

Displacement can offset the apparent benefit; see the displacement note above.