From source files to a report you can check.

Synthetic demonstration. This is an illustrative method, not a client case study or a live connection to a reporting system. Every value is shown so you can reproduce the result.

Try a reporting failure

A useful report knows when to stop.

This synthetic report compares April and May enquiries. Introduce a source problem and inspect the release decision.

Try a reporting failure controls

May source export is missing

The enquiry definition changed

Your choices stay in this page. No sign-up required.

01Import02Validate03Calculate04Release

DRAFT: enquiries increased by 25%

April: 24. May: 30. Increase: 6. Calculation: 6 ÷ 24 × 100 = 25%. The figures show a change; they do not identify its cause.

A human checks definitions, dates, sources and wording before release.

Assumptions and how to use this tool

Synthetic values: April 24 enquiries; May 30. Change = (30 − 24) ÷ 24 × 100 = 25%. This shows a reporting rule, not a live connector or a client result.

1. Define the question before building the report

The example question is: “How did recorded enquiries change between April and May?” The report owner needs a comparable count, the size of the change and any reason to withhold the comparison. It does not need a model to invent why the change happened.

Before implementing this for a team, define what counts as an enquiry, how duplicates are handled, the time zone and period boundaries, and which source is authoritative. If an enquiry becomes a qualified lead later, keep those measures distinct.

2. Keep the source and calculation inspectable

Period	Recorded enquiries	Source status
April	24	Synthetic complete export
May	30	Synthetic complete export

The absolute change is 30 − 24 = 6 enquiries. The percentage change is 6 ÷ 24 × 100 = 25%. These calculations belong in a repeatable formula or transformation. A language model can explain the result, but it should not be the only place the arithmetic exists.

If the earlier value is zero, the percentage change is undefined. Show the two counts and the absolute change instead of dividing by zero or displaying an invented percentage.

Download the synthetic source CSV. It contains the two rows shown above and an explicit synthetic-data label.

3. Decide what prevents release

The switches above cover two different failures. A missing May export means the source is absent, not that May had no enquiries. A changed definition means both counts may exist, but a percentage comparison would be misleading. These conditions need different explanations and the same discipline: do not publish an unsupported conclusion.

• Missing source: hold the affected report, name the missing input and notify its owner.
• Changed definition: resolve the difference or mark a break in the series. Keep the original source values available.
• Unexpected movement: request investigation. Do not automatically treat an unusual result as an error or explain it away.
• Failed delivery: record the failure and alert the responsible person. A generated report is not proof that its recipient received it.

4. Write commentary that stays within the evidence

A supportable draft is: “Recorded enquiries rose from 24 in April to 30 in May, an increase of 6 or 25%, using the same enquiry definition.” If comparability has not been confirmed, say so. “The campaign caused a 25% increase” needs evidence that this example does not contain.

The reviewer checks definitions, periods, source freshness, arithmetic and wording. They also decide whether the change warrants action. Automation can prepare the material for that decision; it does not remove responsibility for it.

5. Compare complete reporting cycles

Measure collection, cleaning, formatting, commentary and review before the change. Afterwards, include exception handling, corrections and maintenance. Use several comparable cycles so a quiet month or a one-off source problem does not distort the conclusion.

Use the reporting capacity calculator to explore a scenario, then replace assumptions with measured values. Time released becomes a financial saving only when an actual cost is avoided.

Practise the review yourself

For a short individual exercise, use Jon Goodey’s first AI reporting workflow lesson. Choose a source condition, decide how the draft should respond and compare your reasoning with the evidence.

A report acceptance checklist to take away

Download the report acceptance checklist. It covers source ownership, definitions, validation, review, delivery and recovery. Adapt it to one report before expanding the scope.

If the figures themselves are not trustworthy yet, start with marketing analytics and measurement. If people need practice checking AI-assisted commentary, compare the separate ChatGPT and Claude training routes.

Questions people ask

Is this a real client reporting result?

No. The figures and source conditions are synthetic. They demonstrate the controls a reporting workflow needs, without using client data or claiming a measured saving.

Why should an automated report stop?

A report should stop or clearly mark an affected section when required inputs are missing, stale or not comparable. A confident-looking output is not useful if its evidence is incomplete.

Where does AI fit in the reporting workflow?

AI may help draft commentary from checked inputs. Reproducible calculations, source validation and human review should remain explicit. The example on this page runs without an AI model.

How should we measure the benefit?

Compare several equivalent reporting cycles before and after the change. Include source collection, preparation, review, corrections and upkeep. Separate capacity released from actual costs avoided.