How to automate marketing reports without losing the judgement

The aim is a report that helps somebody decide what happens next.

Start with the decision the report supports

Different readers need different reports.

A campaign manager may need to identify budget or tracking problems. A commercial director may need to understand acquisition costs and the quality of the resulting business. A client may need a clear explanation of progress against an agreed objective.

Write down the decisions before selecting the charts.

For each metric, ask: what would we do differently if this changed?

If nobody can answer, reconsider its prominence.

Agree the definitions before connecting the data

Reporting becomes unreliable when the same label describes different things.

Does “revenue” mean platform-attributed revenue, gross order value or sales after refunds? Does a “lead” mean every form submission or an enquiry that sales has qualified?

Document the definition, source, reporting period and known limitations of each important measure.

Keep platform-attributed conversions separate from deduplicated business outcomes. Where attribution overlaps, adding together the totals from advertising platforms can misrepresent the number of sales or enquiries received.

Check the data before generating commentary

A reporting workflow should identify missing, stale or inconsistent inputs before attempting to explain performance.

Useful checks include:

• Whether every expected source supplied data.
• Whether the reporting periods, currencies and time zones align.
• Whether the extract is complete.
• Whether conversion definitions changed.
• Whether a zero represents no activity or missing information.
• Whether any comparison period includes an unusual promotion or outage.

When an important source fails, make the limitation visible. The report can still contain useful information, but its conclusions should reflect what is available.

Keep calculation and interpretation distinct

Calculate percentages, totals and ratios through defined formulas. Give those results to the writing stage alongside business context and supporting evidence.

Here is a fictional example:

A useful observation is:

Lead volume fell 10% alongside a 10% reduction in spend. Cost per lead remained £200. The next review should establish why spend fell and whether the quality of leads changed.

The numbers support the observation. They do not establish that competition, creative fatigue or a website change caused it.

An automated draft should distinguish what happened, what might explain it and what needs checking.

Give every recommendation an owner

Use a short decision record:

This gives the next report something useful to revisit. Did the team complete the action? Did the expected result occur? Does the recommendation still hold?

Measure the reporting system itself

Track preparation time, correction rates, delivery reliability and whether the report leads to completed actions.

During a pilot, compare automated drafts with a manually checked version. Include awkward cases such as a missing source, an unusual promotion and a change in tracking.

Peach & Stone’s reporting approach connects data preparation with the commercial questions a team needs to answer. Human review remains part of the operating process.

Next step: Bring one recent report and explain what you wish it helped you decide. That is a practical starting point for improving the workflow.