Build the report around a decision
A report should help someone decide what to do next. For an owner, that may mean following up overdue invoices or understanding a change in enquiries. For an operations manager, it may mean reviewing unfinished work and capacity. Start with the decision, the audience and the reporting period.
Agree the meaning of each measure before connecting data. “Revenue” could refer to invoiced amounts, recognised revenue or cash received. “New customers” needs a definition too. If the team uses different meanings, an automatically produced report will repeat the disagreement faster.
Collect figures from agreed sources
Identify the system responsible for each figure and the cut-off time for the report. Keep the reporting period consistent, including time zone and treatment of late updates. Preserve enough detail to trace a headline number back to the source records.
Missing data should be marked unavailable, not silently replaced with zero. If a connection fails, show that the report is incomplete or keep it in review. A manager should be able to distinguish “no sales” from “sales data did not load” without inspecting the integration.
Use AI for commentary, not arithmetic
Reliable calculations should produce the totals and comparisons before AI receives them. A model can help express those results in plain language, highlight movements and prepare questions for the report owner. Give it the reporting definitions and the verified figures it is permitted to discuss.
A rise in costs does not by itself establish a cause. Commentary should distinguish observed changes from possible explanations that need checking. If there is no supporting evidence, the report should say what changed and ask for investigation rather than inventing a story about why.
An example of weekly operational reporting
A proposed weekly report could bring together new enquiries, accepted work, open tasks and overdue invoices. Each section would show its source and reporting period. A reviewer checks unusual movements, adds context and approves the report for the intended recipients.
Keep access proportional to the audience. A team-level operational update may need task counts rather than individual client financial details. Choose the distribution list explicitly, and avoid automatically broadening access because someone was copied into an earlier email.
Check that automation improves the reporting process
Measure preparation time, corrections after distribution and the number of reports delayed by missing data. Ask recipients whether the report helps them make the intended decisions. A fast report that cannot be reconciled or understood is not a useful improvement.
Bring an existing report, definitions of its measures and a list of source systems to the scoping discussion. Run the automated version alongside the current report and investigate differences before relying on it. Reporting automation should make the numbers easier to verify as well as easier to assemble.
A workflow to discuss
Illustrative steps to adapt to your systems and approval requirements.
- Define the audience, reporting period and measures.
- Collect source figures and check completeness.
- Calculate comparisons and prepare a commentary draft.
- Review exceptions and distribute the approved report.
Common questions
Can AI create financial reports?
AI can help draft commentary and organise information from verified financial data. Totals, reconciliations and accounting decisions still require dependable processes and appropriate review. Keep the distinction between a draft explanation and an approved financial report clear throughout the workflow.
Can we automate spreadsheet reports?
Potentially. First identify the inputs, formulas, manual adjustments and decisions hidden in the current spreadsheet. Some steps can be scheduled directly; others need clearer rules or a review stage. Compare the new output with the existing report across representative reporting periods before replacing the manual process.
What happens when data is missing?
The workflow should flag the missing source and either hold the report or clearly identify the affected sections. Do not convert an unavailable figure into zero. Assign an owner to the exception and define whether a corrected report needs to be distributed when the data becomes available.
Start with one process
Bring the task your team keeps repeating, the tools you use and a few examples with sensitive information removed. Discuss where automation could help and how to assess the result.
Talk to David