Most businesses have at least one: the report somebody assembles every Monday from three systems and a spreadsheet, taking an hour, arriving in time for a meeting where three of its numbers are questioned.
Automating it is straightforward. Making it trusted is the actual work.
Start with the definitions, not the script
The reason manual reports survive is that they encode judgements nobody has written down.
What counts as an active customer? Does a job move to complete when the engineer finishes or when it is invoiced? Is revenue counted when invoiced or when paid? Do internal transfers count?
Whoever prepares the report answers these questions the same way each week, from habit. Automating without capturing those answers produces a report with different numbers, and everyone concludes the automation is wrong.
Sit with the person who prepares it and write the definitions down. This usually takes an afternoon and it is the whole project.
Then find where the numbers actually live
Frequently not where you would assume. The figure comes from a system, gets adjusted for something, has a known exception excluded, and is reconciled against another source.
Trace each figure to its origin. Where an adjustment is applied, decide whether it should be encoded or eliminated — sometimes the adjustment exists because a system is configured wrongly, and fixing that is better than automating around it.
What makes an automated report trusted
It reconciles to something people already believe. Run it in parallel with the manual version for a few weeks and investigate every difference. Some will be the automation; some will be the manual report having been wrong. Both are worth knowing.
Every figure can be drilled into. When somebody asks why revenue is down, "the report says so" is not an answer. Include a link or an attachment listing the underlying records.
It states its own definitions. A line at the bottom saying what period it covers, what is included and what is excluded. This prevents the recurring argument.
It says when the data was taken. A report run at 6am shows a different picture from one run at noon.
It fails loudly. If a source system was unavailable, the report should say so rather than quietly showing zero. A zero that means "no data" and a zero that means "nothing happened" must be distinguishable.
An automated report is trusted when it can be interrogated. One that produces numbers nobody can trace is a slower way to lose an argument.
Keep it short
The strongest temptation once reporting is automated is to add everything, because it is free.
Resist it. A report with forty figures gets skimmed; one with six gets read and acted on.
The test for each figure: has a decision ever been made because of it? If not, remove it. It can always live in a longer report somebody consults when they want detail.
Send it, do not host it
Dashboards require a decision to look. Most people do not make that decision, and a dashboard nobody opens is a dashboard that took a day to build for nothing.
Email the figures that matter, timed to arrive before the meeting or the moment they are useful. Reserve dashboards for exploring detail after the summary has raised a question.
Where a report leads to action, include the action. Not "twelve invoices are overdue" but twelve invoices listed with amounts and contacts, so the recipient can act from the email.
What to automate first
Look for the report that is produced most often, takes the longest, and is genuinely used. Those three together are rarer than they sound — plenty of long-standing reports fail the third test.
Common candidates: weekly sales and pipeline, aged debt, job completion and outstanding work, stock positions, and IT status checks such as those in the scheduled checks every business should run.
Before you build it, ask whether it should exist
Automating a report is a good moment to check whether anybody uses it.
Ask the recipients what decision it informs. A surprising number of recurring reports are produced for somebody who left, or are read but never acted upon.
Discontinuing one is cheaper than automating it, and nobody will notice. If they do notice, you have learnt that it mattered.
Keeping it working
Automated reports break silently when a source system changes. Build in an alert when the report fails to run, and review the definitions annually — businesses change, and a definition that made sense three years ago may not now.
Our automation team replaces manual reporting for UK businesses, starting with the definitions rather than the code. Start a conversation, or read automating business reporting for the wider approach.
