Reporting automation: the weekly number written by the run, not by a person on Monday.
A report assembled by hand is stale before it is read and disputed the moment it is. The fix is to have the work write it.
The same work, twice
- 1Export from the ERPOps · 10m
- 2Export from the WMS and the CRMOps · 15m
- 3Clean the CSVs and fix the date formatsOps · 20m
- 4Rebuild the pivot and check the totalsOps · 25m
- 5Chase finance about a number that movedOps · 20m + wait
- 6Write the commentaryOps · 20m
- 7Paste it into the deck and sendOps · 15m
- 8Answer the follow-up questionsOps · 15m
The shape of the run. Coastline's real report is appended by the order flow itself, which is the cheaper version of this.
Why this queue costs what it costs
Reporting automation usually gets sold as a dashboard, and dashboards are the easy half. The reason Monday costs two hours is not the chart. It is the exporting, the cleaning, the reconciling of two systems that disagree, and the twenty minutes spent finding out why a number moved.
Coastline Logistics did not buy a reporting tool. Their weekly report is appended by the order run itself, so it is current after every batch and it uses the same numbers the operation ran on. The reconciliation meeting stopped happening because there was only one version.
Where a process is not automated yet, we build the reporting run separately: pull the sources, reconcile them, find the variances, trace each one back to its rows, and draft commentary a person edits rather than writes.
The report should be a byproduct, not a task
If a process already runs as an automated flow, the report is nearly free: each run appends what it did, so the weekly view is a query rather than an assembly. That is the cheapest reporting you will ever buy and it is why we ask what else you are automating before quoting this on its own.
Where the process is still manual, the reporting run stands alone and pulls from the systems of record. It is more work and it is still worth it, but the honest sequencing advice is to automate the process first where you can.
Reconciliation is the actual product
Two systems disagreeing is not a data problem to be hidden behind an average. It is usually a real operational fact: an order counted at dispatch in one system and at invoice in another, a timezone boundary, a cancellation applied retroactively.
So the run reconciles explicitly and reports the difference rather than silently picking a side. Teams find this uncomfortable for about two weeks and then rely on it, because the standing argument about whose number is right ends with a link to the rows.
Variance commentary, drafted and attributed
The sentence a report needs is why. Dispatch volume is down 9% week on week, driven by two suppliers confirming late on Tuesday, and here are the eleven orders involved.
A model is good at drafting that sentence from the data and terrible at owning it, so each variance is attributed to a named owner who confirms or corrects the explanation before the report goes out. The draft removes the blank page; the person keeps the judgement.
One number, one definition
Most reporting disputes are definitional. Does an order count when placed, picked or dispatched. Is a lead qualified when scored or when it books. Does churn include downgrades.
The build forces those definitions into one file with a version on it. That file, more than any chart, is what stops three teams reporting three different figures to the same board meeting, and it stays yours after handover.
Where it is delivered matters more than how it looks
The report people read is the one that arrives where they already are: a Slack channel at 06:00, an email with the three numbers in the subject line, a refreshed deck before the Monday call. A dashboard nobody opens is a report nobody reads.
We build to the destination first and the interface second. If a dashboard is genuinely the right surface, we build one, but only after the delivery habit exists.
Cost, timeline and honest sequencing
A reporting pilot starts at the $4,000 floor and typically takes about three weeks: two or three sources, one reconciled view, one scheduled delivery. A broader build with several reports, definitions and variance attribution runs $10,000 to $25,000.
If the same team is also considering automating the underlying process, do that first. Coastline's report cost almost nothing extra because the run that does the work also writes the record of it.
How the build runs
Agree the definitions
What each number means, when it counts, and which system is the source of truth for it. One versioned file.
Pull and reconcile
Sources connected, rows reconciled, differences reported rather than averaged away.
Variance and delivery
Week-on-week comparison, traced variances, drafted commentary, and delivery into the channel people already read.
Hand over the definitions
Editable definitions, an editable schedule, documentation and 30 days of support.
Before and after, in numbers
Coastline Logistics, where the weekly report is appended by the order run. Measured over 90 days against the previous year.
Where we have built this
Related reading
Questions before a pilot
Usually not. The expensive part is exporting, cleaning, reconciling and explaining, and a dashboard does none of those. We build the pipeline and the delivery first, and add a dashboard only if there is a real audience for one.
Yes. If you already run Looker, Power BI, Metabase or similar, the run feeds it a clean reconciled table rather than replacing it. Replacing a BI tool is rarely the cheapest way to fix a reporting problem.
Then the report says so, with both figures and the reason. Hiding a discrepancy behind a chosen source is how a reporting system loses credibility in its first month.
The run drafts it from the variance data and a named owner confirms or corrects it before send. That keeps the blank page away and keeps accountability with a person, which is the only arrangement that survives a bad week.
The summarising is the last five per cent. The value is in defined metrics, reconciled sources, traced variances and a schedule people trust. A summary of unreconciled data is a confident description of the wrong number.
From $4,000 for a pilot covering two or three sources and one scheduled report. A broader programme with several reports, definitions and variance attribution runs $10,000 to $25,000, fixed after the audit week.
Want to see this run on your own reporting?
Bring twenty real examples to the scoping call. The audit week counts the hours, writes the rules and comes back with a fixed price, and you keep the process map whatever you decide.