We spent about four months building a reporting environment covering every channel, with drill-downs, filters, and a great deal of careful work on data consistency.
Usage logs for the following quarter showed eleven sessions, of which seven were by the two people who built it.
Why nobody used it
We asked, which we should have done considerably earlier.
It answered questions nobody was asking. We had built around what the data could show rather than around what decisions people were making.
It required interpretation. Somebody looking at it had to know which metric mattered for their question, and most did not.
It was too comprehensive. Fourteen tabs is a research tool, and the people we built it for had ninety seconds and one question.
And it required them to remember it existed and go and look, which is a behaviour that has to be established rather than assumed.
What people actually wanted
The interviews were clarifying and slightly humbling.
The leadership wanted three numbers, monthly, with a sentence about whether that was good and what we were doing about it.
The channel managers wanted alerts when something changed, not a place to go and check.
Finance wanted a single reconciled spend figure that matched what they had, which we had never provided because it was boring.
Nobody wanted to explore data. That was a thing we wanted and had projected onto everybody else.
What we replaced it with
Considerably less, and it gets used.
A monthly summary of five metrics with written commentary, sent by email rather than hosted anywhere. Delivery rather than availability turned out to be the key difference.
Automated alerts on defined thresholds, sent to the person who would act on them.
One reconciled spend report for finance, in the format they asked for.
And the detailed environment retained for the two of us who genuinely use it, which is a legitimate use and was always the real audience.
The commentary is the product
The finding that changed how I think about reporting generally.
A number without interpretation puts the analytical work on the reader, who lacks the context to do it.
The valuable output is not the metric, it is the sentence explaining what changed, why, and what should happen as a result.
That sentence is expensive to produce, because it requires actually understanding the business, which is why automated dashboards proliferate and useful reporting does not.
It is also the part that gets read, in my experience by a wide margin.
The consistency problem underneath
Worth mentioning because it was the reason the project took four months.
Different systems reported different numbers for the same thing. Sessions did not match between two analytics tools. Spend in the platform did not match the invoice. Conversions were counted differently everywhere.
Resolving those discrepancies was genuinely valuable work and it was invisible, because the output was that numbers agreed, which nobody notices.
If I did it again I would do the reconciliation work and skip the interface entirely, because the reconciliation was the useful part.
What I would advise
Start with the decision, not the data. Ask who will do something differently based on this, and what.
If nobody will act on a metric, do not report it. This eliminates most of what typically goes on a dashboard.
Push rather than pull. Send the report; do not host it and hope.
Write the interpretation, always, and treat it as the deliverable.
And check usage logs after a quarter, which is a slightly painful exercise and the only way to find out whether any of it mattered.
The retention question
One further thing we got wrong that is worth flagging.
We built the environment on top of data with inconsistent retention periods. Some sources held two years, some thirteen months, some ninety days.
Which meant year-on-year comparisons silently broke as data aged out, and nobody noticed until a chart showed a decline that was an artefact.
Establishing retention per source, and warehousing anything you will want to compare against later, is a decision to take at the start rather than to discover eighteen months in.
The one chart that gets looked at
What actually survived from the whole project.
A single chart showing spend and revenue over time, with annotations marking anything significant we did.
It is analytically crude, ignores lag, proves nothing about causation, and it is the only visual anybody has ever asked to see again.
The annotations are the reason. They turn a chart into a narrative, and people engage with narratives about their own business in a way they do not engage with metrics.
If I were rebuilding the whole thing from scratch, I would start there and add nothing until somebody asked for it.
Access, which nobody thinks about
A last practical failure worth naming.
Half the people we built it for could not log in without asking somebody, because access sat behind a licence allocation that had run out.
Nobody reported this. They simply stopped trying, which is what people do when a tool is mildly inconvenient and not essential to their job.
Checking that the intended audience can actually open the thing, before measuring whether they do, is a step so obvious that we skipped it entirely.