Reading what happened

Four screens summarise the same clicks and conversions. They differ by the question they are shaped for, and picking the wrong one is most of what makes reporting feel like work.

All four read the figures defined in the figures, exactly.

#The dashboard

One screen, three questions, in the order people ask them.

How is it going — the headline figures, with profit and ROI given the weight they deserve rather than sitting fourth and seventh in a row of equals.

When — the same figures over time, by hour or by day.

Where — four small tables at once: campaign, landing page, offer and traffic source. Four rather than a tab strip, because comparing them is the whole point and tabs would make that three clicks.

Then the newest clicks at the foot, as the way into the log.

One range control drives all of it, so everything on screen describes the same period. That is worth stating because it has not always been true here: this replaced six components that each fetched their own range and disagreed with one another.

#The report builder

One dimension, every figure, over a range. Change the dimension and look again — that is the move the screen is built around, which is why it is one screen rather than a wizard.

The useful sequence is almost always the same: find the countries that pay, then look at devices inside the good ones by filtering to a campaign and regrouping. Making that a four-step flow would make the cheap part expensive.

Hide empty rows is on by default. A grouping like sub4 on a source that never sets it produces one row of everything and a page of nothing; this keeps the page readable. Turn it off when you are checking whether a slot is being populated at all.

Rows come back best-first on profit, capped at 1000.

#Trends

The same figures grouped by when instead of by what. Four groupings, and the split between them is the point:

• Months and Days lay the range out along the calendar. For "is this getting better".
• Weekdays and Hours fold the range into a cycle — every Monday in one bucket, every 13:00 in another. For "are Sundays worth buying" and "should I pause overnight".

The second pair answers a question a calendar chart cannot answer at all, which is why they are not the same screen twice.

The chart and the table are the same rows: the chart for the shape, the table for the number. Neither computes anything of its own, so they cannot disagree.

#Exported reports

Reports are generated on the server and land here as files.

CSV only. A filter offering PDF and Excel would be offering two things that do not exist — this page previously did exactly that, above a list that was always empty because nothing in the product generated a file at all.

The row count is on each file, which is the fastest way to notice an export ran against the wrong range before you open it.

#What none of these will show you

A summary cannot tell you why one click behaved differently from its neighbours. That is what the logs are for, and it is worth going there sooner than feels reasonable — a campaign behaving oddly usually has a specific cause visible on individual rows and invisible in any total.