How aggregation works
Every analytics tool starts with rows: one per event, click, session or order. Nobody can read a table like that, so the tool groups the rows by something — day, channel, landing page, device, city — and applies a function to each group. It counts them, adds them, averages them, or takes the highest value. What appears on screen is never the raw data. It is the result of a grouping decision somebody made, usually without telling you.
The grouping is the part people forget. Sessions by channel and sessions by landing page come from the same rows, yet the two tables will not line up, because one visit can touch several pages while belonging to a single channel. Change the grouping and the same underlying data produces a different-looking answer, with neither version being wrong.
Why aggregation matters
A summary hides its own composition. A flat monthly total can be made of one source collapsing and another climbing, and the headline stays calm while the business underneath it changes shape. Most disagreements about reports are really disagreements about the level of data granularity someone chose, not about whether the numbers are correct.
It also decides which questions you can still ask later. Once figures have been rolled up and the detail discarded, no filter brings the detail back. That is why a business that only ever keeps monthly totals cannot investigate a bad week, and why the choice of what to store matters more than the choice of what to chart.
Where aggregation goes wrong
The commonest fault is averaging something that is already an average. Taking the conversion rates of ten campaigns and averaging them gives every campaign equal weight, so a tiny campaign with a freak result can drag the figure around. Rates have to be rebuilt from the underlying totals — conversions divided by clicks across the whole set — never averaged from other rates.
The second fault is placing figures side by side that were aggregated at different scopes or over different windows. An event-scoped count next to a session-scoped one, or a platform reporting on the day of the click against another reporting on the day of the conversion, will never reconcile. The mismatch looks like a tracking bug and is usually just two grouping rules.
How to act on it
Decide the grouping before you build the report, and write it on the report itself: what a row represents, what window it covers, and which function was applied. A chart with an unlabelled axis invites everyone to assume their own definition.
Then make a habit of breaking every important total apart at least once before you act on it. Split it by channel, by device and by week; if the story survives all three cuts it is real, and if it does not you have found the segment that was actually moving. Keeping the detailed export alongside your reporting and dashboard setup is what makes that possible.
