What data granularity means
Granularity is the level of detail at which data is kept. At its finest, every individual event is stored with its own timestamp and parameters. Coarser, it becomes daily totals per channel. Coarser still, a single figure per month in a spreadsheet. Each step up trades detail for speed, storage and simplicity.
The rule that matters is one-way. You can always roll fine data up into a coarser summary, and you can never break a summary back down. A monthly total cannot tell you which day the spike happened, and a channel total cannot tell you which campaign caused it.
Why granularity matters
Diagnosis needs detail. A month that looks flat can contain a strong first fortnight and a collapse after a site change, and only daily data shows it. Hourly data shows whether ads are spending overnight when nobody is available to answer the phone. Event-level data shows which step of a form people abandon.
Reporting usually needs the opposite. A monthly summary is easier to read and far less prone to over-reaction than a daily chart swinging with the weekday pattern. The real mistake is having only one of the two available when a question arrives.
Where granularity goes wrong
Retention settings are the quiet trap. Analytics tools keep detailed, user-level data for a limited window and then keep only aggregated reports. The default window is often shorter than people assume, and the detail cannot be recovered once it has gone, so the setting deserves a decision rather than a shrug — the entry on data retention covers what is kept and for how long.
Aggregating too early is the second. A dashboard built only on monthly totals feels tidy until somebody asks why a particular week was bad, and no answer is available at any price.
Over-granular reporting is the third and opposite fault. Daily conversion rates for a low-volume business are mostly noise, and reading that noise as signal produces constant, pointless changes that make the account harder to judge.
How to act on it
Store fine and report coarse. Keep the most detailed data you are allowed and able to keep — extend retention where the tool permits it, and export raw events to a warehouse if the business depends on analysis you cannot afford to lose — then summarise for the reports and dashboards people actually read.
Match the reporting period to how quickly the number can move. Spend and clicks can be reviewed weekly. Organic search and content performance need months before a change means anything at all. Where volume is low, lengthen the window rather than reading daily movement as though it were a trend.
And settle this before you need it. Granularity is one of the few settings where the cost of getting it wrong is paid much later, by a question you can no longer answer. It belongs in the measurement plan alongside the events themselves.