You check into a hotel and get a rate per night. A stock trader watches a price update every millisecond. Both are 'data', but captured at widely different resolutions for very different reasons.
That resolution is what's called data granularity, how fine or coarse your data is at the point it gets recorded. This decision is usually made at the beginning of the collection process, and determines what questions you will be able to answer later on.
This post breaks down what granularity means, and why it's worth paying attention to.
What is Granularity?
Data granularity represents the level of detail, or specificity, of a data set. It measures how finely categorical data is divided within a data structure. For example, let's say that you have published a brand new website, and you now want to analyse the visit data. There are various levels of granularity at which you can record this data:
- Time only: Visits per day → e.g. 1,400 visits on July 7th
- Time + device: Visits per day split by device type → e.g. 950 on desktop and 350 on mobile on July 7th
- Time + device + traffic source: Visits per day, by device, by source → 300 visits on desktop by organic search, 450 visits on desktop by paid ads, 100 visits on mobile by organic search, etc. on July 7th.
Each dimension you add makes the data more granular, even if the time window doesn't change. This can increase the size of the dataset as individual records are expanded into further categories.
Why does it matter?
Granularity matters because it sets a limit on what you will be able to learn from your data later on. In the above example, if website visits were only stored as monthly totals, there would be no way to review what happened on a specific day. You would never know whether mobile traffic increased in the days after a targeted ad campaign as that detail was never captured.
Less granular, or coarse, data sets are smaller and easier to manage, but there is a risk that excluded dimensions become valuable for future analysis.
The inverse also has its drawbacks. Fine-grained data makes detailed analysis possible but it comes at a cost. It requires more storage, is slower to query, and needs more filtering before the data becomes useful. This is why the level of granularity needs to be carefully selected.
Since fine data can always be aggregated (e.g. daily data into monthly), but coarse data can never be broken down, it is generally safer to capture a little more granularity than required. You can then aggregate for reporting, and keep the more detailed version as a reserve.
How Computing Power Shapes Granularity
The granularity that analysts can work with has always been tied to what their computing infrastructure can handle. Decades ago, disk space was expensive and queries over large, fine-grained datasets were slow. As a result, businesses defaulted to coarser data: daily or monthly summaries instead of per-event logs. This was all limited by what the systems at the time could actually store and analyse in a reasonable time.
As computing power and storage has become cheaper with cloud storage, hardware advancements, and software optimisation, the ceiling has moved dramatically. It is now the norm to log every website visit, every app click, every choice, and still be able to query it quickly. This has made fine-grained data the expectation in many industries, since the cost of data capture and storage is now often cheaper than the loss of actionable insights.
Key takeaways
- Granularity is the level of detail at which the data is recorded. In practice, this means that any category you can split your data by can add another level of detail to your data set.
- Fine-grained data offers flexibility but costs more. It requires more storage, more processing, and has more categories to sort through.
- Coarse data is cheaper and simpler, but not always better. You can aggregate data up but you can't break those aggregations down into detail that was never captured.
- Computing power has shifted the way we collect data. Cheaper storage and faster processing means that fine-grained data is now the default in many industries.
- Choosing the right granularity is a trade-off. It should match the questions you expect to ask, now and later.
Next time you are setting up a new data collection pipeline, ask yourself: What is the level of detail I might regret not having in 6 months time?