Definition
Data freshness describes how current a dataset is relative to what is expected of it. Each source has an expected arrival pattern: a sales table loaded every night, an analytics export that appears once the day has been processed, a spreadsheet updated by hand each Monday. A dataset is fresh when its latest date matches that pattern and stale when it falls behind.
A freshness check, or freshness test, turns the expectation into a rule. The tutorial on writing data tests for marketing tables writes it as a query that returns a row, and so fails, when a table's latest date is older than expected. The term applies only to data and reports; whether a published article is up to date is a separate content question.
Freshness among the data checks
Freshness is one of a small family of checks that every data flow needs. The guide to documenting marketing integrations lists them for each integration in the register, alongside the question each one asks:
- Freshness: did it run when expected?
- Volume: did the usual amount arrive?
- Completeness: did the identifiers and required fields arrive filled?
- Reconciliation: do the counts match the source?
The checks catch different failures. A table can be fresh and still half empty, which volume catches; it can have the right number of rows with blank identifiers, which completeness catches. Freshness catches the failure that is easiest to miss by eye: nothing new arrived at all. The same guide records each integration's schedule, because the schedule sets what fresh means for that row.
In practice: per-source expectations, blends and caching
Set the expectation from the source. The data tests tutorial sets the freshness expectation per source from how it actually arrives. Its example is the daily analytics export, which appears once the day has been processed rather than at a fixed hour, so a freshness test on tables built from it should allow for that and run after the usual arrival time. An expectation set too tight produces false alarms that people learn to ignore.
Blends inherit the slowest source. The guide to blending data in Looker Studio warns that a figure built on a blend inherits every source's freshness: the slowest, least frequent source sets the age of the whole number.
Caching is a freshness setting too. Reporting tools hold their own copy of query results. The guide to data sources and connectors explains that in Looker Studio how long cached data counts as fresh is set per data source, not per report, and that a partner connector's data is only as fresh as the vendor's last sync, which can differ from the platform itself.
The business intelligence hub places freshness among the tests that stop a broken table from reaching a dashboard before anyone notices.
A climbing centre's bookings fall that was not a fall
Trezzick Climbing Walls reviewed a weekly report every Monday. One week, bookings showed 212 against a usual range of 340 to 380, and the manager began drafting a promotion to win them back. Before it went out, the analyst ran the freshness check the team had added to the booking table that month. The table's latest booking date was the previous Thursday: the nightly load had failed for three nights after the booking system changed its export password. Once the credential was updated and the load re-run, the week showed 361 bookings, inside the usual range. The team set the freshness test to run each morning and alert the analyst whenever the latest date was more than a day behind.
Related terms
Freshness checks usually live in the Data WarehouseA database designed for analysis that stores data loaded from many systems, such as analytics, ad platforms and the CRM, so it can be joined and queried., where loads run on a schedule and a query can compare the latest date with the expected one. CardinalityThe number of distinct values a dimension or field holds; high-cardinality dimensions, such as page paths with IDs, can exceed report row limits. and Data SamplingWhen an analytics tool calculates a report from a subset of the data and scales the result up, rather than processing every event. are two other reasons a report can disagree with the source, and they are worth ruling out alongside freshness: a high-cardinality dimension can roll values into an "(other)" row, and a sampled report estimates from a subset. Neither is about age; freshness is only about whether the newest data has arrived.
Common confusions
- A dashboard with no errors is up to date. A stalled load leaves charts rendering normally; only a freshness check or a careful reader notices that the latest days are missing.
- One expectation fits every source. Sources arrive differently, so each needs its own rule based on how it really behaves.
- Refreshing the report refreshes the data. Clearing a report's cache fetches what the source holds; if the source or the vendor sync is behind, the numbers stay old.
- A fresh table is a correct table. Freshness says new data arrived, not that it arrived whole; volume, completeness and reconciliation checks cover the rest.
- Data freshness is the same as content freshness. The first is about tables and reports; the second is about whether a published page is current.
