Data engineering · Quality and governance
Knowing a figure is wrong before anyone reads it
We put checks on the data itself, trace each report back to its sources, name an owner for each set of data and build a catalogue people actually use. Governance becomes part of the daily work of the people who maintain the data, rather than a framework written for an audit.
Where this starts
Everything can be running while the data is wrong
A report can load on time, from a platform that is fully working, and still show a wrong figure. A source may have sent only half its rows, or a code may have quietly changed its meaning. System monitoring cannot see this, because nothing has technically failed. Usually, the first person to notice is a senior manager who thinks a figure looks odd, and by then it has already been shared.
What we usually find
- Errors in figures found by the people reading the reports
- No way to see which reports are affected by a broken source
- Data with no owner, so problems are never resolved
- Definitions that exist only in people’s heads
- A data catalogue built once and never opened again
- Governance meetings that produce minutes but no change
Our position
If governance does not change what happens on an ordinary working day, it is only paperwork. Ownership means a named person is told when their data breaks.
What the work covers
Checks, traceability and ownership
Regulatory controls are covered by governance and compliance, under cybersecurity. This work covers the practical side: whether the data is right, where it came from, and who fixes it.
Quality checks
Every load is checked to confirm the data is up to date, complete, correctly structured, free of duplicates and valid. The checks sit at the points where an error would otherwise spread furthest.
Spotting unusual changes
Unusual movements in key figures are flagged automatically. This catches errors that no rule anticipated, such as a region suddenly dropping to zero.
Traceability
Every report can be traced back to its sources, and every source forward to everything that depends on it. The impact of a problem or a change is known before anyone feels it.
Named owners
Each important set of data has an owner, who is told when it fails and has the authority to fix it. It is a real responsibility, not a title on an organisation chart.
A catalogue people use
Data, definitions and owners can be searched in one place. The catalogue is filled automatically from the platform, so it stays current without anyone maintaining it by hand.
Managing data issues
Data problems are logged, prioritised and resolved like any other fault. We track how often each type recurs, so the underlying causes get fixed.
How quality is enforced
Bad data is stopped before it spreads
Checks that only report problems are ignored within a month. These checks take action.
Automatic stops
If a critical check fails, the reports that depend on that data are not refreshed. They keep showing yesterday’s correct figures, rather than today’s wrong ones.
Alerts to owners
Failures go directly to the owner of the data, with the failing records attached, rather than to a shared channel that nobody watches.
Status on reports
Each report shows how up to date its data is and whether it passed its checks. Readers can see when a figure is provisional.
Scorecards
Quality is measured over time for each set of data and each owner. Improvements are visible, and persistent problems become obvious.
Catalogue updates
Descriptions, traceability and ownership are refreshed from the platform automatically, so the catalogue never describes data that no longer exists.
What you are left with
- Checks at every critical load that data is up to date, complete, correctly structured and valid
- A trail from every source to every report
- A named owner for each important set of data, who is alerted when it fails
- A data catalogue filled automatically from the platform
- Reports that show when their data is provisional
Tell us about the last wrong figure that reached a meeting
Tracing how it got there usually shows exactly where the first check should be.