Glossary · The six pillars
Data lineage
Data lineage is whether a number can be traced back to where it came from: which system, which records and which changes produced it.
- Board packARR: €14.2M
- SpreadsheetManual adjustment, undocumented
- Billing exportSubscriptions, monthly
- General ledgerRevenue: doesn’t reproduce the figure
One undocumented step and the board figure can’t be traced.
In plain terms
Lineage asks where did this number come from? It carries 12% of the Truth Score™.
Take a figure in the board pack: annual recurring revenue, say. Can anyone trace it to the system it came from, the records behind it and the steps in between, including the spreadsheet where someone adjusted it? If not, nobody can say whether it is right, and an AI trained on it inherits the same doubt.
The pillar looks at three things: whether each data set has an identified source, whether the transformations between source and report are documented, and whether versions are kept so a past figure can be reproduced.
Lineage is the pillar most exposed in a mid-market company, because the steps tend to live in people’s heads and spreadsheets. It carries less weight than accuracy because a well-kept spreadsheet can still be correct, but it decides whether a board can stand behind a figure, and whether a regulator or buyer can check it.
- See it in a real memoThe sample Truth Score ReadReal engine output on generated test data, not a client
- How we measure itThe method, including the capPillars, weights, bands and what is never guessed
- What happens to your exportsSecurity and data handlingExports only, NDA and data processing agreement first
Questions
Data lineage, answered.
The questions people ask about data lineage, answered in plain English.
Book the callWhy does lineage matter for AI?
Because an AI output is only as defensible as its inputs. If you can’t trace the data a model learnt from, you can’t explain its answers to a board, an auditor or a buyer.
Do we need a lineage tool?
Not necessarily. Many mid-market companies need a short, written map of where their key numbers come from more than a new tool. The roadmap says which.
How do you test lineage from exports?
One practical test: take numbers from the board pack or plan and try to reproduce them from the ledger and system exports. Where they can’t be reproduced, that is a lineage gap.
Related terms
Keep reading.
- The six pillars Data governance Data governance is whether someone owns each record and field, keeps it current and can show what changed: owners, lifecycle, freshness and audit trails. Read the definition
- AI readiness Data quality Data quality is how far data is fit for the job it is about to do: correct, filled in, consistent, owned, connected across systems and traceable to its source. Read the definition
- Method Plan test A plan test reads a business plan, growth strategy, value creation plan or investment thesis, pulls out the AI plays in it and the growth targets that lean on them, and shows play by play whether the data they need is there and can be trusted. Read the definition
- The Truth Score Truth Score™ The Truth Score™ is DataTruth’s score out of 100 for how far a company’s data can be trusted, built from six weighted pillars and placed in one of four bands, A to D. Read the definition
See how your data scores on lineage. Measured, not self-reported.
Exports from two or more systems, matched between them, every red flag checked by hand. A signed memo with your Truth Score and a Leverability read, within 5 working days of complete exports.
Starts with a 30-minute call. Commits you to nothing.