Data lineage in practice: how to know where a data point came from before it reaches your dashboard
When an indicator drops 8%, someone has to reconstruct the data's path. This article shows how to map origin, transformations and dependencies, and what changes when that trail exists.

In this article
Consider the following scenario, probably familiar. At Monday's meeting, the plant manager points at the dashboard and asks: “why did Line 2's efficiency drop 8%?”. It's a simple question. The answer, almost always, is not.
Before answering anything about the operation, the data team needs to answer questions about the data itself:
- Where did this number shown on the dashboard come from?
- Which system originated the records — MES, supervisory system, shift spreadsheet?
- Which table feeds this chart, and which layer does it live in?
- Which transformations were applied between the source and that table?
- Which pipeline processed it, at what frequency, and did it run successfully?
- Which business rule defines 'efficiency' in this calculation?
- When was this data last updated?
Until these questions have answers, no one knows whether there was an efficiency drop or a data-quality drop. That set of answers is exactly what data lineage organizes.
1. What data lineage is
Data lineage is the tracing of the path a piece of data travels from its origin, through every transformation, to the point of consumption. It answers three questions at any point in the architecture: where this data came from, what was done to it, and who depends on it.
- 01SourceSensor, PLC, MES, ERP, spreadsheet
- 02IngestionCollection, broker, connector, load
- 03TransformationCleaning, joins, calculation, rules
- 04StorageData lake, warehouse, layers
- 05ConsumptionDashboard, report, API, model
In an industrial architecture, the concrete path of a process variable usually looks like this:
Sensor → Gateway → Broker/Pipeline → Data Lake → Transformation → Data Warehouse → Dashboard
Each arrow is a place where the data can be changed, delayed, duplicated or lost. Lineage is the explicit record of those arrows — not the assumption that they exist.
2. Why the origin of data matters
Reliability
Two reports diverge by 3%. With lineage, you can show that one uses a manual shift entry and the other uses the supervisory system's counter. The divergence stops being a mystery and becomes a choice of source.
Troubleshooting
An indicator freezes. Lineage points, within minutes, to the fact that the last load of the source table failed, instead of the team opening six systems to find that out.
Impact analysis
Before renaming a column or changing a tag's unit, you can list every dashboard, extraction and model that depends on it.
Change control
When the definition of 'net production' changes, you can identify every calculation that needs to be reviewed in the same window.
Governance
Every data asset has a declared origin and an identifiable owner, which makes it possible to assign maintenance and fixes.
Audit and compliance
Audits ask for evidence of how a number was produced. Lineage is that evidence, documented instead of reconstructed from memory.
Inconsistency investigation
When a negative value shows up in a total, the trail shows at which transformation the sign was inverted.
Note that all the items above are activities the team already does. The difference is that, without lineage, they're done through manual investigation every time; with lineage, they're queries.
3. Data lineage in practice
Let's trace a concrete indicator: Line 2's efficiency shown on the management dashboard. Read backwards — from consumption to origin — the path looks like this:
| Layer | Asset | What happens here |
|---|---|---|
| Dashboard | Operational Panel / 'Line 2 Efficiency' card | Displays the value aggregated by shift, with the period filter applied at the visualization layer |
| Indicator | 'line_efficiency' metric | Documented formula: productive time x performance x quality, using the plant's definition of scheduled time |
| Analytical table | fact_oee_shift | One row per production line and shift, already with targets and historical reference |
| Transformation | OEE calculation job | Joins entries, stoppages and scrap; applies rounding rules and excludes scheduled stoppages |
| Pipeline | incremental production load | Runs every 15 minutes; logs success, volume and delay |
| Storage | raw layer in the data lake / warehouse | Keeps the original records unchanged, with an ingestion date |
| Production system | plant MES | Origin of order, quantity and stoppage entries |
| Original source | Line counters and stoppage records from the supervisory system | Physical signal collected and associated with the asset |
With this map, the nature of the meeting's question changes. The team no longer investigates “is the dashboard wrong?”, but rather “at which layer did the behavior change?”. If the raw layer kept the expected volume and the calculation job started excluding a type of stoppage that used to count, the problem is a rule issue, not an operational one. If the raw layer's volume dropped, the problem is in collection.
4. Technical lineage and business lineage
Both exist, serve different audiences, and don't replace each other. Technical lineage describes objects:
Table A → View B → Table C → Dashboard D
Business lineage describes meanings:
Equipment temperature → Operating condition → Process efficiency → Management indicator
| Aspect | Technical lineage | Business lineage |
|---|---|---|
| Question it answers | Which objects feed this object? | Which business concept does this number represent? |
| Audience | Data engineering, platform team | Operations, process engineering, management |
| Typical use | Debugging, impact analysis, migration | Interpreting the indicator, aligning definitions |
| How it's obtained | Largely automatable from metadata and queries | Depends on human definition and documented rules |
Having only the technical version produces a correct graph that no one from operations can read: the answer “the number comes from fact_oee_shift” doesn't explain why efficiency dropped. Having only the business version produces a coherent narrative with no verification: no one can confirm whether the described formula is what's actually running. Both together let operations and data teams have the conversation about the same object.
5. What happens when there is no lineage
Consider the real cost of not having it. Monday, 9 AM: the efficiency indicator shows up 8% lower. Operations says the week was normal. Manual investigation begins.
- An analyst opens the dashboard and tries to figure out which query feeds it — the documentation is outdated.
- They find the analytical table, but don't know which job writes it or how often.
- They ask the platform team, who find three jobs with similar names, two of them inherited from a previous project.
- They discover that a stoppage-exclusion rule was adjusted last week, with no record of who asked for it or why.
- No one knows which other dashboards use the same table, so the fix is applied without knowing what else will be affected.
- Two days later, a cost report shows values inconsistent with the corrected dashboard.
None of this sequence is a skill failure. It is the predictable outcome of an architecture where the data's path only exists implicitly, inside the code and people's memory. The cost shows up as hours of engineering consumed by repeated investigation, decisions delayed by distrust, and — most expensive of all — a gradual loss of credibility for the dashboards. Once operations starts keeping its own spreadsheets “to double-check”, the platform investment is already being paid for twice.
6. How to implement a lineage strategy
You don't need to map everything. You need to first map what supports decisions. A roadmap that works well in industrial environments:
1. Map the sources
List source systems with owner, protocol, frequency and criticality: supervisory systems, MES, ERP, laboratory, spreadsheets that still feed official reports.
2. Identify data assets
Catalog the tables, views, topics and files that are actually consumed. Start with those that show up in the most-used dashboards.
3. Map pipelines
Record what moves data between layers, with scheduling, dependencies and failure behavior.
4. Document transformations
Document what each step does in verifiable language — filters, joins, unit conversions, aggregations, rounding.
5. Identify dependencies
For each asset, know who consumes it: dashboards, extractions, APIs, models. Without this there is no impact analysis.
6. Link technical rule and business rule
Every indicator needs a business-language definition tied to the calculation that actually runs.
7. Assign owners
Every source and every indicator needs a named owner — someone who decides on changes and answers for inconsistencies.
8. Monitor changes
Version definitions and schema changes, recording who changed what, when and why. Without a change history, lineage goes stale in weeks.
7. Data lineage checklist
- Can I identify the origin of the main data that supports decisions?
- Do I know which transformations were applied between the source and the indicator?
- Do I know which dashboards and reports depend on a given table?
- Can I predict the impact of a change before applying it?
- Is there a named owner for each source and each critical indicator?
- Is there up-to-date documentation of the indicators' business rules?
- Can I investigate an inconsistency in minutes, not days?
- Are schema and rule changes recorded with author and date?
Conclusion
Data lineage is not technical documentation to satisfy an audit. It is the infrastructure that lets you quickly answer two questions that come up every week: “is this number reliable?” and “what happens if I change this?”. Without it, every answer costs hours of investigation and arrives too late for the decision that prompted it.
The more complex the data architecture, the more important it becomes to know exactly where each piece of information came from and where it is used. It's worth starting with the indicator that generates the most discussion — mapping it end to end usually reveals, at once, both the immediate problem and the architecture's weak points.
See how Athenas organizes your operation's data
A roughly 40-minute demo showing source connectivity, the catalog, the ISA-95 UNS and real-time data distribution.
Keep reading

