Data Lineage Tracking: Prove Where Every Number Came From
What is data lineage tracking?
Data lineage tracking is the practice of recording the full path every piece of data takes from its source system to the place it is consumed, through every transformation, join and calculation along the way. Done properly, it lets you point at any figure in any report and answer, with evidence: which source it came from, what happened to it on the journey, when each step last ran, and who changed anything in between.
That is the whole discipline in one sentence: lineage is how you prove where a number came from. Everything else in this guide is about why that proof has become urgent, and how to build it so the answer takes minutes rather than days.
Why did “where did this number come from” become a board question?
For years, lineage was a technical nicety, something data teams wanted and budgets ignored. Two forces changed that.
The first is audit and regulatory pressure. When a figure in a board pack is challenged, in an audit, a regulatory query or a plain disagreement between executives, the organisation has to reconstruct how that figure was produced, sometimes months after the event. Without lineage, that reconstruction is archaeology. With it, the answer is a lookup. We covered the evidence side of this in our guide to compliance monitoring for Power BI, and lineage is the thread that ties the whole audit trail together.
The second force is AI. Copilots and AI assistants answer questions using whatever your estate publishes, and they inherit every ambiguity in it. When five teams calculate revenue five different ways, the AI picks one, confidently, without telling anyone which. The only defence is knowing exactly which dataset feeds what, which calculation is certified, and where every variant lives. AI-first ambitions depend on data-first foundations, and lineage is the data-first discipline that makes the others enforceable.
There is also a quieter, expensive reason: impact analysis. When a source column changes or a system migrates, lineage tells you every downstream report that will break before it breaks. Without it, migrations run on guesswork and the breakage arrives in production, usually in front of the people you least wanted to see it.
What does tracing a number by hand actually involve?
The honest answer: days of skilled people’s time, and a result you still cannot fully evidence.
The pattern our consultants see in Australian mid-market estates is consistent. A senior stakeholder questions a figure. An analyst opens the report and finds the measure. The measure references a dataset, which pulls from a warehouse view, which is fed by a pipeline someone built in 2021, which reads from two source systems, one of which was migrated last year. Each hop means opening another tool, reading someone else’s logic, and asking around for the person who remembers. New executives make the problem visible fast: they ask how many reports exist, what feeds them and how often they refresh, and admin teams disappear for most of a week assembling half an answer.
Some teams script their way around this with API extracts and dependency crawlers. It works until it quietly stops working, platforms change underneath the scripts, the author moves on, and the organisation discovers the gap at exactly the wrong moment. We unpacked the same failure pattern for audit data in our retention guide.
Manual lineage is not impossible. It is unaffordable at the frequency the questions now arrive.
What are the levels of data lineage?
Lineage is not one thing at one depth, and most estates need two layers of it.
Estate-level lineage maps how data flows between systems: source databases into pipelines, pipelines into the warehouse or lakehouse, the warehouse into analytics platforms. This is the territory of Microsoft Purview, which catalogues your entire estate and draws the system-to-system map across Azure, Fabric, Microsoft 365 and beyond. It answers questions like “which systems feed our reporting platform” and “where does customer data travel”. If you run the Microsoft stack, you should be using it, and our Purview explainer covers what it does and where its scope deliberately ends.
Platform-depth lineage operates inside a specific platform, down to the table, view, column and measure. For a Power BI estate this means knowing which database column feeds which semantic model, which measures reference it, which reports render those measures, and what the calculation was on any given date. This is where purpose-built tooling such as Power BI Sentinel earns its place, mapping granular lineage across the estate continuously and pairing it with the change history that turns a lineage map into evidence.
The two layers are complementary, the same breadth-plus-depth pattern that runs through everything we write about governance tooling. A useful test: if the question names two systems, it is an estate-level question. If it names a specific number, it is a platform-depth question, and platform depth is where most board questions land.
How do you implement data lineage tracking?
A sequence that works in practice, without turning into a transformation program:
- Start from the reports that matter. Lineage for everything is a long project; lineage for the twenty reports in executive and regulatory use is a sprint. Work backwards from those.
- Inventory before you map. You cannot trace what you have not found, and estates reliably contain several times more content than anyone expects. Automated discovery beats a survey.
- Switch on automated capture. Manual lineage documentation is out of date the week it is written. Estate-level mapping comes from your cataloguing tool; Power BI depth comes from governance tooling that reads the estate continuously.
- Pair lineage with change history. A lineage map shows where a number comes from today. Audits ask where it came from in March. Capturing changes to measures, models and pipelines is what makes lineage answer historical questions, and it is the same infrastructure that supports recovery when a deployment breaks. The quality half of that loop rules and monitoring, is covered in our data quality management guide.
- Certify the endpoints. Lineage is most valuable when it terminates at certified definitions: one approved calculation for revenue, margin and the other board metrics. Mark the certified path and the seventeen unofficial versions lose their audience.
- Make it the front door for impact analysis. Before any migration, schema change or decommissioning, the lineage map is the first consultation. This is where the investment pays for itself in avoided breakage.
Most of this is achievable inside a quarter for a focused scope, and step 1 is the discipline that keeps it there.
The five-minute test
A working lineage capability passes a simple test: pick any number in last month’s board pack and trace it to source, with evidence of every hop and every change along the way, in five minutes. If that trace takes a meeting, a script and a week, the capability exists on a diagram somewhere rather than in the estate. The organisations that pass this test answer audits from their desks, migrate platforms without surprises, and give their AI tools a foundation worth building on.
Frequently asked questions
What is data lineage, with an example?
Data lineage is the documented path data takes from its source to where it is consumed, showing every system and transformation along the way. For example, a “monthly revenue” figure on a dashboard might trace back through a Power BI measure, to a semantic model, to a warehouse view, to a sales system table, and lineage records every one of those hops so the figure can be verified.
What is the difference between data lineage and data provenance?
The terms overlap and are often used interchangeably. In practice, lineage emphasises the path and transformations data travels through systems, while provenance emphasises origin and custody, where data began and who has handled it. A complete lineage capability covers both: the route and the history.
Does Power BI have built-in data lineage?
Yes, the Power BI service includes a lineage view that shows how dataflows, semantic models and reports connect within a workspace. It is genuinely useful for current-state visibility, and its scope is the present: it shows how things connect today, per workspace, without column and measure-level history. Estate-wide, historical lineage requires cataloguing or governance tooling on top.
Does Microsoft Purview track data lineage?
Yes. Purview maps lineage across your entire data estate at a system level, showing how data moves between sources, pipelines and analytics platforms. It is designed for breadth rather than platform depth, so detailed lineage inside Power BI, down to table, view and measure level, comes from dedicated tooling such as Power BI Sentinel. The two work together.
What is column-level lineage?
Column-level lineage traces an individual column or field from its source through every transformation to the reports that consume it, rather than mapping only system-to-system connections. It is the depth that answers “which reports break if we change this column” and “which source field produced this figure”, which makes it the level that impact analysis and audit responses actually need.
MIP has spent 30 years building and governing data platforms for Australian organisations. If tracing a board number to source currently takes your team days, talk to us, that is a solvable problem with a short payback.

