Most dashboards describe quantities that sit still: how much is in each region, each team, each stage. Flow analysis asks where things go instead: which route carries the most, who keeps handing work back to whom, which team everything passes through. Five charts do this work, and two questions tell you which one to start with.
The same table of from-to pairs can feed a Sankey, a dependency wheel, or an arc diagram, and each shows it differently, so the choice depends on what you know about the nodes and what you need to read from them. Below, the two questions come first, then the five charts, then a worked example that runs one data set through three of them.
Two Questions, Five Answers
Start with two questions: does the position of the nodes matter, and do you need to see how much moves or what connects to what?
| Does the position of the nodes matter? | How much moves | What connects to what |
|---|---|---|
| Yes, stages or a sequence | Sankey diagram | Arc diagram |
| Yes, places on a map | Flow map | Flow map |
| No, or they have none | Dependency wheel | Network graph |
The position matters when the question depends on where the nodes sit: in stages the flow passes through, in a sequence such as steps or time, or on a map. Countries always have a place on a map, but if the question is only who trades with whom, and how much, the map adds distances that the eye mistakes for volume, and a dependency wheel is often the better start. Stages also mean the flow moves one way. If it loops back, the stages no longer hold, and the first question needs a new answer.
The second question is about what you need to read. All five charts can size their links by volume, but the Sankey and the dependency wheel are built around amounts, and the arc diagram and the network graph around the pattern of links. When the geography matters, a flow map does both. The network graph is also where you land when you are still exploring.
Shortlist: Which Chart Answers Which Question
| Chart Type | Question |
|---|---|
| Sankey diagram | Which paths carry the most, and where does the flow narrow? |
| Flow map | Where does the volume move geographically? |
| Arc diagram | Which nodes connect to which, when their order is fixed? |
| Dependency wheel | Which pairs exchange the most with each other? |
| Network graph | What is the shape of the whole connection structure? |
Five Charts, Five Questions
Sankey Diagram: Which paths carry the most, and where does the flow narrow?
A Sankey diagram draws flows between stages, with link width sized to volume.
Best for a process with several routes through it, where the question is which route dominates. It is the default for a reason: it is the only one of the five with direction, volume, and sequence all built into its layout.
Watch out: Too many nodes and it turns to spaghetti. Best with a clear direction and a few stages.
Cousins: A funnel chart is the linear, single-path case of the same idea.
Flow Map: Where does the volume move geographically?
Mexico’s flow is more than twice China’s, yet its line is the shortest on the map.
A flow map shows lines drawn between locations on a map, weighted by the volume moving along them.
Best for movement between places: shipments, migration, traffic, trade. When the nodes are locations, the map does the layout work for free: an arrangement you invent will rarely place them better than the one people already have in their heads.
Watch out: Long lines look important because they are long. Distance is not volume, and the eye keeps confusing the two.
Cousins: A choropleth map shows what sits in each place; a flow map shows what travels between them. A connector map draws the same kind of lines to show which places are connected; size them by volume and add direction, and it becomes a flow map.
Arc Diagram: Which nodes connect to which, when their order is fixed?
An arc diagram has nodes placed along a single line, with connections drawn as arcs above it.
Best for sequences where position already means something and you also need to see the links: text, code, timelines. It is the chart for when you cannot move the nodes because their order is the data. Nodes with no order of their own can be given one: sort them by group, and clusters show up as nested arcs. The arcs above are also sized by weight, so the diagram shows volume too. But a long arc stands out for its height whatever it carries, which makes the diagram easier to read for what connects than for how much moves.
Watch out: The layout is only as good as the node order. Sort them badly and every arc crosses every other one.
Cousins: A network graph drops the fixed order and lets a layout algorithm find the structure instead.
Dependency Wheel: Which pairs exchange the most with each other?
Color shows direction: each ribbon takes the color of the league the players leave.
A dependency wheel, also known as a chord diagram, has nodes arranged around a circle, with ribbons between them sized to the volume exchanged.
Best for a closed set of players where the traffic between them, rather than through them, is the story. A circle has no beginning, which is exactly right when nothing about your nodes suggests a first or a last.
Watch out: A ribbon has no arrowhead. Unless its color shows where it starts, A to B reads the same as B to A.
Cousins: A Sankey with origins on the left and destinations on the right shows the same volumes with an unambiguous direction, giving up the circular overview.
Network Graph: What is the shape of the whole connection structure?
A network graph has nodes and edges positioned by a layout algorithm rather than by data values.
Best for finding hubs, clusters, and isolated pockets when you do not yet know what you are looking for. It is the only one of the five that imposes no order on the nodes, which is both why it can surprise you and why it is so hard to read precisely.
Watch out: Position comes from the layout, not from any data value. The same graph re-laid out looks like a different network, and neither picture is a measurement.
Cousins: An arc diagram fixes the node order, trading the free layout for readability.
How They Chain: A Worked Example
The interesting thing about this set is that each chart is limited by exactly the assumption that makes it readable. Drop the assumption and you see the next thing.
Take work items moving between teams: intake, then whoever handles it, then an outcome.
One set of handoffs through three charts, and what each one adds.
Start with the Sankey. Left to right, sized by volume. Suppose most items go in through one channel, land with one team, and resolve. A visible slice peels off into a node called reassigned.
That node is an artifact. A Sankey needs a direction and stages, so work that goes back where it came from has nowhere to live, and it ends up in a catch-all node. The chart is not wrong. Its stages run one way, so the loop has to go somewhere. Repeating each team at every step can show it, but after a round trip or two the diagram turns to spaghetti.
Drop the stages and keep the volumes: a dependency wheel of team-to-team handoffs. Suppose most of that reassigned volume turns out to be two teams passing the same work back and forth between each other. On the Sankey that was one gray ribbon into a generic node. On the wheel it is a nameable, fixable relationship between two specific teams.
Now drop the volumes too: a network graph. No stages, no circle, no imposed order. Suppose one team turns out to be connected to almost every other. It is not the largest by volume, and both previous charts size a node by its volume, so it never stood out on either. Nearly every path in the organization passes through it.
Three charts, two assumptions dropped, and each drop revealed something the previous layout had no room for. That is the argument for treating the two questions at the top as a starting point instead of a verdict. The chart that answers your question best is often the one you reach for second.
What Is Not Here
You won’t find a bar chart of volume per node, a flowchart, or a table of from-to pairs.
A bar chart of volume per node describes quantities that sit still, and flow is about what moves between things. A flowchart draws the steps and decisions of a process, not how much actually moves through them. The from-to table is a different case: it is complete, precise, and unreadable past about twenty rows. Every chart above trades some of that precision for a structure you can actually see.
The table is worth dwelling on because it is usually where flow data starts, and it is frequently where it stops. Two hundred rows of origin, destination, and volume contain every fact that a Sankey contains, plus the exact numbers that a Sankey turns into widths. What it does not contain is the shape. Nobody reads two hundred rows and notices that two of the entries form a loop, or that one node appears in a third of them. The chart’s whole value is that it loses the decimals and surfaces the structure, and the table’s whole value is the reverse. Keep both, and know which one you are asking.
What to Do With This
If you have a from-to table sitting in a warehouse somewhere and no chart built on it, the two questions at the top of this page will point you to a first chart in about a minute. Position first, then volume or connections.
Many teams have that table and have never drawn it, because the table looks like it already answers the question.
These five answer how much moves and what connects to what. Why a flow changed is a question for root cause analysis charts. When the process is a schedule with deadlines and dependencies, start with project and resource planning charts.
Top comments (0)