Sankey Diagram for Application, Claim and Alert Pipelines
How to chart a decision pipeline as a Sankey: count each item once, give every way out its own node, split automatic from manual decisions and keep each stage balanced. Worked example.
Applications, claims, alerts and tickets all move through a pipeline where each item is decided once. A Sankey diagram shows how many items took each path and where they ended, which a funnel chart cannot do when paths merge or loop back to review.
The example has 400 applications. 250 are decided automatically and 150 go to manual review. Automatic decisions approve 200 and decline 50; manual review approves 90, declines 40 and refers 20 further.
The loan application funnel, insurance claims flow, fraud alert triage and KYC onboarding templates all follow the same pattern. This guide explains the rules that make such a chart honest.
Count each item once
Decide the period and the unit first. An application counted when it arrives and again when it is decided will not balance. Use the date of the first event for every item, and keep items that are still open in a node of their own.
Give every way out its own node
Declined, withdrawn, abandoned, referred and still open are different outcomes and should not be lumped into one. Where several paths end the same way, merge them into one node: a Declined node with arrows from the automatic and the manual route shows how much of each end there.
Do not leave a ribbon ending in nothing. A stage that sends on fewer items than it received is flagged by the editor, and that gap is usually the item you forgot.
Split the routes before the outcomes
Showing the route (automatic, manual, investigation, fast-track) as a stage before the outcome lets the reader compare how routes end. That is the part a single approval rate hides: in the example, 80% of automatic decisions approve, against 60% of manual ones.
- Keep the routes to three or four.
- Use the same outcome nodes for every route.
- Do not draw loops back to an earlier stage; count the item where it ended, and say so in the notes.
What it does not show
A Sankey shows counts or amounts for one period. It does not show how long each item took or why it was decided, and it should not be used to judge individual decisions. If time matters, draw a second chart with days as the unit.
Key points
- 250 of 400 applications (63%) are decided automatically and 150 (37%) go to manual review.
- 80% of automatic decisions are approvals (200 of 250), against 60% of manual ones (90 of 150).
- Manual review passes 20 applications on to another team, which ends their path in this diagram.
- In total 290 are approved, 90 declined and 20 referred.
The example’s data
Invented counts for illustration. Not data about any real organisation.
| From | To | Amount (applications) |
|---|---|---|
| Received | Decided automatically | 250 |
| Received | Manual review | 150 |
| Decided automatically | Approved | 200 |
| Decided automatically | Declined | 50 |
| Manual review | Approved | 90 |
| Manual review | Declined | 40 |
| Manual review | Referred | 20 |
The same data as CSV
source,target,value
Received,Decided automatically,250
Received,Manual review,150
Decided automatically,Approved,200
Decided automatically,Declined,50
Manual review,Approved,90
Manual review,Declined,40
Manual review,Referred,20How it works
Choose the period and the unit
Count each item once, in the period of its first event.
Draw the routes, then the outcomes
One node per route and one per outcome, with a flow for every path that occurred.
Check that every stage balances
What enters a stage must leave it. Open items need their own node.
Frequently asked questions
Why not use a funnel chart?
A funnel shows a single line of stages. A pipeline has routes that merge and outcomes reached by more than one path, which a Sankey shows and a funnel cannot.
Can I include time in the same chart?
Not usefully. Use counts or amounts in the Sankey and make a second chart for time, with the same stages.
Open this pipeline and explore it
Open the example as your own chart and change an amount to see how the routes end.