What a delivery digital twin actually models

A dashboard shows state, a twin answers questions

A dashboard renders what is true right now: burndown, capacity, blocked items. It is a mirror. Useful, and entirely descriptive.

A twin holds the relationships between those things — how capacity, dependencies, readiness and historical throughput interact — well enough that you can change one and see what happens to the others. The difference between the two is whether you can ask a question the builder did not anticipate.

What has to be modelled for that to work

Four things, at minimum. Capacity as it actually behaves, including the unplanned work that reliably consumes it. Dependencies as a graph rather than a list, because the second-order effects are where dates die. Readiness, so the model knows the difference between committed work and work that merely has a ticket. And history, so the forecast is calibrated against what this organisation delivers rather than a benchmark.

Miss any one and the model produces confident nonsense. Capacity without history over-promises. Dependencies without readiness under-estimates rework.

The questions worth asking it

If we move two engineers to the payments epic, what happens to the release date, and what breaks downstream. If this vendor dependency slips a fortnight, which commitments are exposed. If we trim scope here, does the constraint move to QA. Which of the risks currently open have actually changed in the last sprint.

These are the questions asked in every planning meeting already. They are usually answered by whoever has the strongest intuition in the room. A twin does not remove that judgment; it gives it something to argue with.

Where it should run

Delivery, people and governance data is among the most sensitive an organisation holds — it describes what is late, who is overloaded and where the money went. A model over that data should run inside the tenant, without third-party model calls, or the security review will end the conversation before the value is ever tested.