When a digital twin is worth building — and when it is not

There is a wide gap between how digital twins are marketed and how they are operated. The marketing shows a rotating building with live data flowing through it. The operation, more often, is a dashboard someone opened enthusiastically for a month and now opens when a consultant asks whether they still use it.

The failure is rarely technical. Integrations work. Sensors report. The failure is that nobody ever specified a decision the twin was supposed to support, so the twin supports none.

Three tests, applied before anyone scopes a platform, will tell you whether a twin is worth building.

First: is there a named person who will act on it? Not a department — a person, with a job that involves making a recurring decision the twin would inform. If you cannot name them, you are building a demonstration. Demonstrations have their uses, but they should be budgeted as marketing rather than as operations.

Second: is there a live data source, and does someone already own it? A twin fed by a manual annual survey is a report with better graphics. The value comes from the gap between how often the data changes and how often you could previously see it change. If the BMS, the meters or the occupancy sensors are already reporting and already owned by an accountable team, the twin has something to be.

Third: is the decision frequent enough to justify the integration cost? A choice made once every five years does not need a live model; it needs a good survey when the time comes. A choice made weekly — plant scheduling, space allocation, fault triage — compounds, and that compounding is what pays for the integration.

Pass all three and the scope tends to define itself, because you now know which data must be live, which can be static, and which was never needed. Fail any of them and the honest recommendation is a well-structured as-built model with clean asset data, which costs a fraction as much and is the foundation you would need for a twin anyway.

The minimum viable twin is smaller than most proposals suggest: one decision, one live feed, one owner. Prove that loop works before extending it. Portfolio-wide rollouts fail for the same reason single-asset twins do, only more expensively.