Why Most Digital Twin Projects Fail, and How to Avoid It
When a digital twin project fails, the simulation itself is rarely the problem. The usual causes are organizational: no concrete decision behind the model, a scope that is too big, a 3D picture instead of a comparison of options, months of waiting for perfect data, and no owner once the first project ends. The result is an expensive model that nobody uses for the next decision.
All five causes are visible before the project starts, and all five can be avoided.
What a failed digital twin looks like
Failure is rarely dramatic. Picture, for illustration, a plant that spends a long time building a model of the whole site. The 3D view looks impressive and gets shown to the board and to visitors. Then the next real question comes up: should the new packing line go in hall B or next to the warehouse? The team opens a spreadsheet, because the model cannot answer that quickly and nobody remembers how to update it.
The budget for the model is only part of the cost. When a change goes ahead without testing, the bill arrives later and in pieces: a rack moved twice, a forklift route redesigned after go-live, overtime during a slow ramp-up. No single budget line shows it. Each of those corrections paid for a decision that was tried on the running plant. A model that runs the flow over time, with your cycle times, routes and forklift waits, would have shown the problem before the money was spent.
Cause 1: No decision behind the model
The most common starting goal is "we need a digital twin." That names a tool and leaves the purpose open. Without a decision to support, nobody can say which data matters or when the model is good enough.
A working goal has three parts: the decision, the options and the date. For example: "By the end of March, choose between two layouts for the new assembly area, based on output per shift and forklift load." That one sentence tells the team what to model, what to measure and when to stop.
Write the decision sentence before anyone opens a modeling tool. If nobody can write it, the project is not ready to start.
Cause 2: Scope too big for the first project
The second trap is modeling the whole plant at once: every line, every product variant, every exception, every integration. By the time such a model is ready, the decision it was meant to support has already been made, usually on gut feeling.
A good first scope covers the area where the decision lives, plus whatever shares resources with it. If two lines share forklifts or a buffer, both belong in the model. A paint shop on the other side of the site that never interacts with them can stay out.
A small scope can still carry a serious decision. In one DBR77 case study, a multi-variant production line layout analysis stayed small because the scope was set by one question. We describe the steps and roles in How to Run Your First Simulation Project.
So start with one area and one decision, and extend the model when the next decision needs it.
Cause 3: A picture instead of a comparison
Some projects deliver a beautiful 3D model that cannot answer "which option is better?" The forklifts move, but there are no numbers to compare. The board expected answers about output and cost and got a video.
The opposite mistake is just as common: a model that runs only one scenario, the planned one, on average values. It confirms the plan instead of testing it. Averages hide queues, while a model that runs the whole week with fixed cycle times can show a line that works on paper and struggles in the first busy week (see Averages vs simulation over time).
A useful model compares at least two options, usually today's state and one or two variants, on the same measures: units per shift, waiting time in front of the main stations, buffer levels, forklift or cart utilization. 3D and VR then help people understand the result.
Agree on the options and the measures before the model is built, and plan the 3D view as a way to explain results once they exist.
See what a working model is used for
Each DBR77 Digital Twin use case starts from one question a plant had to answer, from a layout change to warehouse picking. Look at how the scope was set before you set your own.
Cause 4: Waiting for perfect data
Many projects stall at the data stage. The team decides the model first needs live machine data and complete master data. The integration project grows, and the decision window closes.
Most first models run without live data. They need a process sequence, nominal cycle times, changeovers, transport speeds, shift rules and the order mix for a typical week. Most of it sits in ERP, routing sheets and CAD layouts. The rest can be measured with a stopwatch, and nominal values are fine to start. The first results usually surprise the team: the gap between the model and what the floor really produces shows where the real bottleneck is.
Where data is missing, write down an assumption, label it as one, and test whether the result changes when the assumption moves. If the ranking of options holds, the gap does not matter for this decision. If it flips, you know exactly which data to collect. Live machine data, for example from DBR77 IIoT, is the next stage of that learning, once the model is used regularly and you want it to follow the plant.
The practical rule: start from the data you have, label every assumption, and connect live data when there is a clear reason.
Cause 5: Nobody owns the model after the first project
A twin that answered one question well can still fail. The first project ends, the engineer who built the model moves to other work, and months later the plant has new machines, a new shift pattern and a new product. The model no longer matches the floor, so nobody trusts it and nobody uses it.
This slow drift turns a twin into a one-time study. Someone has to own the model: update the base model after each change made on the floor, keep the list of assumptions, and decide when a refresh is needed. Engineering, operations and finance each have a part in that. We cover how to split the roles in Who Should Own the Digital Twin Model?.
Before the first project closes, put a name next to the model and agree on the refresh rule.
A quick check before you start
Use this table at kickoff. Fix any warning sign before modeling begins.
| Cause | Warning sign | What to do instead |
|---|---|---|
| No decision | The goal is "build a digital twin" | Write one decision sentence with options and a date |
| Scope too big | The first model covers the whole site | Model one area plus the resources it shares |
| Picture, not comparison | The deliverable is a 3D walk-through | Agree on options and measures first |
| Waiting for data | Modeling waits for an integration project | Start with ERP, routings, CAD and floor measurements |
| No owner | Nobody is named for the time after the project | Name an owner and a refresh rule at kickoff |
FAQ
What is the most common reason digital twin projects fail?
Starting without a concrete decision. When the goal is "have a twin" instead of "choose between these options by this date", the model has no clear scope, no finish line and no user.
How big should a first digital twin project be?
Big enough to answer one real decision, and no bigger. Model the area where the change happens, plus anything that shares forklifts, buffers or people with it.
Do we need live machine data to avoid failure?
No. Waiting for live data is itself a common cause of failure. Most first models run on ERP data, routings, CAD layouts and floor measurements, with assumptions clearly labeled.
How do we keep a digital twin matching the plant after changes?
Name an owner, keep a list of assumptions, and agree on which plant changes trigger an update of the base model. Options you test run on copies and leave the base model as it is. Without that, the model stops matching the floor and people stop trusting it.
Conclusion
If a twin project is already planned, run the kickoff table above against it this week. The first row decides the most: if nobody can write the decision sentence with options and a date, hold the modeling budget until someone can.
Start with one decision
Come with the decision sentence, or with the question you are still trying to turn into one. We will show how DBR77 Digital Twin would scope a model around that question and which data it needs.
Sources
- DBR77, Case study: Production line design (summary on this site)
- DBR77 Digital Twin, Averages vs simulation over time
- DBR77 Digital Twin, Use cases
Want to test a decision from your plant?
Describe the decision to a DBR77 expert and we will show how DBR77 Digital Twin compares the options on output, waiting time and transport load.
Questions
Have a question about this topic?
Tell us what it is about. We will reply within one business day.
Digital Twin newsletter
Short notes on testing layout and flow decisions on a model, and news from DBR77 Digital Twin. You can unsubscribe at any time.