Data Department, working through a partner in the United States, built a programme delivery simulation for a global networking company. The company was migrating a large telecom carrier's sites from legacy equipment to newer hardware, and earned revenue only when a site's migration was complete. Our simulation showed early that, with current resources, next year's site-migration targets would be missed by a wide margin. That forecast proved correct.
The decision
The networking company had a migration contract with a large telecom carrier. The work was to replace legacy circuit-emulation equipment, which carries older dedicated-line style services over a modern network, with newer and more efficient hardware, site by site.
Two features of the contract shaped the problem:
- Revenue came only on completion. The company was paid when a site's migration was finished, not for work in progress.
- Targets were time-based. The contract set targets for how many sites should be migrated by when.
So the real question was not "how busy are we?" but "how much of today's workload will turn into completed migrations, and when?" A full pipeline can still produce too few finished sites if work stalls or falls out between steps.
What we built
We used the programme's own records, taken from its project-management software, and built a simulation in four steps.
| Step | What we did |
|---|---|
| Map the process | Broke the site-migration process into its individual steps |
| Estimate transitions | Estimated the conditional probability of moving from each step to the next, including fall-outs |
| Estimate durations | Estimated how long sites typically spent in each step |
| Simulate | Ran a Monte Carlo simulation of how the current workload converts into completed migrations, week by week |
What a Monte Carlo simulation is. It runs the same process many times, each time drawing durations and outcomes at random from the estimated distributions. Counting the results across all runs gives a range of likely outcomes, not a single optimistic number.
A deliberately simple method. We did not build an elaborate model of every resource and dependency. Step-to-step probabilities and durations from real project data were enough to answer the question. The value was not sophistication; it was making the gap visible early.
Results
- Finding: with current resources, the programme would miss its site-migration targets for the coming year by a wide margin.
- Outcome: the forecast turned out to be correct.
In the same partnership we also built service value analysis and anomaly detection on service-request, returns and installed-base data for the same company.
Lessons
Model conversion, not activity. When revenue depends on completion, the useful question is how work in progress converts into finished units. Busy teams and full pipelines can hide a shortfall.
Your project data already holds the answer. Project-management records show how work really flows between steps, including what falls out. That is often enough to build a credible forecast.
Simple and early beats elaborate and late. A plain simulation that flags a gap in time is worth more than a perfect model that arrives after the targets are missed.
Not every AI (artificial intelligence) project is machine learning. This was a simulation and operations research problem. Simulation sits alongside optimisation in our optimisation and simulation service; see also telecom and networking.
Frequently asked questions
What did the programme delivery simulation show?
It showed that, with current resources, the migration programme would miss its site-migration targets for the coming year by a wide margin. That forecast proved correct.
What data did the simulation use?
The programme's records from its project-management software, used to estimate step-to-step conditional probabilities, fall-outs and step durations.
Why use a simple method?
Because it answered the question in time. Real step-to-step data and a Monte Carlo simulation were enough to make the gap visible early.
Can this approach work outside telecom?
Yes. Any process where work moves through steps, and value is earned only on completion, can be simulated the same way.