Completion-Time Simulation
A single declared SLA hides the truth. Real processes finish in a range, not at a number: most instances land near the middle, some finish fast, and a few run long. The instances that run long are the ones that miss the SLA, trigger escalations, and generate complaints. A static diagram cannot show you that risk. Logic Insights makes it visible with a Monte Carlo simulation of every process.
What the simulation does
For each process, the engine runs thousands of simulated instances over the process graph. On every iteration it:
- Samples each stage duration from a triangular distribution, using an AI-inferred minimum and maximum with the stage SLA as the most likely value.
- Runs a longest-path pass in topological order using those sampled durations, which yields the total completion time for that instance.
- Backtracks the path to record exactly which stages sat on the critical path for that particular run.
Across all iterations it aggregates the results into a forecast.
What you get
Percentile completion times
Instead of one SLA number, you get the shape of the whole distribution:
| Percentile | Meaning |
|---|---|
| P50 | Half of instances finish by this time (the typical case) |
| P75 | Three quarters finish by this time |
| P85 | The common planning threshold for a comfortable commitment |
| P95 | Only one in twenty runs longer than this |
| P99 | The near worst case |
The gap between P50 and P95 is your exposure. A tight distribution is predictable. A wide one means the process is a coin flip under load, even if its average looks fine.
Per-stage bottleneck probability
For every stage, the simulation reports the fraction of runs in which that stage was on the critical path. This is the single most useful output for improvement work: it tells you which stage is most often the reason an instance runs long, ranked by how often it actually matters, not by its average duration. Shortening the stage with the highest bottleneck probability moves the whole distribution.
Probability of meeting the SLA
When an overall SLA is defined, the simulation reports the probability of meeting it. This turns a binary claim ("our SLA is ten days") into a risk statement ("we meet the ten-day SLA about 82 percent of the time"), which is exactly the language a service owner needs to make a commitment they can keep.
Why it is trustworthy
The simulation is seeded. Every process uses the same seed, so the forecast for a given process is stable across runs and comparable over time. Re-analyzing a process after a change produces a forecast you can put side by side with the previous one to prove the change helped.
How to read it
- Wide distribution: high variance. The process is unpredictable under load. Look at the stages with the highest bottleneck probability and the highest SLA variance.
- Low SLA-meeting probability: the target is aggressive relative to reality. Either the SLA needs to change or the critical path needs to shorten.
- One stage dominating bottleneck probability: you have a clear, high-leverage target. This often lines up with a god stage anti-pattern (see Defects, Anti-Patterns & Standards).