> For the complete documentation index, see [llms.txt](https://doc.duaer.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.duaer.com/build/process-predictions.md).

# Case predictions, delay drivers and what-if: when a case will finish and what slows the process

From the finished cases of the same source, Duaer predicts when an open case will most likely finish and how likely it is to take longer than usual; on the process page it lists the column values that slow cases down and simulates how many days would be saved if they ran like the rest. No model is trained; every figure comes from your own past cases.
## Expected to finish

- When an open case is opened on the employee portal, its timeline shows Expected to finish: the most likely date, a range (25th to 75th percentile) and the chance of taking longer than usual.
- It compares finished cases of the same source that reached the case’s current step and took at least as long from there as the case has already waited. They share as many of the source’s breakdown columns (such as supplier or region) with the case as still leaves 10; the last columns are dropped first.
- Expected finish = when the case reached its step + the median of their remaining time. The usual time is the 80th percentile of the source’s finished cycles.
- The basis (how many cases, which step, which columns matched) is shown with it. With fewer than 10 comparable cases it says too few, with the count; a finished case shows none. When the organization counts working days, so do the figures.

## See the delay drivers and simulate

1. Open the process page of a cases table and, on a published source, click View next to Delay drivers.
2. The list is ranked by the extra days in all, at most 10: a value of a column (such as “Supplier = A”) with at least 10 finished cases whose median cycle is longer than the others’.
3. Click Simulate: if those cases took at most the others’ median, how the source’s median cycle and its share within the usual time would change, and the days saved in all.

## Who sees what

- A case’s prediction goes only to people who see the case. Delay drivers and what-if go to people who may read the project’s process data, as with metrics.
- Every case of the source counts, but only medians, ranges and shares are shown, never another case.
- Reading Cases through [OData](/api/process-signals.md) adds PredictedFinish, PredictedFrom, PredictedTo and RiskScore, under the same rules as the portal.
## Questions

### Does Duaer train a model on my data?

No. Predictions and drivers are worked out each time from the finished cases of the same source; no model is trained or stored.

### Why does my case have no prediction?

Fewer than 10 comparable finished cases, or the case is finished. It shows once the source has more finished cases.

### Does a simulation change my data?

No. It only calculates; no case is changed.

## Related

- [See the process in your data and generate workflows in Duaer](https://doc.duaer.com/build/process.md)
- [Add a process source in Duaer](https://doc.duaer.com/build/process-sources.md)
- [Process boards: built once by an admin, each employee sees their own figures](https://doc.duaer.com/build/process-boards.md)
- [Send process signals to other systems: subscriptions, OData and the process API](https://doc.duaer.com/api/process-signals.md)