All analyses
AN-2026-07Sample analysis · synthetic data

The loudest process in the building is not the most expensive

An engineering consultancy wants to digitalise leave requests. The measurement agrees with the complaint and still not with the investment. The process is slow, but cheap.

In short

Self-service would work. Lead time would fall by 73 %. It still would not pay: 310 requests a year at 18 minutes each cannot carry €11,500 of setup plus €2,600 of annual licence.

How long it takes

4.8working days

In 8 of 10 cases between 2.7 and 10.2 days.

Where it sticks

Genehmigung Teamleitung

In 61 of 100 simulated runs this step was the hold-up.

What it costs

€4,000

Per year. Estimated between €3,500 and €4,500.

What the fix would return

€1,500

Per year — against €11,500 of one-off cost. It does not pay off.

01Starting point

Leave request

Industry
Engineering consultancy
Size
48 Mitarbeitende
Cases per year
310
Fully loaded rate
€62 / h
Simulated runs
120,000
Date
August 2026

The leave request is the in-house exhibit for “nothing here is digital”. Four paper hops, a median of almost five working days to an answer, considerably longer in summer. A vendor offers a self-service module: €11,500 setup, €2,600 licence a year.

Before signing, the sums get done: 310 requests a year, five steps, three roles, a €62 fully loaded rate. Requests arrive seasonally. Almost twice as many arrive before the summer holidays and before Christmas as in February. Modelled and simulated over 120,000 runs.

02The model

Six steps, four roles, one measurable path

The process modelled in FlowVisual. ↯ marks a media break — the point where data is retyped from one system into the next.

Every value in the model, as a table.
StepRoleSystemDuration P10–P90Bottleneck
01Antrag ausfüllenMitarbeitendePapierformular37 min0 %
02Genehmigung TeamleitungBottleneckTeamleitungPapier14 min61 %
03Prüfung ResturlaubPersonalExcel219 min39 %
04Eintrag ZeitwirtschaftPersonalZeiterfassung26 min0 %
05RückmeldungPersonalE-Mail13 min0 %
03The measurement

120,000 runs, one clear answer

Every chart below shows before against after. Values are labelled directly — the colour is a second signal, never the only one.

Abb. 1Probability of each step being the bottleneck in a run. Before against after.
Abb. 2Lead time as a P10–P90 range. The light tick marks the median (P50).
Abb. 3Utilisation per role. Everything right of the red line is structural overload.
Throughput

20

Anträge/Woche

Days over capacity

2 %

Range P10–P90

2.7–10.2working days

04The finding

Genehmigung Teamleitung — 61 %

The complaint is justified. Median lead time is 4.8 working days, the P90 is 10.2. In 61 % of runs the bottleneck is approval by the team lead. The reason sits inside the process itself: in peak season the approver is away more often than usual, because they are on leave too.

The second wait is a rhythm, not a load: HR works through leave requests in batches on Tuesdays and Thursdays. That step is the bottleneck in 39 % of runs.

And now the number this all turns on. One request costs 18 minutes of work, spread across four people. No role sits above 30 % utilisation; on 2 % of working days more is left open than can be cleared.

Of those 18 minutes, roughly 13 are avoidable: the paper form, the spreadsheet check, the retyping, the reply. At 310 requests a year that is around €4,000. That is the entire attributable cost of this process.

05The intervention

The simulated self-service scenario

Instead of buying and hoping, the offer was run through the model, and as generously as the vendor describes it: request in the portal, remaining entitlement calculated by the system, the entry into time tracking handled overnight in a batch, the reply sent automatically. Only the approval stays human.

One request in twelve would still be a case for HR. Entitlement exceeded, special leave, a correction. That remainder is deducted in the counterfactual; without it, the saving would come out too high by exactly that much.

06The re-measurement

Median lead time down 73 %

How long it takes
4.81.3 working days
Throughput
2049
Days over capacity
2 % → 0 %
Saving per year
€1,500

The intervention would work, and clearly so. Median lead time would fall from 4.8 to 1.3 working days. Down 73 %. The P90 would fall from 10.2 to 2.2, handling time from 18 to 5 minutes per request. The bottleneck would stay with the approval, because it is the only human step left; beside it sits a step that costs zero minutes of work and half a day anyway: the overnight transfer into time tracking.

And still. The labour saving would be around €4,000 a year. From that, €2,600 of licence comes off. Roughly €1,500 would remain. Against that would stand €11,500 of setup. Payback would sit at close to eight years.

The recommendation would therefore be: do not buy it, at least not on this reasoning. The arithmetic flips at around 490 requests a year, roughly 75 employees. Below that, the leave request is the loudest process in the building, not the most expensive. And the better question is not “how do we speed it up” but “which of our processes is actually the most expensive”.

Two exceptions belong right next to that. First: if time tracking is being replaced anyway, the module is a surcharge rather than a setup, and the arithmetic changes completely. Second: four days less waiting cannot be written in euros and counts nonetheless. This analysis does not dispute the benefit, it puts a number on it.

The case is in the archive because the fourth step of the method is called “weigh”, not “justify”. Doing nothing is a result too.

07Limits of this analysis

What this analysis cannot tell you

Every measurement has limits. A measurement that hides them is advertising.

  1. 01

    This is a sample analysis. Process, roles, volumes and vendor prices are constructed, not collected at a client.

  2. 02

    The avoidable minutes are measured against an explicit counterfactual (the same request in self-service, step by step), not against zero. A request costs time digitally too.

  3. 03

    Setup and licence costs are assumed vendor figures. A cheaper offer moves the tipping point down; interfaces move it up.

  4. 04

    Soft effects such as satisfaction, predictability and fewer questions in the team are not priced in euros. They may still carry the decision. They just make a different case than the one shown here.

  5. 05

    “Days over capacity” here means: on that share of working days, at least one role ends the day with more work outstanding than it can clear in a day.

  6. 06

    “Throughput” is the ceiling the scarcest role allows, not the actual volume. That is just over 6 requests a week. The jump to 49 does not mean more requests would arrive. It means the summer peak would stop being noticeable.

  7. 07

    P10–P90 is not a worst case. In 10 % of cases it takes longer than the P90 value.

Your process will look different.

This analysis is a sample. Your numbers are not. With FlowVisual you model your own process and get the same evaluation — on your machine, with your values.