Process Optimization KPIs: Which Metrics Indicate Real Improvement?

Unlabeled line, bar and donut graphics suggest why several process metrics must be interpreted together.

Approval has become faster in a case-handling process, yet customers still receive no earlier response. More cases are being closed, while requests for missing information are increasing. Contradictions like these show why relying on a single performance metric is not enough. Real improvement is reflected in the outcome of the entire process: operations become faster, more accurate and more predictable.

Choosing the right process optimization KPI therefore does not begin with reporting. First, we need to clarify the business outcome we expect from fine-tuning an existing workflow. Using an illustrative case-handling example, we walk through how to calculate lead time, first-pass yield and throughput, then explain how to interpret these metrics together.

1. A good KPI answers a specific business question

Process assessment reveals the steps that make up operations. Measurement makes the results of those steps visible. Reducing customer waiting times requires different metrics from those used to reduce errors or a growing case backlog.

  • Lead time: how much time passes between receiving a case and actually completing it?
  • First-pass yield: what proportion of cases reach closure without corrections or being sent back?
  • Throughput: how many cases do we complete within a defined period?
  • Open case backlog: how much work remains in progress at the end of the period?

Together, these provide a useful picture. Greater speed is not necessarily an improvement if it causes more errors. Higher throughput does not necessarily mean more efficient operations either, if it is achieved solely through additional overtime. This is why resource input and case complexity are also worth tracking.

Every KPI should have a definition, data source, time window and owner. We should also specify what decision a deterioration in the metric will trigger. If we cannot say, we may simply be collecting interesting data rather than using a genuine management tool.

A miniature conveyor inspection station includes a sensor, timing dial and operator-controlled diversion for cases needing review.

2. Lead time: measure the entire case journey, not just the work

The following data serves solely to illustrate the calculation; it does not represent customer results or industry benchmarks. In this example, an application goes through review, approval and finalization. The workflow event log contains the following timestamps:

  • Monday 08:00: the application is received.
  • Monday 10:00: the review begins.
  • Monday 10:20: the review ends.
  • Tuesday 09:00–09:20: approval takes place.
  • Wednesday 07:40: finalization begins.
  • Wednesday 08:00: the response is sent and the case is closed.

Lead time = completion timestamp − receipt timestamp. In this case, the lead time is 48 calendar hours. The three work stages take a total of 60 minutes, so the remaining 47 hours do not involve active processing.

However, this difference does not automatically represent 47 hours of avoidable waste. It includes both overnight periods and organizational waiting time. There is no parallel work in this example, and we have recorded every active stage; in more complex processes, simply subtracting active time can be misleading.

What event data is needed?

At a minimum, a case ID, event type, timestamp, process step and status are required. The responsible team and the reason for the pause are also useful for analyzing waiting time. Assigning a task does not mean work has actually started: active working time cannot be reliably calculated solely from the difference between assignment and closure.

Decide in advance whether to measure calendar time or working time. The former often provides a better reference for customer experience, while the latter is more useful for internal capacity planning. If both are used, distinguish them clearly in reports.

A limestone clock sculpture has a continuous outer rim and three small red inlays suggesting active work within elapsed time.

3. First-pass yield: how many cases avoid rework?

Closure alone does not reveal how much correction was needed. First-pass yield, often abbreviated as FPY, shows what proportion of the cases examined move through the defined process without a correction cycle.

Suppose we closed 80 cases during a five-day working week. Of these, 68 required no requests for missing information, error corrections or returns due to errors. The remaining 12 cases required correction at least once.

First-pass yield = cases closed correctly on the first attempt / total cases closed × 100. In this example: 68 / 80 × 100 = 85%.

The unit counted here is the case, not the correction event. If an application was returned three times, it counts as one case that failed to pass the first time in the FPY calculation. The three returns should be recorded separately when measuring rework frequency.

The definition must specify what counts as an error. A mandatory second approval is not rework. Providing a missing mandatory attachment, however, may count as rework when examining the entire case journey. It is advisable to use separate reason codes for customer-side omissions and internal processing errors, as they require different interventions.

In this example, the 85% applies to cases closed during the week in question. It does not mean that 85% of applications received that week were error-free: several may still be in progress. Data on closed cases therefore needs to be supplemented by an examination of open, stalled cases.

4. Throughput: monitor incoming cases as well as closures

Throughput = number of completed cases / measurement period. The 80 cases closed over five working days represent an average throughput of 16 cases/working day. This is actual performance data, not the process's theoretical maximum capacity.

The figure of 16 alone does not tell us whether we are keeping pace with demand. Continuing the example: we had 20 open cases at the start of the week, received 100 new applications during the week and closed 80.

Closing backlog = opening backlog + incoming cases − closed cases. In other words: 20 + 100 − 80 = 40 open cases. We assume there were no deletions, mergers or reopenings; if such events occur, the balance must also be adjusted accordingly.

The team therefore completed substantial work, but the backlog grew. If incoming cases consistently outnumber closures, the backlog will typically put future lead times at risk as well.

Break throughput down by case type as well. Sixteen simple data updates do not represent the same workload as sixteen complex authorization cases. Direct comparisons of case counts provide a meaningful picture of performance only when the case mix is similar.

5. When do KPIs demonstrate real improvement?

Values before and after a change must be measured using the same definitions under comparable conditions. Otherwise, the impact of a new form, a simplified approval procedure or an automated check may be confused with seasonality or changes in staffing.

  • Establish a baseline: use several periods representative of normal operations.
  • Segment into relevant groups: for example, by case type, complexity and organizational unit.
  • Examine the distribution: use the median and an upper percentile alongside the average.
  • Track side effects: rework, overtime, complaints and the open case backlog.

The median provides a picture of the typical case lead time. The 90th percentile, or P90, indicates the time within which approximately 90% of cases are closed. If the average improves but P90 worsens, problems with slower-moving cases may persist.

Watch for measurement bias as well: if simple cases are closed quickly while difficult ones remain open, the average for closed cases may appear more favorable. Tracking the age of open cases and overdue deadlines makes this visible. Improvement is supported not by a single favorable week, but by consistent measurement and explanations of deviations.

Two office workers review a thick unresolved case folder beside trays of open and completed cases.

6. Turn measurement into regular process fine-tuning

When setting up measurement for an xFLOWer-based workflow, it is advisable to define measurement points, status meanings and reasons for rework within the process model itself. This allows KPIs to be calculated from consistently interpreted events rather than retrospective estimates.

The review can follow a simple set of questions: where did results deteriorate, which case group is affected, what is the cause and what change will we test? Each intervention should have an owner and a review date. Assess the success of automation using the same criteria as for any other process change.

In summary: lead time indicates speed, first-pass yield indicates quality, and throughput indicates the volume of completed work. Combined with the open case backlog and resource input, they reveal whether operations have genuinely improved.

Would you like to make improvements in your own case-handling process measurable? Consult xFLOWer experts on where to place measurement points and which KPIs should support your next process fine-tuning decision.

This article is part of our complete guide: Business Process Optimization: A Complete Guide to Improving Existing Workflows

A charcoal technical schematic shows a case-handling process with measurement points, a rework loop, an open-work reservoir and a feedback path.

xFLOWer.ai

Design the process behind this article – for free!

AI reads the processes hidden in this article; open any of them in xFLOWer.ai with one click.

Contact us

Let's talk about your processes

Tell us briefly which process you'd like to automate – we'll get back to you within 1 business day.