Invoices arrive, employees are working, yet payments still regularly need chasing. Finance is waiting for approvers, approvers are looking for missing information, and all the manager sees is a growing number of open cases. Bringing in another person or sending more reminders may seem like an obvious solution. Yet neither necessarily addresses the real problem.
The identification of bottlenecks is one starting point for fine-tuning existing processes. Process mapping shows the steps involved in the work; workflow data shows how it actually progresses. Using an illustrative invoice processing example, we explain how event logs and capacity data can support informed improvement decisions.
1. What Makes a Step a Genuine Bottleneck?
A bottleneck is a process element that limits the throughput of the entire process under a given workload. It does not necessarily involve the most work, nor is it always where the fewest people work. It may be an overloaded role, a decision-maker who is rarely available, missing data, or a rule that only allows cases to move forward at specific times.
A long task queue alone does not prove there is a bottleneck. A backlog after a temporary surge in incoming work may be normal if the team subsequently clears it. A persistent problem is indicated, however, when the queue before the same step regularly grows, open cases age, and the next stage of work is left waiting for tasks to process.
Another important distinction is whether something reduces throughput or primarily increases lead time. Weekly approval can cause long waits even if every invoice is eventually processed. We need visibility into both to make improvements, but they may require different interventions. So first, we should clarify whether we want to close more cases, work to shorter deadlines, or both.
2. What Data Should We Look for in the Workflow Event Log?
The event log records the history of each case: when a status changed, what changed, and which role received the task. In invoice processing, such events might include receipt, completion of data entry, submission for business approval, a request for missing information, or handover to accounting. The final completion date alone is not enough for analysis; intermediate handovers are also needed.
A useful dataset should link at least the following:
- Case ID: to track the events associated with the same invoice from start to finish.
- Event and timestamp: to reconstruct the sequence of steps.
- Task assignment, work start, and completion: if the system records these separately.
- Responsible role or team: to show where work is accumulating.
- Case attributes: such as invoice type, value threshold, organizational unit, and reason for requesting missing information.
- Deadline and returns: to enable analysis of rework and delays.
A particularly important measurement limitation is that the time between two status changes is not the same as active working time. If an invoice is submitted for approval on Monday and approved on Wednesday, this does not reveal how much time was spent working on it. Without data on when work started and was interrupted, only the time spent in that status can be measured reliably.
Before interpreting the data, check for missing events, time zones, and the working calendar. For example, completions recorded retrospectively in a single batch can give a misleading picture of daily performance. Data extracted from the system is therefore not automatically a ready-to-use metric.
3. How to Trace the Path of Backlogged Invoices
Consider an illustrative scenario: invoice verification is followed by business approval, then financial processing. A significant share of open invoices is awaiting approval. At first glance, slow approvers appear to be causing the delay. However, the purpose of the analysis is not to confirm this assumption, but to test alternative explanations.
Examine Queues, Waiting Times, and Returns Together
First, examine each step to see how many cases arrive, how many leave, and how the number of open cases changes. Then look at the age of cases that are still open. If we calculate lead time only for completed invoices, the cases that have been stuck the longest may be excluded from the assessment.
Alongside the average, it is worth including the median and a higher percentile. The median approximates the typical case, while the 90th percentile indicates the time within which 90 percent of the cases examined are completed. This ensures that the impact of lengthy cases is not hidden behind a single average.
Break the Process Down into Comparable Case Groups
Separate standard invoices from exceptions: for example, cases linked to purchase orders, those with missing information, or those requiring multiple approvers. Complete invoices may move quickly, while items arriving without confirmation of delivery or service completion cause the backlog. In this situation, the approval queue is where the problem becomes visible, but its root cause arises earlier.
Finally, walk through a few specific cases with the employees involved. The event sequence shows what happened; the discussion can help explain why. Email exchanges or access issues that are not visible in the system can thus become explanations that can be verified.
4. What Does Capacity Data Add?
To understand a backlog, we need to know how much actual processing capacity is available. This is not the same as team headcount. Part-time work, leave, cover arrangements, time spent on other tasks, and who is authorized to handle which cases all matter.
Use a consistent basis for comparison: examine incoming and processed work for the same period and the same case groups. For invoices of varying complexity, counts alone can be misleading. An increase in the number of straightforward cases completed, for example, does not necessarily mean that the backlog of problematic cases is also shrinking.
- Persistent capacity shortage: more work regularly arrives than the role can handle.
- Uneven workload: incoming volumes spike on certain days or during certain periods.
- Assignment issue: cases accumulate with one employee while others authorized to handle them still have spare capacity.
- Availability constraint: the required decision or backup coverage is only rarely available.
Where possible, focus the analysis on processes and roles rather than individual rankings. Access personal data only for a justified purpose and with appropriate authorization. The goal is to improve operations, not to assess performance without context.
5. From Identification to Measurable Process Improvement
Match the intervention to the identified cause. For example, if confirmation of service delivery is missing, preliminary data validation may be effective. If cases accumulate with a single approver, adjusting backup arrangements or task assignments may help. If an approval step is unnecessary, review the decision rules while retaining the required financial and compliance controls.
In an xFLOWer-based environment, process improvement should start with a review of existing workflow settings and the log data actually available. Check whether statuses, responsibilities, and exception-handling paths accurately reflect operations. If not, improve measurability first; otherwise, the impact of any changes cannot be reliably assessed.
Before piloting an improvement, document:
- The suspected cause: exactly why are cases piling up?
- The change: which rule, handoff, or task assignment will you modify?
- Success metrics: for example, waiting time, age of open cases, and on-time completion.
- Guardrail metrics: the number of errors, returned cases, or control gaps must not increase.
- The evaluation period: ensure comparability with the baseline conditions and workload.
Where possible, change only a few things at a time. If a new team, a new rule, and new automation are introduced simultaneously, it becomes harder to determine what produced the results. Measure the entire process as well: speeding up approval is not a genuine gain if the backlog simply shifts to the queue for accounting.
6. Summary: Verify the Cause Before Taking Action
Identifying bottlenecks is not simply a matter of finding the longest task queue. Event logs show how cases move through the process, capacity data helps explain the workload, and case-by-case checks can reveal underlying obstacles. Together, they provide a basis for process improvement grounded in verifiable relationships rather than assumptions.
Would you like to see where and why work is piling up in your company? Start with a clearly defined process, such as invoice approval, and consult xFLOWer experts to identify the available data, measurement points, and scope of an initial improvement pilot.
This article is part of our comprehensive guide: Business Process Optimization: A Complete Guide to Improving Existing Workflows
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.
