Dark processing rate explained & evaluated in a business context

The Dark processing describes the process in which a document is processed fully automatically without the need for human intervention. The dark processing rate (DVQ) in turn measures the proportion of documents that are processed correctly without manual adjustments.

This technical article explains the dark processing rate and evaluates the key figure in a business context.

Calculation of the dark processing rate

Each digitized document consists of several fields that need to be recognized and read by artificial intelligence (AI). The probability of a document being processed without errors depends on how accurately the AI recognizes each individual field.

Example:

• A bank statement has 50 account numbers and 50 amounts, i.e. 100 fields.
• The AI recognizes each field with 99.5 % accuracy.

The probability that all 100 fields are recognized without error is calculated as follows:

0.995 to the power of 100 results in approx. 61 %

This means:

• The document is recognized correctly in 61 out of 100 cases.
• In 39 out of 100 cases, there is at least one error that needs to be corrected manually.

Strategies for improving the dark processing rate

1. increase in AI accuracy

• If the detection accuracy per field increases from 99.5 % to 99.8 %, the rate improves significantly.
• With 99.8 % accuracy, the probability of an error-free document would increase to 83 %.

2. reduction of the fields to be checked per document

• The fewer fields that have to be extracted, the higher the probability that all of them will be recognized without errors.

3. better AI models through more training data

• If the AI knows more examples, recognition becomes more accurate.
• By training the AI with 100 different examples of the account number, the recognition accuracy increases.

Correlation - field accuracy & dark processing rate

Dark processing vs. actual efficiency gains

The dark processing rate only indicates whether a document was processed without manual correction, but leaves open whether the processing was actually efficient or whether errors occurred later. We therefore make a distinction:

• Dark processing measures quantity, not quality → A document can be considered "dark processed" even if it has to be checked manually later.
• No statement about the correction effort → If the AI saves incorrect values that need to be corrected later, this is not taken into account.
• No direct correlation with process optimization → A high rate does not necessarily mean that less time is needed for processing.

Alternative metrics for a real increase in efficiency 

Time saving per document

The time saved per document measures how much working time is saved by Automation is actually saved compared to manual processing. A high dark processing rate does not necessarily lead to high time savings, as incorrect extractions often require manual corrections. The decisive factor is how much time a manual correction costs and whether automation speeds up the entire process. The better the AI works, the less post-processing is required and the greater the actual time saving.

Evaluation of confidence

Confidence evaluation ensures that the system only automatically accepts reliably recognized values, while marking uncertain fields for manual checking. Instead of processing incorrect data unnoticed, the system recognizes low confidence values and displays them for review. This approach prevents incorrect information from entering the process and at the same time optimizes the dark processing rate.

Evaluation of the DVQ in a business context

The evaluation of the dark processing rate (DVQ) plays a key role in a business context, as it reflects the effectiveness of automated processes. This effectiveness can be analyzed by forecasting error probabilities and their impact on the overall process.

Process reliability through probabilities

The metaphor of lamps connected in series proves helpful in the DVQ assessment. Each lamp represents a specific activity within the business process. The entire process only remains robust if all lamps - and therefore all activities - are active without errors.

I. The metaphor of the lamps

In the context of this view, there is a room with three lamps: lamp A, lamp B and lamp C. These lamps each represent a specific activity within a business process. In terms of the probability of the lamps remaining functional over the course of a year, it is assumed that the entire process is only considered successful if each lamp is intact. The failure of a lamp leads to darkness in the room, which means that the process chain is disrupted and cannot be completed.

This metaphor illustrates the need for continuous monitoring and evaluation of the functionality of each individual activity in the process.

II. Probabilities and error rates

To measure reliability, the probabilities of each lamp working over a year are analyzed. The analysis is performed using two different implementations to capture the variation in reliability and potential failure rates.

Implementation 1

• Lamp A
  Error rate of 0.05 (functional probability of 1 - 0.05 = 0.95 or 95 %)
• Lamp B
  Error rate of 0.05 (functional probability of 1 - 0.05 = 0.95 or 95 %)
• Lamp C
  Error rate of 0.2 (functional probability of 1 - 0.2 = 0.8 or 80 %)

Implementation 2

• Lamp A
  Error rate of 0.1 (functional probability of 1 - 0.1 = 0.9 or 90 %)
• Lamp B
  Error rate of 0.1 (functional probability of 1 - 0.1 = 0.9 or 90 %)
• Lamp C
  Error rate of 0.1 (functional probability of 1 - 0.1 = 0.9 or 90 %)

III Calculation of the overall probability

The overall functional probability of the process is determined by multiplying the functional probabilities of the individual lamps in each implementation.

Implementation 1

P(implementation 1)=P(A)×P(B)×P(C)=0.95×0.95×0.8

Calculation:

0,95×0,95=0,9025
0,9025×0,8=0,722=72,2%

Implementation 2

P(implementation 2)=P(A)×P(B)×P(C)=0.9×0.9×0.9

Calculation:

0,9×0,9=0,81
0,81×0,9=0,729=72,9%

IV. Analysis of the results

Overall probability

• Implementation 1 has an overall functional probability of 0.722 or 72.2 %.
• Implementation 2 achieves an overall functional probability of 0.729 or 72.9 %.

Dark processing and its limitations

Although the dark processing rate is almost identical for both implementations, this highlights the limitations of this metric. While the dark processing rate primarily measures the success of a process, it does not capture the full picture. It is unsuitable for identifying deeper problems that occur when a single activity within the process is faulty.

Portfolio of risky actions

The actions of a process can be compared to a portfolio of risky investments. Each "lamp" or action contributes to the overall value of this portfolio. However, there is a risk: if one lamp fails, the remaining lamps may not achieve the expected value.

The dark processing rate merely indicates that the process works smoothly in most cases. However, it does not capture the potential costs and damage caused by necessary manual rework if an action fails. Such rework can take up considerable human and financial resources that are not included in the original success of dark processing.

Costs of reworking

If a lamp fails in the process, additional measures are required to rectify the fault. This rework may include the following:

• Additional working hours
• Material costs
• Customer dissatisfaction

V. Practical implications for companies

The findings from this analysis have a significant impact on process optimization in companies:

• Focus on critical process activities
  Companies should prioritize those process steps with the greatest impact on overall success and continuously monitor their performance in order to identify optimization potential. The analysis shows that a failure at critical points can have systemic effects.
• Implementation of proactive risk management strategies
  Regardless of the high probability of success, companies should establish regular maintenance and audits for critical process steps in order to ensure sustainable risk reduction.
• Long-term strategic orientation
  The results of the study indicate that a focus on continuous process stability is advantageous over short-term efficiency gains or cost-cutting measures.
• Resource allocation based on critical process relevance
  Resources should be allocated in accordance with the strategic importance of the process steps in order to guarantee the operational excellence of all components.

Limits of the dark processing rate

The analysis of the lamp metaphor for process reliability illustrates that while the DVQ provides valuable insights into the success of a process, it does not capture its full complexity. The dark processing ratio provides an overall probability, but neglects the potential costs and risks associated with the failure of individual actions, see V. Practical implications for companies.

A holistic understanding of processes is a basic prerequisite for securing sustainable competitive advantages in a volatile market environment. The dark processing rate (DVQ) acts as an indicator of process reliability, but is not sufficient to fully capture the complete opportunity costs or the transactional expenses for the manual reworking of faulty process steps.

For effective process optimization, companies should understand the DVQ as an initial data point and at the same time implement a detailed risk analysis and a cost-benefit assessment of possible post-processing scenarios in their strategic planning. In this way, companies can ensure that their operational processes offer robust and scalable efficiency even in the context of dynamic market conditions.

Variability of post-processing

Dark processing only captures the probability of success, but not the variability of the "repair effort". In the event of a failure, the steps required for recovery vary considerably depending on the type and complexity of the respective action:

• Repair costs
• Time required
• Complexity of the actions

Conclusion

The dark processing rate shows how many documents the AI recognizes correctly without manual correction. A high rate saves time and costs because less post-processing is required. Better training data, higher recognition accuracy and targeted optimizations increase the dark processing rate.

In order to maximize process reliability and efficiency, companies should use the dark processing rate (DVQ) as an initial indicator, but at the same time systematically include specific risks and potential rework costs in their strategic planning. This allows a holistic assessment of process performance and ensures adaptability in a dynamic market environment.

Reduce post-processing costs through strategic risk evaluation and targeted process improvements. Contact us to increase your operational excellence.