Key Takeaways
- Stagnation is expensive: Many companies are stuck at a dark processing rate of 60-70% because they confuse “reading” with “processing”.
- The semantic shift: True dark processing requires the transition from pure “OCR data grabbing” to AI-supported, contextual document understanding.
- Master data as a lever: Without in-depth, autonomous master data matching (fuzzy matching), too many documents end up in manual exception handling.
- Rethinking ROI: The true value lies not only in saved working time (FTE), but also in the avoidance of scaling costs and drastically reduced throughput times.
- The 90% strategy: By shifting the decision-making logic to a state-of-the-art intelligent document processing platform and consistently minimizing exceptions, touchless rates of over 90% can be achieved.
Introduction: The expensive plateau of automation
It’s an open secret in management and IT departments in the DACH region: the digital transformation in input management is faltering. Investments have been made in expensive OCR software and (supposedly) advanced intelligent document processing (IDP) solutions, the go-live parties have been celebrated, and yet people look at the dashboards with disillusionment. The dark processing rate – the “holy grail of process automation”, in which a document passes from receipt to final posting without human intervention – generally stagnates somewhere between 60 and 70%.
For business analysts and IT decision-makers, this is an untenable situation. A process that requires 30% manual intervention is not automated; it is merely digitally supported. The remaining exceptions eat up ROI calculations, tie up qualified resources and prevent much-needed scalability. Until you break through this glass ceiling, operational costs will remain high and lead times volatile. It’s time for an honest stocktaking and a radical change in strategy to sustainably raise the dark processing rate to over 90 %.
When “readout” is falsely sold as “processing”
Why do so many projects fail at the “70% hurdle”? The main reason lies in a fundamental misunderstanding of the terms. Most of the systems established on the market are excellent “readout machines”. They use OCR and machine learning to find and extract IBANs, invoice amounts and data on invoices. This is necessary, but by no means sufficient for a high dark processing rate.
The problem is not finding the data, but validating and contextualizing it. A system that correctly reads a vendor number from paper, but cannot autonomously decide whether this number matches the order reference at hand, inevitably forces a human to interact and validate it.
The resulting problems of this approach are fatal:
- Explosion of exception costs: Each document that requires manual checking costs between 5 and 15 euros on average in processing time. With high document volumes, the exception queue becomes a bottleneck and cost driver.
- Scaling paradox: If your business volume grows by 20 %, your need for input management clerks also grows by almost 20 %, as the dark processing rate remains constantly low. This is the opposite of economies of scale.
- Latency times: Manual rework interrupts the digital flow. What could be done in seconds takes days, which jeopardizes discounts and strains supplier relationships.
The current solutions from legacy providers try to solve this problem with even more complex sets of rules (RegEx) or client-side scripting. The result is an unmaintainable “spaghetti code” monster that collapses at the slightest layout change by a supplier.
The tech shift: dark processing quota as a target
What has changed? Why is a dark processing rate of over 90% no longer a utopian dream, but a realistic target? The market and technology have undergone a paradigm shift that many established IDP providers have missed. Earlier systems were based on predefined templates or statistical machine learning, which had to be laboriously trained for each document type. Modern IDP architectures, such as those we are driving forward at Parashift, use large language models (LLMs) and advanced computer vision in combination.
This new approach is no longer primarily concerned with where a date is located on the page, but what it means in context. The system understands the semantics of an invoice. It understands the difference between a delivery address and a billing address, even if they have never been seen in this layout before. This deep, pre-trained understanding drastically reduces field extraction errors – the first prerequisite for a high dark processing rate.
The strategy: master data reconciliation as an autonomous decision-making center
The mere error-free extraction of data does not lead to dark processing, as explained above. The decisive step towards a dark processing rate of over 90 % lies in the optimization of master data matching.
The system must be able to autonomously validate the extracted raw data against the leading systems (ERP, CRM) and make final decisions. This is where the wheat is separated from the chaff. Simple 1:1 matching is not enough. If the supplier writes “Müller GmbH” on the invoice, but “Peter Müller GmbH & Co. KG” is stored in the ERP, classic systems fail and throw an exception.
The solution is high-performance, AI-supported fuzzy matching that takes place directly in the IDP platform. The system must be able to weight: If the supplier name, address and bank details correlate 95%, the system must autonomously set the vendor number and forward the document in the dark.
To maximize the dark processing rate, the following strategies must be implemented:
- Autonomous vendor matching: Use of all available data points (name, address, VAT ID no., IBAN, telephone number) for unique identification, even with incomplete master data.
- Purchase order reference validation (PO matching): The IDP system must check the item level against open purchase orders in the ERP. Do the quantity and price match within defined tolerances? If yes: dark processing.
- Tolerance management: Implementation of intelligent commercial rounding rules and cent tolerances directly in the IDP process so that minor deviations do not lead to manual checks.
By moving these logical checks from the backend application (where they often lead to errors late in the process) to the front of the IDP process, we radically minimize exceptions and drive up the dark processing rate.
ROI of a high dark processing rate
Switching to an IDP approach that is consistently optimized for a dark processing rate of over 90% delivers measurable results. Customers who switch from legacy OCR to modern, semantic IDP platforms often report a halving of manual interventions within the first three months – without extensive training.
But the ROI leverage is greater than just the FTE effect in input management. The actual business case results from the following two aspects:
- Avoidance of scaling costs: As your company grows, your back office remains lean. The costs per processed document decrease degressively as the volume increases.
- Drastically reduced throughput times: If these are reduced from days to minutes, liquidity planning improves and cash discounts can be used consistently.
Conclusion
No longer settle for a dark processing rate of 70%. It’s an expensive comfort zone that is technologically outdated. True touchless processing requires the courage to question legacy systems and rely on modern IDP platforms that can semantically understand documents and autonomously synchronize master data.
The difference between mere reading and real processing is the difference between stagnant efficiency and exponential scalability. Set the bar higher. 90% is the new normal.
