Confidence scoring for your extractions
Why did Koncile implement a confidence score?
Koncile assigns a confidence score to every extracted data point, such as unit prices, total amounts, identities, or reference numbers, to provide a clear indicator of the reliability of the detected information. This approach is fully aligned with an Intelligent Document Processing strategy, where data quality is just as critical as data extraction.
The confidence score makes it easy to distinguish between data that can be processed automatically and data that requires verification, especially in high-volume or high-impact business contexts. It helps secure automation workflows, reduces the risk of errors, and allows teams to prioritize controls without slowing down integration processes.
What is the confidence score?
The confidence score is an indicator used to assess the reliability level of an extracted data point. It adds an additional layer of control by helping identify which information can be processed automatically and which requires manual verification.
How does the confidence score work?
The Koncile confidence score is calculated using an algorithm embedded in the application. It is assigned to each extracted data point based on the clarity of the source document, its complexity, and the extraction conditions, in order to estimate the reliability of the detected information.
Its goal is to make it easier to identify potentially incorrect data and to prioritize verification when necessary.
Examples of confidence score usage
In high-volume order scenarios, the confidence score helps secure sensitive financial data. For example, a unit price detected at €6.20 with a confidence score of 0.85 should be reviewed, as even a minor error can have a significant impact on the total amount.
The confidence score is also essential for identifying low-quality documents. Poorly scanned or low-resolution images typically result in lower scores, making it easier to quickly flag documents that require manual review without disrupting the overall processing flow.
More broadly, the confidence score applies across all business functions. In finance and accounting, it helps prioritize checks on amounts and taxes. In procurement, it secures the analysis of prices and order lines. In logistics, it facilitates the verification of quantities and references. In human resources, it assesses the reliability of data extracted from administrative documents. In legal and compliance contexts, it highlights high-risk information that requires special attention.
By centralizing these signals into a single indicator, the confidence score helps ensure reliable data usage, regardless of document type or processing volume.
Type : Automation
User : Business
Complexity : Low
Automate document processing at scale
All your questions about the Koncile confidence score
The confidence score takes into account multiple signals derived from both the document and the extraction context to provide a consistent and usable estimate of the confidence level associated with each piece of information. Used as a decision-support indicator, it helps secure automation workflows while maintaining strong control over data quality.
The confidence score is an indicator, not an absolute truth. It helps assess the reliability level of extracted data, but it must always be interpreted in context.
A confidence score of 95 percent does not have the same meaning for every data point. For example, a unit price or financial amount at 95 percent confidence is more critical than a descriptive or textual field, whose wording can vary without impacting business usage.
Binary fields require special attention. For true or false, yes or no, or present or absent fields, an error leaves no room for approximation. The value is either correct or incorrect, which can directly impact business rules or automation workflows.
Overall, the confidence score should be analyzed on a field-by-field basis, taking into account the real business impact of the data rather than treating it as a uniform metric across the entire document.
There is no single threshold that applies to all cases. In general, any data point with a confidence score at or below 0.95 deserves attention. Below 0.85, verification is strongly recommended, especially for data with high business impact.
Yes, document quality directly influences the assigned confidence score. Blurry, poorly scanned, or heavily compressed documents make extraction more complex and often result in lower confidence scores.
Not necessarily. A low confidence score may be related to extraction difficulty, particularly when the document is of poor quality or has a complex structure. However, in most cases, a low score indicates that verification is recommended, as the probability of error is higher.
Yes, each extracted data field includes its own confidence score.
Yes, the confidence score is also applicable to complex or unstructured documents. However, its interpretation strongly depends on the document context and quality, as each case has its own specific characteristics.
The confidence score is designed to guide automation decisions rather than replace human judgment. It can be used to define validation rules, trigger manual reviews, or route data differently depending on risk level, allowing you to automate with confidence while keeping full control over critical data.
.png)
