Anyone looking into intelligent document processing (IDP) quickly comes across many technical terms, often in English. This glossary explains the 52 most important ones in plain language.
Metrics
Accuracy: how often a system gets it right. For text recognition, the output is compared with the correct text (ground truth), usually measured as character error rate and word error rate. Accuracy can also be measured for individual tasks such as classification or field extraction.
Automation rate: the share of documents processed completely without manual intervention. The higher it is, the less manual work remains.
Average handling time (AHT): the average time a person needs to process one document or transaction.
Character error rate (CER): the share of wrongly recognised characters compared with the correct text. The lower, the better.
Confidence: a value between 0 and 1 that indicates how certain an AI model is that its result is correct.
F1 score: a value between 0 and 1 that combines precision and recall into one figure.
Manual effort: the work people still have to do by hand, as opposed to automated steps.
Precision: the share of a model’s results that are correct. A measure of quality.
Recall: the share of all correct results that a model actually finds. A measure of completeness.
Return on investment (ROI): the ratio of net profit to the investment over a given period.
Straight-through processing (STP): a document passes through the whole process automatically and reaches the target system without anyone checking it.
Threshold: the confidence value above which a result is accepted automatically. Below it, a person checks.
Time to value (TTV): the time it takes until an investment delivers the expected benefit, in IDP for example the target accuracy.
Word error rate (WER): the share of wrongly recognised words compared with the correct text. The lower, the better.
Technology
Barcode recognition: finds and reads barcodes in documents and images.
Capture: the first stage of document processing: reading incoming documents, sorting them and turning them into usable data. More in our blog post on digital document capture.
Classification: assigning documents to document types based on layout and content. More in our blog post on PDF classification.
Data augmentation: automatically creating slightly modified copies of training data, so that a model learns more variety from fewer examples.
Data extraction: reading specific information from documents and passing it on as data. More in our blog post on PDF data extraction.
Data fields: the individual values extracted from a document, for example in forms or tables.
Data ingestion: collecting and loading data from different sources into a system, in real time or in batches.
Data labelling: adding labels or annotations to raw data so that a model can learn from it.
Data validation: checking whether the right data was extracted and whether it is correct, either by a person or automatically against databases or rules.
Deep learning: a method of machine learning based on artificial neural networks with many layers.
Document splitting: dividing a multi-page file into its individual documents. More in our blog post on PDF splitting.
Entity extraction: finding names, dates, amounts, addresses and similar information in text and assigning them a type. More in our blog post on entity extraction.
Ground truth: the known correct result, used to train and test models.
Human in the loop (HITL): a person checks and corrects results that did not go through automatically.
Intelligent character recognition (ICR): recognition of handwritten characters, for example in form fields.
Key-value pair extraction: reading pairs of label and value from documents, such as “invoice number: 2024-0815”.
Low-code platform: a development environment in which applications are built largely with visual tools and little code.
Model: a program trained on data to recognise certain patterns.
Neural network: a computing model loosely based on the brain and the basis of deep learning.
No code: building applications entirely with visual tools, without writing code.
Optical character recognition (OCR): recognition of printed text and its conversion into machine-readable text. More in our blog post on PDF text recognition.
Redaction: finding and blacking out sensitive information in documents. More in our blog post on redacting PDFs.
Signature detection: finding signatures in documents and extracting them as image sections.
Signature verification: comparing a signature with stored examples to confirm a person’s identity.
Table recognition: recognising tables in documents and reading them out row by row and column by column.
Operation and integration
API (application programming interface): an interface through which two software systems exchange data.
DLL (dynamic link library): a program library that other programs load when they need it.
On-premises: software runs on the organisation’s own servers instead of in the cloud. More in our blog post on on-premises or cloud.
SDK (software development kit): a set of tools and libraries for developing software for a specific platform.
Web client: an application in the browser that communicates with a server.
Fields of application
Automated email processing: reading, sorting and routing incoming emails automatically.
Digitisation: converting paper and other analogue information into digital form.
Document archiving: storing documents long-term so that they can be found again and remain legally secure.
Enterprise content management (ECM): systems for storing, distributing, archiving and finding unstructured content such as scans, emails and reports.
Indexing: labelling documents with attributes such as customer number or date so that they can be found later.
Mailroom automation: digitising incoming paper mail and routing it automatically. More in our blog post on the digital inbox.
Records management: rules for how long documents are kept and when they are deleted, often as part of an ECM.
Scanning service provider: a company that scans and indexes documents on behalf of other organisations.
See it with your documents
In a demo, we show you what these terms mean in practice, with your documents.