Document Conversion Process: Quality Control Metrics

Our quality control process is composed of severalsteps that have been developed and refined over the last decade to limit oreliminate errors in the conversion process. In addition to our own in-house tools, we utilize industry-leadingsoftware to manage the process from start to finish.  Our software provides batch tracking andworkflow management throughout the entire conversion process.
DocumentPreparation
Documentsare separated by specially coded sheets referred to as separator sheets.  These sheets include a patch code on all 4edges and are generally yellow in color. Additionally, the sheets are typically 8.5” x 14” in size so that theycan be easily removed and reused to keep customer costs down.  Because these sheets are added to a batch forthe purposes of separating documents, they can be counted.  This count can be subsequently used to verifythat the same number of documents are processed at both the Scanning andIndexing stages of the process.DocumentScanning
Documentsare fed into our high speed scanners in sets of up to 500 pages, depending onthe paper quality and page weight.  Thescanner operator ensures that every page is captured and that the image drawnon their monitor is an accurate representation of the physical page.  This process returns a total page countgenerated by the document capture system, and verifies the document count withthose provided during the Preparation stage.
Indexing/Classification
Eachdocument is displayed on the monitor of our data entry staff.  A predefined schema—specific to the documenttype—is used to generate a form to prompt the operator for the index values.  The operator captures each field from thedocument and saves those values to the database using a streamlined data-entryworkflow.  We have the capability to“double key verify” this information, which requires each document pass throughanother data entry station.  On thesecond pass, a different operator repeats this step.  The system responds to the second operatorwhenever an index value does not match what was entered by the firstoperator.  The second operator must thenreview the two entries and make a decision as to the correct value.  The purpose of this “double key verification”process is to minimize or eliminate data entry errors, which produces accuracyrates exceeding 99.99%.
FinalQuality Control ReviewOnce eachstep of the conversion process is complete, every batch is passed to a specialteam whose sole task is to find discrepancies in the process.  Our QC team consists of team members withyears of experience in all three steps. They follow a strict procedure for verifying the accuracy of the processas a whole, as well as ensure that every document has been scanned and classified as accurately as possible.
Countvalidation
Countsfrom each process are compared to ensure that every document was   handled at each step.  Any count discrepancy triggers a full boxreview where every document is reviewed for accuracy and completeness.  Issues at this stage of the quality controlprocess are generally related to errors in the recognition of separator sheetsand can be resolved quickly by a trained QC team member.
RandomSampling
Documentsin a batch/box are randomly pulled and compared with the corresponding imagefiles.  QC team members review every pageof the document to make sure that the image is present, its quality is good,and that the document is complete.  Anadditional check is made to verify that the data entry that was completed forthe document is accurate and complete. Any errors caught during the random sampling triggers a full batch/boxreview.

Three Types of AccuracyMetrics Measured During Conversion
Completeness: everydocument was captured.  This measurementis made to determine whether all documents present in the batch/box arecaptured and represented as an image file in the final deliverable.  Metric target level is 100% of all documentscaptured.
ImageQuality: imagesare accurate representation of the original file.  This is a measurement of the difference betweenthe quality of the original document and its electronic representation.  This measurement is adjusted for concessionsmade in consideration of cost and usability. For example, the best representation would be made using 2400 dpi fullcolor scans that were then run through a gamut of image enhancementalgorithms.  The resulting images howeverwould be unusably large and cost multiples of the acceptable rate to capture.  Capture settings are therefore adjusted topresets for the types of documents being scanned.  Metric target level varies depending on thedocument type.
IndexAccuracy: imagesare classified and indexed accurately for search functionality.  This is a measurement of verification alertevents and final QC findings.  This is themost difficult metric to measure as some index information may be left to interpretation, such as handwrittendata.  We recommend capturing a minimumof 3 fields per document to ensure that each document can be found even incases where index values are subjectively interpreted.  Fields that are “double-key verified” achieve99.99% accuracy.