（非）智能文档处理.pdf

1、(Un)intelligent Document ProcessingIntroductionDisclaimerWhat are my goals here today?What this IsntExplicit vendor comparisonsRecommendationsTelling you what to doWhat this IsThings to think about when approaching IDPReality vs marketingReality vs wishful thinkingData EvolutionStart with DataWhat i

2、s this?Or this?Data EvolutionStart with DataWhat is this?Or this?Make It InformationConvert it into something actionable:“8675309”Or:“631021815”Data EvolutionStart with DataWhat is this?Or this?Make It InformationConvert it into something actionable:“8675309”Or:“631021815”And then MetadataGive it co

3、ntext:Or:Data EvolutionIDP evolves Data into MetadataBy turning non-actionable DATAInto actionable INFORMATIONAnd,finally,with context into METADATAWhat is IDP?AI-powered(isnt everything these days?)Multi-channel captureFull text extractionContent ClassificationField ExtractionIntelligent processing

4、 of fields and contextWhat is IDP?Machine learning of forms through training setsUsing AI to classify content that defies classificationFalling back on“human-in-the-loop”(eyeballs and brain)Improving classification and extraction from interactive feedbackNOTE:Some IDP solutions lack a HITL UX.How Ca

5、n IDP Fail?Poor form designPoor classification rulesOver-extractionOverconfidenceNone of these are product failuresHow Can IDP Fail?Low code/no code solutionsOperations and supportSaaS vs on-premIDP Strengths&WeaknessesExtractionIs only going to keep getting better as technology/AI improvesBut can s

6、till be a victim toPoor form designPoor captureAnti-photocopying watermarksIDP Strengths&WeaknessesValidationWho builds the solution?IDP Strengths&WeaknessesWhat is your goal for your IDP solution?Simple mailroom automationGoal is high accuracy and high confidence extraction of a few fieldsIDP Stren