Applied Intelligence Observations on the use of RPA and AI in the Public Sector
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
AI – CORE COMPONENTS SENSE COMPREHEND ACT LEARN See Make Connections Physical Support Structured decision making Computer Vision Graph Analysis Robotics Machine learning Read Decision Support Automate Unstructured decision making Natural Language Processing Analyse, Predict, Recommend Robotic process automation Deep learning Hear Audio Processing Copyright © 2021 Accenture All rights reserved. 2
AI-Powered Virtual Workforce What if a virtual workforce could run part of your operations? An AI-powered, virtual workforce, able to emulate many types of human worker activities, can complement the human workforce by: - automating high-volume complex* repetitive tasks, - augmenting decision-making with collective experience and data insights, - scaling new, innovative business services. * Involving semi-structured or unstructured content, interactions, judgement calls Copyright © 2021 Accenture All rights reserved. 4
AI-Powered Bots have an Increasing Number of Human-Like Skills DISCOVER AND READ AND UNDERSTAND AUTOMATE PROCESSES CONTENT Record, edit and replay processes Interpret the content of a by having a robot interact with document, image or video applications’ graphical user Group documents by category interfaces Recognize text and visual Mimic human interactions markings in content Applicable on any system Extract and validate key information ROBOT DIALOGUE INTEGRATE IN ENTERPRISE Interact by using natural language OPERATIONS speech and text Schedule and execute multiple Understand context and ask robots follow-up questions Monitor performance and escalations, provide full auditability Copyright © 2021 Accenture All rights reserved. 5
SVA – AUSTRIAN SOCIAL SECURITY Classification of scanned letters For this client, 500,000 letters are categorized manually each year. The letters are scanned into PDFs, and with OCR the corresponding text is extracted. This text is classified into one of 56 categories. Based on this categorization, further processing is carried out by the responsible employees. • 500,000 LETTERS • AUTOMATED TEXT CLASSIFICATION • AVERAGE ACCURACIES OF 83% Copyright © 2021 Accenture All rights reserved. 6
MAJOR INSURANCE COMPANY Classification of emails with high precision requirements Client required classification of 4M annual incoming customer emails in 3 different languages. Because of far-reaching consequences in case of error, the client required a precision of 98%. The model was trained on 400,000 emails. Language was detected, and OCR was applied on the email’s attachments. This enables fully automatic classification of 4M emails each year, with high accuracy. • 3 DIFFERENT LANGUAGES WITH AUTOMATIC LANGUAGE DETECTION • ACCURACY OF 98%+ • 4 MILLION MESSAGES CAN BE PROCESSED EACH YEAR Copyright © 2021 Accenture All rights reserved. 7
EUROPEAN LAND REGISTRY DEEP LEARNING PROVES IT’S THE SMART WAY TO SEE HOW THE LAND LIES We helped a European land registry build a proof of concept to show how deep learning could help their surveyors transform the laborious process of updating land records. By applying advanced deep learning algorithms to satellite imagery, we were able to train a model capable of alerting surveyors to undocumented changes with over 90% accuracy—in close to real time. In a single pilot study it uncovered over 16 extensions and over 22 structures the authorities knew nothing about—amounting to over €6 million in uncharged land tax. OVER 90% ACCURACY AUTOMATION Uncovered OVER spotting undocumented of annual surveys €6 MILLION in tax revenues changes Copyright © 2021 Accenture All rights reserved. 8
Clinical chart review solution overview Multiple Data Sense Comprehend Act Files & Formats 1 Optical Character 2 Text Analytics & 3 Machine Learning 4 Interactive User Recognition Natural Language Platform Interface JPEG Processing PNG Labs XP PDFs Claims S Auths Converts unstructured Leverages admin & clinical Deduces the correct A human reviews the documents (e.g. PDFs, JPEG, knowledge libraries with 1B answer and continually output, validates the Medical XP Records CCD-A TXT, etc.) to a structured format that computer clinical combinations to identify key words / phrases adapts the algorithm based on patterns of successes results, and creates actions for operations S applications can read such as names, diagnoses, etc. and failures teams if necessary Learn Provides feedback loops to machine learning to enhance the model based on human validation
Taking a Holistic Approach to Augment Employees Identifying right approach: Program Governance 1. Have an organisation-wide strategy for AI. Don’t do it piecemeal. 2. Think of the downstream impacts. Discovery Implementation Run 3. What benefits will be achieved? Automation 4. What processes do I target to achieve maximum benefit? Strategy 5. Should those process be transformed or tweaked before automation? Center of Excellence 6. What are the technical and data protection implications? Change Management 7. What are the optical and change management implications of using AI? Technology Innovation Copyright © 2021 Accenture All rights reserved. 10
Intelligent Technology + Human Ingenuity Scaling Responsible AI AI Virtual Workforce daniel.sheils@accenture.com
You can also read