安永:机器人技术与智能自动化——人类与机器力量的结合(英文版)(32页)
Robotics and intelligent automation Combining the power of human and machine Contents 20 How “Big AI” delivers value 24 Top tips for successful intelligent automation projects 28 Conclusion 29 Next steps/ contacts 2 RPA makes its mark 4 AI: transformational power but greater complexity 10 The relationship between RPA and AI 12 Moving to intelligent automation 16 How “RPA+” delivers value Robotics and intelligent automation: Combining the power of human and machine 1 Financial services companies have been embracing automation for some time now. Robotic process automation (RPA) has already made its mark, but artificial intelligence (AI) is now attracting huge interest as businesses explore the potential to unlock value in the form of improved revenue, customer service, efficiency and risk management. AI is widely seen as the next business disrupter and advancing quickly. Those working in the AI field believe it is the most disruptive technology we will see in the next three to five years. And there’s more to AI than driverless cars. Financial services companies are already seeing how AI tools can improve revenue, efficiency and risk management. However, AI is one part of the future: its real power comes in combination not only with RPA and digitization in general, but also with the input of people. Neither human nor machine alone can outperform human and machine working together. We believe financial services companies can thrive through “intelligent automation” — the intelligent use of multiple tools and automation approaches. This latest paper in our series on robotics and intelligent automation looks beyond the hype to explain what AI means for financial services businesses and how RPA and AI complement each other. We also identify the types of activities and processes that are enabled by different forms of intelligent automation, from processing “margin call” emails to credit risk management, product pricing and fraud detection. Intelligent automation can bring benefits for financial services businesses, but there are also challenges to address in order to unlock those benefits. We highlight the most important of these and suggest how management teams begin exploring the transformative potential of intelligent automation in their own operations. Robotics and intelligent automation (IA) offers multiple benefits for financial services companies — but success depends on understanding the strengths of different tools and automation approaches. RPA makes its mark In the last few years we have seen an explosion in take-up of RPA. Once an obscure tool, every big financial services organization is now running an RPA program. This high interest is reflected in the rapid growth of EY’s specialist RPA team — from a handful of people to over 2,000 delivering RPA projects across 40 countries. 2 Robotics and intelligent automation: Combining the power of human and machine 3 RPA creates numerous opportunities for automation and we are, so far, only scratching the surface of its potential. However, traditional RPA has its limits — it can only process information in digital form e.g., in systems, spreadsheets, web, and can only undertake simple defined decision making e.g., based on a “decision tree”, or more complex when rules based. Combined, these limitations mean that, although extremely powerful, RPA can only be used on certain processes — and more often sub-processes, thereby limiting end-to-end process automation. Traditional RPA is now being combined with AI and other digital automation tools e.g., optical character recognition (OCR), digital forms, workflow, chatbots, human-in-the-loop processing, to help overcome these limitations. This creates the potential for end-to-end process transformation across any process and irrespective of legacy system environments. Traditional RPA is transforming processes but has key limitations when used in isolation. AI: transformational power but greater complexity AI refers to the development of computer systems able to perform tasks normally requiring human intelligence where judgement is applied beyond simple decision trees, such as visual perception, chat and messaging dialog, reading emails, speech recognition, decision making and translation between languages. 4 Robotics and intelligent automation: Combining the power of human and machine 5 AI is real and has the potential to transform business operations today. In the last few years three key components have come together to enable this: ? new AI model algorithms ? move data available digitally ? new cloud computing capabilities and processing power However, the world of AI can be confusing for businesses, even those experienced with RPA. This is partly because AI encompasses many different types of technology with numerous vendors — far more than are evident in the RPA space. As a result, the AI spectrum is huge, and describing all the flavors of AI is beyond the scope of this document. However, AI can often be categorized into distinct types or use cases and implemented using multiple techniques and tools — today, no one tool can deliver all AI-based use cases. AI is increasingly recognized by financial services organizations as a transformative force. Natural Language Processing (NLP) and “Written Chatbots” AI categories Speech Recognition and “Virtual Assistants” Computer Vision and OCR Supervised Learning for prediction and decisions Deep Learning for simple questions with complex answers 6 Robotics and intelligent automation: Combining the power of human and machine 7 Not only is the complexity of potential use cases and corresponding skills an issue, the application of each use case can require significant investment. Take the example of a recent EY project to automate the processing of “margin calls” — broker requests to investors to increase their cash deposits to meet prescribed minimum levels. Today margin call responses come in through entirely unstructured emails, and in their tens of thousands — making this an area ripe for automation. The table on page 8 summarizes the pros and cons of using AI to automate this process, based on our experience. This is a great example of the power of AI (95% processing of emails in an eight-week project is excellent), but also the cost and complexity, and uniqueness of each case. Within just one vendor, there can be multiple different tools, demonstrating specifically: ? The breadth of use cases ? The number of tools ? The skills required, given the number of tools ? The training, testing and technical environments needed ? The delivery operating model This illustrates the complexity of the AI landscape. AI comes in many flavors and is implemented using multiple techniques and tools AI may need other tools and people to be effective Margin analysts retrain models, as necessary Robot records the successful booking Robot emails any exceptions to agents Robot approves the calls Robot inputs into collateral management system ML models compute results and store them into a database for RPA Use NLP to split the email into sentences Retrieve email from email server and extract metadata ML/NLP Robot People (human- in-the-loop) 8 Pros and cons of using AI to automate margin call processing Pros Cons 1. Delivered fully working email parsing and processing in eight weeks 2. Within four weeks had 60% accuracy — sufficient to pilot 3. Within eight weeks had 95% accuracy — sufficient to go live at scale 4. Had to be combined with RPA to enable actual booking of margin call responses in existing systems 5. Significant return on investment (ROI) — payback started after three months and full return in less than 12 months 1. Required 30,000 emails (six months’ worth) in order to get to desired accuracy 2. Still required human intervention — both to process exceptions, and to continue to teach machine learning (ML) model 3. Required natural language processing (NLP) and ML skills (in our case using Python) and programming skills 4. This only worked on a specific type of email, where we knew it related to margin calls. The ML would have to be trained again for any other type of email with similar data volumes e.g., complaints and chasing issues 5. Where data volumes are small, accuracy of ML will be much lower, and hence ROI may be lower as more human intervention is needed 6. By using open-source (free) tools we could achieve appropriate ROI — using commercial tools payback would have been two or more years Therefore, although AI has the potential to transform an organization’s processes, it can be significantly more complex to implement than RPA. So, realizing the promise of intelligent automation lies in finding the right balance of RPA, AI, digital tools and people to maximize return on investment (ROI), while minimizing complexity and risk. Robotics and intelligent automation: Combining the power of human and machine 9 The relationship between RPA and AI RPA and AI are highly complementary solutions — for example, RPA can be thought of as the oxygen that feeds data into AI, and enacts the decisions or insights that AI delivers. 10 Robotics and intelligent automation: Combining the power of human and machine 11 Based on our experience, we believe there are a number of misconceptions about the relationship between RPA and AI. For example, AI isn’t simply replacing RPA and RPA isn’t just “old tech”. In fact, one does not replace the other. They can be used in isolation or together. AI can significantly increase the value of RPA tools, and vice versa. For example, know your customer (KYC) and credit-risk modeling can be supported through AI without RPA — but with RPA, the insights from the modeling can be immediately actioned. Identifying when and how to combine RPA and AI — and other tools — makes for intelligent automation. RPA is “old tech” and AI is replacing it 1 Today RPA and AI are primarily separate, but complementary, technologies RPA and AI are similar technologies in terms of cost, complexity and skills 2 AI is very much an IT/Data Science-based set of technologies, with RPA an order of magnitude simpler and business-skills based Everything can be automated by AI 3 While AI is transformational, allowing increased scale of automation, today the cost and risk would not stack up for all processes Misconceptions about the relationship between RPA and AI RPA and AI are complementary technologies that can be used in isolation or together, depending on specific circumstances. Intelligent automation is the intelligent use of multiple tools. It can span not just RPA, but Digital and AI enablers, human-in-the- loop and “Big AI” concepts. Moving to intelligent automation 12 Type of automation Description Examples RPA A virtual workforce automating highly repetitive tasks, based on defined processes and decision making ? Data entry across multiple systems ? Checking multiple data sources for KYC ? Running reports for multiple countries and collating into a single summary RPA+ RPA combined with add-on capability e.g., RPA combined with adaptors to improve the productivity of RPA Digital adaptors significantly increase the scope of automation when combined with RPA: ? Web or mobile self-service dynamic digital forms e.g., removing need for paper or telephone applications ? Chatbots or voice recognition integrated with RPA to provide real-time self-service e.g., “how much is in my account” against legacy systems ? Employee digital portals for human-in-the-loop work hand-off from RPA to agents e.g., for approvals or highlighting exceptions Where we cannot use digital adaptors, we could look to use AI adaptors for RPA including: ? Next generation optical character recognition (OCR) that incorporates machine learning to read scanned images e.g., application forms ? NLP and machine learning to read emails e.g., margin call processing “Big AI” Gives computers the ability to learn and predict, to make decisions, as well as the ability to mimic human interactions e.g., predict or recommend, virtual assistants. May use advanced analytics and big data, decision engines, machine learning or deep learning algorithms for certain processes Examples include: ? Machine learning to predict churn ? Expert systems to assist with product recommendations e.g., robo advice ? Advanced analytics combined with ML to look for patterns of fraud ? ML for making decisions on loan applications ? Expert systems or ML for making decisions on medical issues for critical illness insurance claims At EY we distinguish between three types of automation: Robotics and intelligent automation: Combining the power of human and machine 13 14 Digitize processes right from the start Emulate procedural manual tasks via front- end interaction Use AI techniques to derive structure from unstructured data Digital enablers RPA AI enablers Passing work from robot to human and back again to optimize use of human skills and experience Identify patterns and insights to drive decisions and new sources of value Human-in-the-loop and process mgmt “Big AI” RPA+ Tools in the intelligent automation toolbox ? Dynamic digital forms for humans to review or action via Business Process Management (BPM) or digital agent portals “Human-in-the-loop” and process mgmt. ? ML convert to defined/structured data ? Digital forms ? Chatbots ? Voice recognition Emulate procedural manual tasks via front-end interaction ? RPA Identify patterns and insights to drive decisions and new sources of value ? Intelligent optical character recognition (IOCR) Digital enablement Robotics AI/adaptors The conversation is intelligent automation, not just robotics: the suite of capabilities to help our clients change the way they work. However, while the potential of AI is enormous, the practical implications for its delivery today mean that it needs to be well targeted and prioritized. To achieve a rapid and significant ROI from intelligent automation, you need a full understanding both of business processes and how they could be transformed, and of the associated benefits for RPA, digital enablers and AI. As one of the world’s premier financial services consultancies, one of the world’s biggest users of RPA, and having invested significantly in combining RPA with digital enablers and AI, we believe EY has unparalleled intelligent automation delivery experience. We hope this paper has given you useful insights into the practical implications of intelligent automation: how you could transform your own organization’s operations while also optimizing the performance and potential of your people. Next steps To discuss how EY can help accelerate the benefits of intelligent automation in financial services, please get in touch. Contacts Dynamically created digital forms within key markets with BPM/ Agent portals, allow RPA to interact with people Human-in-the-loop About EY EY is a global leader in assurance, tax, transaction and advisory services.