INTELLIGENT AUTOMATION: Learn how to harness Artificial Intelligence to boost business & make our world more human - by Pascal Bornet
Date read: 2023-03-05How strongly I recommend it: 8/10
(See my list of 150+ books, for more.)
Go to the Amazon page for details and reviews.
Great book that goes over the benefits and usage of a subset of AI called Intelligent Automation (IA), recommended by someone who is actively implementing IA within companies. Biggest takeaway: the types of IA are more advanced than I originally thought and the best solutions combine multiple IA tools to implement.
Contents:
- CAPABILITIES - VISION
- CAPABILITIES - EXECUTION
- CAPABILITIES - LANGUAGE
- CAPABILITIES - THINKING & LEARNING
- IMPLEMENTING IA
- IMPACTS TO SOCIETY
My Notes
IA, also called Hyperautomation, is a concept leveraging a new generation of software-based automation. It combines methods and technologies to execute business processes automatically on behalf of knowledge workers. This automation is achieved by mimicking the capabilities that knowledge workers use in performing their work activities (e.g., language, vision, execution, and thinking & learning).
86% of global business leaders recently surveyed believe that to stay ahead in their given domains, their organizations must deploy IA in the next five years. Another survey by Gartner found that 42% of CEOs have already begun the process of digital transformation, and 56% reported gains after implementing IA.
According to a Deloitte survey, IA already has an adoption rate of over 50%. This rate is expected to increase to more than 70% in the next two years.
According to the McKinsey Global Institute, the adoption speed of artificial intelligence is already a serious competitive differentiator. In essence, early adopters experience higher profit margins (+10–15% on average compared to laggards), and their free cash flows are expected to accelerate much faster.
Based on our experience, 80% of a customer journey can become digitally touchless and fully omnichannel. As an impact, we have seen companies increase customer satisfaction as measured by the NPS by more than 15 percentage points, reduce the contact center workload by over 50%, and quintuple the “service to sales” ratio (number of up- and cross-sales during a service interaction).
Vision - This capability allows computers to perceive, interpret, and understand elements of the visual world, like environments, objects, signs, or letters. In the context of IA, it mainly enables the processing of documents, images, videos, and biometric information. For example, it helps to automate the processing of invoices or to identify anomalies, such as signs of diseases on medical images.
The overarching technology is called computer vision (CV).
Optical character recognition (OCR) - Character recognition This technology is used to identify and digitize alphabetical or numerical characters presented in images (e.g., a scanned copy of a contract). It avoids this data needing to be typed manually by a human operator on a computer. These digitalized characters can then be used by other digital technologies leveraging, for example, smart workflow platforms, natural language processing, or machine learning.
OCR - The limits of using OCR as a standalone technology to capture and process documents is its inability to manage unstructured documents. This means documents that are not presented using exactly the same standards and layout (e.g., invoices that are different from one vendor to another) won’t be processed accurately by OCR.
Intelligent character recognition (ICR) - This technology unlocks the capacity for IA to manage and process scanned images or documents (e.g., invoices, people’s IDs, contracts, financial information, receipts). Such a platform allows the processing of about 80% of documents to be automated.
Image and video analysis - It includes both static images and videos. Key benefits include processing speed and accuracy, providing the capacity to process an enormous amount of images and videos in a short time. The range of applications is broad, across several industries, including medical image diagnosis, retail store automation, business processes documentation, and autonomous vehicles.
Retail store automation - When applied in the retail industry, image and video analysis are used to support understanding of customer behavior, automate assessment of the level of product inventories, automate check-out, and analyze surveillance videos.
Business processes documentation - The technology leverages computer vision to detect the applications and objects used by computer users. The user just presses a button at the start and at the end of performing a process. The output is an automatically generated flowchart, including screenshots of each step of the process.
Execution (or action) is about “doing”, actually accomplishing step-by-step process tasks in digital environments (e.g., on computers or servers). Such activities include typing text, clicking on buttons, routing information, filling in forms, and authenticating users. Typical automated processes involve logging in and out of systems, compiling data and preparing reports, reconciling data, sending emails, enabling interfaces between systems, and performing spreadsheet analyses.
Smart workflow - A platform that is a prepackaged solution that is configured to support business processes by driving specific data and action flows. It includes data entry and data routing according to a predefined flow. It involves rules, automated steps, manual steps, and decision points.
Low-code platforms - Offer the capacity for business users (also called “citizen developers”) to develop their own programs using a user-friendly visual development environment (e.g., drag-and-drop). Low-code platforms allow anyone in a company to create applications without specific coding skills. For example, they enable any user to create interfaces between systems and workflows, such as forms to collect data, reports, and approval processes.
Robotic process automation (RPA) - This involves a configurable software tool (also called a “software robot”) that uses business rules and sequences of actions. It interacts with the computer the same way a human does and automatically completes processes in any number of different applications. Effectively, any action that a person can do on a computer using a mouse or keyboard can be accomplished by such a “robot”.
RPA helps to produce two types of automation: assisted (also called attended) and unassisted (or unattended) automation.
Assisted automation (also referred to as “robotic desktop automation”, or RDA) runs on people’s desktops and executes actions hand-in-hand with the employee. For example, from our experience, any call center agent needs to open 5 to 8 different applications when serving a client during a call. Opening these applications and looking for the name of the client in each of them takes 10 to 20% of the total time of a call. RDA helps to pre-open these applications at the right screen.
Unassisted automation works in the background and does not typically involve interactions with employees. For example, it is used to manage the client information collected from webforms to feed a workflow system while onboarding clients in a bank.
The leading practice is to use smart workflow and low-code platforms as a foundation of the overall automation platform. RPA is used when IA needs to be integrated with legacy systems or automation of bespoke processes.
For a well-defined, standard, and stable process involving hand-offs between people and systems, it is preferable to use a smart workflow platform. Such platforms offer pre-developed modules. These are ready-to-use automation programs customized by industry and by business function (e.g., onboarding of clients in retail banking).
There are no ready-to-use modules with RPA. Most of the development is bespoke, and all process flows need to be built almost from scratch. The connections also need to be constructed.
As an outcome, based on our experience, the leading practice is to use low-code and smart workflow platforms as a foundation of the overall automation platform. In contrast, RPA is used for any integration of the overall platform with legacy systems or for automation of bespoke processes.
This capability gives machines the ability to read, speak, write, interact, interpret, and derive meaning from human language. In the context of IA, it mainly concerns language interactions with employees, clients, suppliers, and partners through diverse channels, including phone, messaging, and interactions with smart devices. Key functionalities include text translation, information extraction, information summary, information categorization (e.g., spam filters), sentiment analysis, speech-to-text, text-to-speech, predictive text typing, or voice understanding.
The overarching technology is called natural language processing (NLP).
Intelligent chatbots - The most advanced type of chatbot is called a “cognitive agent”. On top of the standard chatbot capabilities described above, cognitive agents include the ability to learn from conversations between people, execute actions (like processing changes in systems), and improve themselves over time. In addition, they can observe people’s reactions using a webcam, recognize sentiments, and adjust their behavior accordingly. Each conversation can be captured, analyzed, and aggregated to deliver real-time insights.
Unstructured information management (UIM) platforms - UIM platforms are used to extract the meaning from large amounts of data and create insights from them. In particular, they help to extract, categorize, and classify data. They transform unstructured data into structured information that is readable and searchable, and they leverage it to create meaningful insights.
In the human resources industry, UIM systems support the search and selection phases of talent recruitment. They allow the analysis of large quantities of resumes and job advertisements in different formats.
Sentiment analysis - Refers to contextual text mining that identifies and extracts subjective information to understand social sentiment. Sentiment analysis involves various models. Some simply focus on polarity (e.g., positive, negative, or neutral). Others detect feelings and emotions (e.g., angry, happy, proud, or sad), and some can even identify intentions (e.g., interested or not interested).
Identification of negative mentions in social media (e.g., alerts) in order to be proactive in answering complaints and prevent a possible crisis. This can be for a product, a brand, or a company.
Speech analytics - Also known as customer engagement analytics or interaction analytics, speech analytics combines the capabilities of UIM platforms and sentiment analysis platforms. It is often specifically designed for call centers. Speech analytics can capture conversations held via phone, email, text, webchat, and social media. The system transcribes all discussions using speech-to-text and turns them into searchable, structured data that machine learning can then use to gain insight into what customers feel or think.
This capability is about creating insight from data through analysis and prediction to support decision-making. It allows employees to be augmented, providing them with insights from data to guide their actions and decisions. It can also trigger automated process activities. For example, the prediction that a client might be fraudulent could trigger a message to a customer representative to investigate the case.
Three main technologies support this capability: big data management, machine learning, and data visualization. Big data management allows the extraction and preparation of data. Data is then fed to algorithms to generate insights (using machine learning) required for actions or decisions made by computers (again using machine learning) or by humans (based on data visualization).
Big data management - Data quality is a crucial prerequisite to ensure the quality of predictions and other insights from data. The best machine learning algorithm can’t produce valuable insight if it uses low-quality data.
Machine learning - This technology uses a set of algorithms that parse data, learn from it, and then apply what has been learned to make informed decisions.
Regression analysis - It is a form of predictive modeling technique that investigates the relationship between a dependent variable (the target) and one or more independent variable(s) (the predictor). This technique is used for forecasting, prediction, and time series modeling, by finding and leveraging the causal effect relationship between the variables. It will continually adjust its forecast as new actual outcomes are injected into the model, making it learn continuously.
Decision trees - They are graphical representations of a decision based on certain conditions. They are useful in a decision-making process, providing a simple representation of the flow of questions to be answered to get to a decision. They also provide an easy way to explain an ultimate decision.
Clustering - It can be defined as the task of identifying subgroups in the data such that data points in the same subgroup (cluster) are as similar as possible and are as different as possible to other clusters. In other words, it is about finding homogeneous subgroups within the data. Some uses of clustering algorithms include network traffic classification, document categorization, gene clustering, or product categorization.
Deep learning - Deep learning algorithms are inspired by how the human brain works. They are more powerful than other machine learning algorithms because they are made of numerous layers, each providing a different interpretation of the data analyzed. Deep learning techniques are used when data features are numerous and there is a lack of domain understanding to identify and understand them.
Open-source platforms (e.g., Scikit Learn, Keras, PyTorch or R-Project) are freely available and may be redistributed and modified. Communities support the code. As these platforms are broadly used, it might be easier to hire talents with the required skills to use them. They are well-suited for smaller initiatives with lower requirements in terms of reliability and availability of the platform (where potential downtimes are less important).
In contrast, commercial platforms are developed and maintained by a single company. Commercial software includes two main categories: traditional platforms (e.g., SAS or Matlab) and cloud-only ones (e.g., AWS, Microsoft Azure, and Google Cloud).
Data visualization - The presentation of information and data in a pictorial or graphical format. It enables decision-makers to grasp difficult concepts and patterns more easily. By using visual elements such as charts, graphs, and maps, data visualization provides an accessible way to see and understand trends, outliers, and patterns in data.
The short answer is that IA acts on top of these foundational systems. The focus of IA applications is process activities that are currently performed by human workers who interact with and operate these foundational systems. These systems are tools. IA is not meant to replace them but use them as any human worker would. IA just helps to add intelligence and automation on top of foundational systems.
It is critical to prioritize the implementation of the IA initiatives which create a sufficient business impact to convince management and employees of the power of IA.
By combining talents and technology levers and targeting end-to-end processes, the organization will create synergies, build economies of scale, and remove potential bottlenecks.
An IA leadership committee that includes top management of the company and people from all key functions taking part in the transformation. It is in charge of overseeing, owning, and sponsoring the achievement of the vision, business case, and roadmap.
The project delivery is performed by an IA center of excellence (CoE), under the supervision of the IA leadership committee. Third, the CoE is headed by the IA CoE leader, who is responsible for supervising the CoE and monitoring the achievement of the IA leadership committee’s vision.
When the first IA applications are delivered into the production environment, successful companies often set up an “automation operation center” (AOC). Distinct from the CoE, it is in charge of maintaining the IA applications and managing the escalation of issues and changes to the application.
An effective IA transformation starts with an IA opportunity assessment: qualifying processes for automation and prioritizing them. The main prioritization criteria generally used include feasibility (is it technically feasible to automate this task?), complexity (how much effort is necessary to automate this task?), and expected benefits (what are the quantitative and qualitative benefits generated from this automation?)
Identifying a compelling collection of IA opportunities in a company is critical to make sure that the most strategic ones are present and prioritized in the roadmap. Based on our experience, it is wise to start implementing the most impactful opportunities to demonstrate the value of IA to management and employees.
Based on our experience, at most companies, more than 50% of the potential transformation value comes from the top 10-20 end-to-end processes.
Be sure to get help from external vendors - Leveraging the support from consulting and delivery partners (they can be the same) is critical for gaining the right momentum in the organization with minimum investment.
Enabler 1: Democratization of IA - Using technologies that require limited skills to design and build IA applications.
Low-code applications provide the opportunity for almost anyone in an organization to automate work activities with only limited training and skills. Low-code applications enable business users to easily connect systems, drag and drop actions, build rules, and relate machine learning applications to process-automation programs. These programs are created through a user-friendly interface that uses tools such as graphical functional blocks that connect together to build an entire program.
No-code technologies are business-driven building blocks that leverage reusable low-code components to build IA applications.
IA learns by doing or being instructed by the user. For example, chatbots learning thru natural language processing.
UiPath Go not only helps developers share best practices, but it also helps organizations to support each other by sharing automation workflows member companies have already built.
Enabler 2: Convergence of technologies - It's expected that multiple IA tools are needed to implement these initiatives. Companies have started creating suite of AI tools, similar to what Microsoft has done with Microsoft Office.
Enabler 3: IA generated by IA - A few consulting or technology companies recently launched solutions to automate this assessment. They involve algorithms that have been trained on hundreds of transformations. As input, the model uses drivers like job descriptions, task descriptions, headcount, and organizational charts. The outcome is an estimation of the expected benefits at the organization level, which is then broken down by technology lever (e.g., machine learning, RPA, and computer vision), by function, and by team.
Process analysis, documentation, and prioritization of opportunities are time-consuming and workload-intensive. There are tools available to document processes and identify opportunities for automation.
Technology vendors have started to create applications that can generate the code of RPA robots directly by using the outcome from process discovery and process mining solutions (discussed earlier in the same subsection). What is so exciting about these applications is that they automatically create and add automation workflows directly into the automation design studio. Users can then further refine the code. Based on our experience, about 60% of the code for most IA projects can be pre-generated, doubling or tripling the speed of implementation.
Enabler 4: Symbiosis of people and IA - Verbal command enabled technology (VCET): dialogue-enabled interaction. For example, in the coming years, we expect these IA applications will be able to be triggered by full sentences, understand plain language, and be able to propose actions. For example, technology could say: “I heard that you haven’t yet received the progress report on this initiative. Should I get you connected via a call with your business partner?”
In the context of IA, sensors, cameras, computers and other technologies can collect data. For example, some organizations have started to use wristbands, cameras, and GPS to collect data about the tasks performed by employees across space and time.
Augmented Reality (AR) and Virtual Reality (VR) are commonly referred to as XR technologies.
The internet of bodies (IoB) connects objects to the human body through devices such as sensors and chipsets that are ingested, implanted, or worn. In other words, the IoB is the capacity for humans to tighten their connection with technology by embedding technology in their bodies.
Reinforcement learning (RL) is a system that learns by doing. RL refers to goal-oriented algorithms that learn how to attain a complex objective (goal) over many steps.
According to the World Economic Forum, machines and algorithms in the workplace are expected to create 133 million new roles by 2022, while only causing 75 million jobs to be displaced.210 This means that the potential number of net new jobs generated by AI could be 58 million over the next few years.
We believe that adaptability and “learning how to learn” will be the most important skills to acquire in the future.
In the near future, traditional subjects at school will be replaced by the 4 Cs: communication, collaboration, creativity, and critical thinking.
We also believe that we need to transition the workforce from “knowledge workers” to “insight workers". Machines can build knowledge (i.e., skills acquired through a learning process). But only humans can generate a real insight (i.e., one that involves gaining an intuitive understanding of something).
Which countries today are the readiest for automation? The Automation Readiness Index (ARI), created by the Intelligence Unit at The Economist, assesses how well 25 countries are prepared for the challenges and opportunities of IA. The top countries on the list are South Korea, Singapore, Germany, Japan, and Canada.
We are convinced that this type of technology, able to ingest and create instant insight from companies’ data, represents a potential future disruption to business consulting. Such an application could consist of a software-as-a-service platform that could plug into any company’s data. It could identify areas of improvement in real time, and automatically build or execute programs to solve these issues. This could mean the end of the current costly armies of consultants spending months in companies, interviewing, observing, and crunching data.