Key Takeaways
Combine AI with workflow tools to automate the whole customer journey.
IA means coordinated automation. Intelligent Automation, or IA, combines software and methods to perform knowledge work with little human intervention. Rather than equating it with a single AI model or robot, the book treats IA as a way to redesign processes around a business outcome. Its aim is not simply to make one task faster, but to connect the steps and systems that deliver a service.
The fraud case shows why. A bank used machine learning to identify more fraudulent transactions, but the added cases worsened the experience for customers and employees. A broader redesign combined machine learning with tools that handled documents, decisions, and routine actions. The bank automated over 80% of the process, solved 70% more fraudulent transactions, and reported more than $100 million in annual savings. The lesson is that a clever model cannot fix a broken journey by itself.
The example captures a distinction often blurred in technology programs: improving a prediction is not the same as improving a service. A model can make one step more accurate while pushing more work onto customers or staff elsewhere. Process thinking makes those transferred costs visible. This is also why end-to-end results need to be measured across the whole experience, not only through the accuracy or speed of the automated component. The bank’s reported gains are persuasive, though a single case cannot establish what every organization should expect.
Think of automation as four linked abilities, not one clever algorithm.
Four capabilities make IA legible. The book groups the work of digital systems into vision, language, execution, and thinking and learning. Vision reads documents or interprets images. Language processes speech and text. Execution carries out digital actions such as filling forms or routing information. Thinking and learning use data to generate predictions or support decisions. Different tools provide these abilities, and a process may need several of them working together.
The combination matters more than the label. A bank onboarding process, for example, can use chatbots to answer questions, document recognition to read identity papers, machine learning to assess risk, and software automation to update systems. In the book’s case, this integrated redesign cut process time by over 90% and cost by over 80%. IA generally sits on top of foundational systems such as ERPs rather than replacing them.
This capability map is useful because it shifts attention from fashionable product names to the work a system must actually perform. It also highlights a practical source of failure: organizations may buy a strong tool for one capability without arranging the data and handoffs needed by the others. The map is a guide, not a strict taxonomy. In real deployments, functions overlap, and a process’s success depends as much on interfaces, data quality, and exception handling as on the sophistication of any individual technology.
Redesign the process before automating it, or scale its waste.
Automate tasks, not job titles. The book cites a McKinsey estimate that about half of work tasks could be automated with technologies then available, while only 5% of entire jobs could be fully automated. Most roles mix automatable tasks with work that requires judgment, interaction, or handling unusual cases. Treating a job as one indivisible unit therefore hides both opportunities and risks.
Fix the route before speeding it up. Process redesign means examining how work moves from beginning to end, removing unnecessary steps, and deciding where people still add value. Lean methods make incremental improvements; a zero-based approach imagines a better process without assuming the existing one must remain. The authors report that, in many companies, the ten to twenty highest-value end-to-end processes account for more than half of the potential transformation value. These deserve deeper analysis than a long list of minor task automations.
This approach aligns with a basic insight from operations research: optimizing one station does not necessarily improve the whole system. A faster step can simply move the bottleneck downstream. The task-level distinction also encourages more careful workforce planning, since roles are bundles of activities rather than single repetitive units. Still, redesign is not neutral. The definition of waste depends on whose time and experience are being counted. A process that looks efficient to management may shift friction onto customers, contractors, or employees who have little say in its design.
Use IA to remove routine work, augment judgment, and abandon waste.
The Triple-A artifact sorts the work. This named framework proposes three choices: automate repetitive activities, augment employees with information or decision support, and abandon practices that add little value. The authors estimate that 42% of the work activities they analyzed could be automated, 32% augmented, and 26% eliminated. These are their assessment figures, not a universal formula for every organization.
Invoice processing makes the shift concrete. One manufacturer received 400,000 invoices a year and employed 29 people to enter them manually. Intelligent document processing automated over 70% of the capture work in about four months. Six employees learned to train the system and check exceptions, 21 moved into other roles, and two joined the transformation team. The point is not that every worker will keep the same job, but that automation should include a deliberate plan for the work and people left behind.
The framework is stronger than a narrow cost-cutting lens because it asks whether an activity should exist at all before deciding whether software can perform it. That question can expose burdensome meetings, duplicate approvals, or data entry that automation alone would preserve. The invoice example also shows why redeployment requires planning and investment, rather than assuming saved time automatically becomes better work. The percentages are best treated as prompts for local assessment: the balance among automation, augmentation, and elimination will vary by task, workplace, and employee needs.
Make customer service continuous, contextual, and available beyond office hours.
Customers value continuity and speed. An omnichannel service lets someone move between phone, chat, email, or staff without having to repeat the story. IA can keep customer information available across those channels, answer routine questions through chatbots, and send complex cases to employees. The authors estimate that chatbots can handle 30% to 50% of common customer interactions, while self-service tools may resolve about half of customer requests.
Vodafone’s example shows the operational effect. After automation, average call time reportedly fell from 10 to 20 minutes to 4 to 8 minutes. Call wrap-up, which took 5 to 10 minutes to record, was automated. Personalization can also protect revenue: Netflix estimated that its recommendation system saved $1 billion a year in revenue otherwise lost to cancellations. These cases show IA supporting both lower service friction and more relevant experiences.
The strongest customer applications remove repetition without forcing every interaction into a bot. Context carried across channels prevents customers from doing the organization’s data-entry work, while automated first responses can make help available when staff are not. Personalization can be useful, but it can also feel intrusive or narrow a customer’s choices if recommendation systems optimize only for engagement or sales. A sound service design therefore treats automation as a way to improve access and resolution, not as an excuse to make human assistance difficult to reach.
Scale through leadership, sound data, and governance, not a string of pilots.
Pilots are easier than transformation. The authors describe IA as a business change involving people, processes, and technology. Their suggested sequence is to launch with a clear management vision and business case, prepare by selecting opportunities and testing a pilot, scale deployment in waves, and manage employee communication and reskilling throughout. A small demonstration may work while the organization still fails to achieve broad benefits.
Create ownership and keep systems reliable. A Center of Excellence coordinates standards, reusable tools, and delivery. An Automation Operation Center maintains programs once they enter production. Both need business and IT participation. Data quality, secure access, clear roles, and monitoring matter because a poor data feed or a changed application can break an automated process. The authors advise beginning with high-impact processes and avoiding long vendor-selection exercises that delay learning and delivery.
This implementation advice reflects a broader pattern in organizational change: technology adoption is not the same as institutional capability. Governance can make successful components reusable, while ongoing maintenance prevents an automation portfolio from becoming a fragile collection of scripts. Yet central oversight can become a bottleneck if it controls every small experiment. The useful balance is consistent standards for risk, security, and production quality, paired with room for teams closest to the work to identify problems and test improvements.
Let employees build automations, but govern what reaches production.
Democratization widens participation. Low-code tools let business users build applications with limited programming skills; no-code tools aim to remove even more technical barriers. Wider participation can surface practical opportunities that a small specialist team might miss and can help employees feel ownership of change. The authors suggest communities, training, incentives, and reusable libraries to support this spread.
Access still needs guardrails. Their proposed model gives employees room to create, asks experienced champions to review their work, and reserves production approval for the Center of Excellence. IA can also help create IA: process discovery records how people work, process mining analyzes system logs, and automated code generation or AutoML can speed development. The authors report that discovery tools can make implementations three to five times faster, though such performance claims depend on the tool and setting.
Participation is more than a way to increase output. It can change the social meaning of automation from something done to workers into something they help shape. The safeguard is essential, because a convenient personal automation may create security, compliance, or maintenance problems for everyone else. Democratization therefore works best as distributed problem-finding with shared standards, not a free-for-all. It also requires honest incentives: if employees fear that revealing an automation will cost them their jobs, they have a rational reason to hide it.
Use automation to strengthen public services, not merely to shrink budgets.
The social case reaches beyond efficiency. The book argues that IA could help detect disease earlier, reduce administrative errors, extend services to places with too few professionals, and keep operations running during crises. For example, a system trained on records from 700,000 patients reportedly predicted acute kidney injury up to two days before standard detection. Zipline’s drone service had made 13,000 medical deliveries by 2019, including blood and other essential supplies to remote areas.
Treat the biggest totals as projections. The authors estimate that IA might avert more than 10 million early deaths and reduce roughly $10 trillion in annual losses from fraud, errors, accidents, and work-related illness. They describe these as long-term potential, not immediate or guaranteed outcomes. Better diagnosis and fewer clerical mistakes are credible aims, but any deployment must account for clinical oversight, unequal access, privacy, and what happens when a system is wrong.
The public-benefit argument is most compelling when tied to a specific bottleneck, such as an early warning that gives clinicians time to act or a delivery route that reaches an isolated community. Broad estimates can motivate attention, but they combine different problems and assumptions, so they should not be mistaken for measured savings already achieved. High-stakes systems also need a clear path for human review and appeal. Reliability is not only a technical property; it includes whether people can understand and contest consequential decisions.
Plan for job growth and job loss at once.
The employment outcome is uncertain. The book contrasts a hopeful scenario, in which new roles replace those displaced, with a harsher one, in which automation creates too few jobs quickly enough for workers to transition. It cites a World Economic Forum forecast of 133 million roles created and 75 million displaced by 2022. That was a forecast available when the book was written, not a settled account of what happened afterward.
Prepare people to adapt. The authors favor lifelong learning, creativity, critical thinking, collaboration, and the ability to learn new skills, rather than training only for today’s software. Companies can assess how roles are changing, retrain workers, and redeploy them. One example is Sumitomo Mitsui Banking Corporation, which reported eliminating 1.6 million work hours through automation while keeping headcount stable and moving employees into a new productivity-focused business.
Planning for both outcomes is more robust than assuming history will repeat itself or that mass unemployment is inevitable. The key variable is not only how many jobs appear, but how quickly workers can access them, whether they pay adequately, and who bears the transition costs. Reskilling helps, but it cannot guarantee a suitable role for every person, especially when change is rapid or concentrated in particular regions. Forecasts should therefore guide preparedness and scenario testing, not be treated as precise predictions.
Share automation’s gains before productivity deepens inequality.
Productivity does not distribute itself. The authors warn that automation can increase output while concentrating the gains among owners of capital and highly skilled workers. They discuss Universal Basic Income, or UBI, as one possible way to ensure people can meet basic needs regardless of employment. A Finnish experiment described in the book replaced benefits for 2,000 participants with an unconditional payment of about $30 per month. It found no significant employment or income difference from the comparison group, but participants reported better mental and physical well-being.
Reconsider what counts as work. The book argues that automation could make more time available for care, education, creative activity, and community life, while questioning why paid employment is the main measure of contribution. Shorter workweeks, retraining, and income supports are possible choices, not automatic results of better technology. Public policy and workplace decisions determine whether productivity gains become broadly shared security or simply higher returns for a few.
This is the book’s most consequential move: it treats automation as a choice about social arrangements, not merely a technical forecast. A basic income is one policy proposal among many, and the small Finnish payment cited here cannot settle the case for a full-scale program. But the underlying question remains important: who receives the time and wealth technology creates? Any serious evaluation should consider income, bargaining power, access to meaningful activity, and the value of unpaid care, rather than judging success only by output per worker.
Analysis
The book’s central contribution is its insistence that intelligent automation is a coordinated redesign of work, not a synonym for artificial intelligence or a collection of software robots. Its most practical model is the combination of capabilities: systems must interpret information, communicate, make or support decisions, and execute actions across a process. The bank fraud and onboarding examples make that claim tangible. They show why isolated automation can shift work and frustration rather than remove them, while connected redesign can improve customer outcomes and operating performance.
The strongest guidance concerns implementation. The authors connect leadership, process redesign, data management, governance, and employee transition, and they recognize that a successful pilot proves little about the ability to maintain thousands of programs. Their emphasis on task-level change is also more careful than claims that whole occupations simply disappear. A job is a bundle of activities, and automation changes that bundle unevenly.
The book is less conclusive when it moves from organizational cases to global totals. The estimates of trillions in avoided losses and millions of lives saved combine multiple domains, timelines, and assumptions. They work best as hypotheses that invite investment and scrutiny, not as established results. Similarly, forecasts about future employment are scenarios, not evidence that one outcome is assured. The text was written in 2020, so its market projections and some technology descriptions should be read in their historical context.
The deeper argument is that productivity is not the same as human benefit. Automation can improve service, reduce errors, and free people from routine tasks, but those gains do not automatically become better jobs, shorter hours, equal opportunity, or public goods. The book’s optimism is most useful when paired with its own institutional questions: who controls the systems, whose data they use, who can challenge their decisions, and who shares the value created? Those questions determine whether IA augments human capacity or merely makes organizations more efficient at distributing rewards unevenly.
Review Summary
Intelligent Automation receives mostly positive reviews, praised for its comprehensive coverage of IA and its potential impact on business and society. Readers appreciate its insights into implementation challenges and future scenarios. Some find it optimistic about IA's benefits, while others value its balanced approach. Critics note disorganization and overreliance on IA as a solution. The book is recommended for those interested in understanding the Fourth Industrial Revolution and its implications for various professions.
People Also Read
Glossary
Intelligent Automation (IA)
Connected knowledge-work automationA combination of software technologies and methods that performs business processes on behalf of knowledge workers, with little or no human intervention. IA can combine tools for interpreting documents, understanding language, making predictions, and carrying out digital actions. The book distinguishes it from AI alone by emphasizing process execution and the coordination of technologies toward a business outcome.
Four IA capabilities
Four kinds of digital workThe book’s organizing model for IA technologies: vision, language, execution, and thinking and learning. Vision interprets images and documents; language processes text and speech; execution performs digital actions; thinking and learning analyze data and support predictions or decisions. A complex process combines several capabilities rather than relying on one tool.
Triple-A artifact
A framework for redesigning workThe authors’ framework for improving employee work through three choices: automate routine activities, augment employees with tools or insights, and abandon practices that add little value. It is meant to help organizations redesign how workers spend their time rather than focus only on replacing manual steps. The authors’ estimates of each category are based on their own research.
Robotic process automation (RPA)
Software for rule-based actionsA configurable software robot that performs routine, rule-based tasks through computer interfaces, such as copying information, opening applications, and sending emails. RPA imitates a person’s clicks and keystrokes but does not independently understand or reason about a task. The book presents it as useful for repetitive work and for connecting automation to legacy systems.
Intelligent character recognition (ICR)
Automated document understandingA document-processing approach that combines optical character recognition with language processing and machine learning. It reads text from documents, interprets the meaning and location of information, and learns patterns across different layouts. Unlike basic character recognition, ICR can handle less standardized documents such as invoices from many suppliers, while routing difficult exceptions for human review.
Center of Excellence (CoE)
Central automation governance teamA team that coordinates an organization’s automation program. The book assigns it responsibility for delivery standards, reusable tools, policies, training, innovation, and oversight of applications entering production. A CoE typically combines business and IT expertise, and can work with local champions so that employee-led development follows common safeguards.
Automation Operation Center (AOC)
Automation maintenance functionA group responsible for keeping deployed automation applications working and managing issues, changes, and escalation. The authors distinguish it from a Center of Excellence, which develops and scales the transformation. Separating delivery from ongoing operations helps ensure that maintenance does not consume the capacity needed to build new automation programs.
Process discovery
Observation of computer-based workA method that records user interactions with computer applications, including clicks and process steps, to document how work is actually performed. Software analyzes these observations to map processes, identify variations, and estimate automation potential. The resulting process maps can help teams review current practice and may provide a starting point for generating automation code.
Process mining
Analysis of system event logsA method for reconstructing and analyzing business processes from event logs produced by systems such as enterprise resource planning software. It can show process paths, variations, and potential noncompliance. Unlike process discovery, it depends on structured logs and therefore cannot capture work that occurs entirely in tools such as email or spreadsheets.
Democratization of IA
Broader access to automation toolsThe effort to make IA development accessible to more employees through low-code and no-code tools, training, communities, and reusable applications. Wider access can surface opportunities and increase employee ownership. The book pairs this approach with governance: users can create and test applications, while expert teams review and authorize deployment into production.
IA generated by IA
Automation that builds automationThe use of automation technologies to speed the creation and operation of other automation applications. Examples include tools that observe work to discover processes, analyze system logs, generate portions of code, automate machine-learning development, or predict application failures. The aim is to reduce the manual cost and time required to identify, build, and maintain IA.
Insight workers
Workers focused on human judgmentThe authors’ proposed evolution from knowledge workers, who primarily handle and apply information, toward people who use judgment, intuition, critical thinking, and creativity to interpret it. The distinction reflects the book’s view that machines can increasingly process knowledge, while people should focus on generating meaning and making judgments in complex or novel situations.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.