Welcome to AI book reviews, a series of posts that explore the latest literature on artificial intelligence.
While there is a lot of excitement about how advances in artificial intelligence will help the enterprise sector, the reality is that most efforts fail. Study after study shows that organizations of different sizes are struggling to bring machine learning into their operations, and many initiatives end up being shelved or used in a very limited capacity.
The adoption of applied AI is very difficult and costly, wrought with pitfalls, and requires fundamental changes at different levels. However, as the tools and processes mature, more companies will be able to take advantage of enterprise AI while reducing the risks and costs of adoption.
Demystifying AI for the Enterprise, a book written by six experienced executives and thought leaders, brings to light some of the ways that organizations can ease their way toward applied AI. The book shares best practices and practical case studies of how AI is being applied to different industries.
Chapter 1 of the book, written by Prashant Natarajan, Vice President of Strategy and Products at H2O.ai, introduces IMPACT, a general framework that can guide the adoption of AI in the enterprise. Natarajan describes IMPACT as “a checklist to create business value and ensure the lasting success of digital transformation with data, analytics, and AI.”
Great products are built on top of great visions to address unsolved problems or provide solutions that are better than the incumbents. AI doesn’t change this. You’ll still need to have an eye for new markets, gaps in existing markets, and pain points in the lives of people and organizations. As Natarajan writes, “The most successful organizations—be they enterprises like Amazon or Tesla or Apple, or several start-ups and unicorns—thrive on creating new markets, new users, or even new uses for existing products.”
As other experienced people in the field suggest, a good AI strategy starts with knowing what problem you’re trying to solve. However, having a good understanding of the capabilities and limits of current AI technology will help you look at problems from a new perspective.
For example, the traditional e-commerce business model revolves around the “shop then ship” process. But with the power of predictive models, you can think about moving toward a “ship then shop” model. This means that items arrive at the customers’ homes before they even shop, and the machine learning model’s predictions are accurate enough to make the experience both convenient for the shopper and profitable for the seller. Another example is the healthcare industry, where service providers can think about improving the quality and reducing the costs of care and insurance through predictive care powered by machine learning.
“AI presents organizations with tried and tested methods to apply imagination by enabling leaders to leverage new data-driven insights and amplified/augmented intelligence… to go beyond current business value streams and processes, incorporate insights via new operational solutions and processes, or even refreshing existing ones to reach new users and markets,” Natarajan writes.
Thinking about the opportunities of enterprise AI is one thing. Having an organization that is ready to adopt AI technologies is another. Natarajan tracks an organization’s “maturity” to adopt AI across different dimensions, including strategy, leadership, processes, and data.
Writes Natrajan, “Evidence shows us that moving up the maturity curve is not only possible, but is necessary to transform your enterprise from an information-rich but insights-poor organization to one that is not only insights-rich but also an integrated cognitive enterprise where AI enhances and supplements human intelligence and experience to create an organization that can make use of the advantages of both human and machine intelligence to their utmost.”
In terms of maturity, organizations can range anywhere between “nascent” and “symbiosis.” At the “nascent” stage, there is no AI strategy, the leadership does not support the adoption of AI, the data infrastructure is not ready to train and feed machine learning models, and the organization uses very rudimentary analytics tools (e.g., spreadsheets and simple business intelligence dashboards).
At the “symbiosis” stage, the leadership is fully aware of the latest trends in enterprise AI and how they can fit into the organization’s business model. AI is integrated across the organization’s operations and business outcomes and is directly tied to the goals and needs of the people the organization serves.
At the symbiosis stage, AI helps drive efficiency and gives the organization a competitive advantage (remember the “ship then shop” example above). Data is managed in a way that reduces the barriers to consumption by AI systems. And as discussed in Competing in the Age of AI (another highly recommended book), the company has established an “AI factory” that is constantly gathering new data, learning, and updating its models to improve its AI systems.
Digital native organizations have an easier time reaching symbiosis because they started their operations in the age of cloud computing and advanced analytics tools. For them, success depends on finding the right opportunities, bringing together qualitative and qualitative information, testing and failing fast, and pivoting to the right directions where AI can help them better fulfill the needs of their customers.
Legacy organizations often get stuck in the nascent stage because their processes, workflows, organizational structure, and technical infrastructure were conceived and designed before the age of AI and data-driven product management. This is especially true of long-standing industries such as education and healthcare, where data has mostly been collected for administrative purposes and not to be fed into AI models. Data exists in silos and stored in inconsistent formats and conventions. The workflows, established over decades, have been designed to rely purely on the intuition and experience of human experts, much of which is not digitized. For these organizations, moving up the AI maturity ladder is often more difficult but not impossible.
Natarajan provides some extremely important guidelines, such as finding easily accessible and high-value use cases to automate. These early successes can help to improve leadership and organizational buy-in and create a launchpad to further create wider AI transformation.
You can’t have a successful enterprise AI adoption strategy without putting people at the center. “When investing in AI, think of people first, not technology first. Technology is important, but people will create value, drive adoption, and ensure human–AI symbiosis for your business,” Natarajan writes.
This requires organizations to invest in their people as much as (or even more than) they invest in technology. Admittedly, bringing the right people together to create an AI team is easier said than done. But Natarajan mentions some of the main practices that have remained consistent across organizations that have carried out successful enterprise AI strategies.
Naturally, executive and leadership buy-in is one of the key components of any successful enterprise strategy. As mentioned in the previous section, small pilot projects that show success can help make progress in this regard. But you must also help raise awareness on the resources, staffing, and funding required for your AI projects, and provide a realistic view of expectations your organization can have.
One of the key positions that Natarajan mentions is the “AI/analytics translator” role, which he describes as “a professional who can understand business & product life cycle management and can tie them to AI capabilities.” The AI translator is the person who can facilitate communications and bring your ML engineers, domain experts, and executives on the same page. This person should be able to translate business requirements into AI solutions while at the same time explaining AI solutions to executives and non-technical people.
Finally, Natarajan stresses the importance of tackling AI projects as a team operation in which people from different disciplines and backgrounds work together. Data scientists might be too focused on model accuracy, F1 scores, and other metrics and not enough on the goals behind those metrics. Business leaders and domain experts, on the other hand, might not be able to articulate their goals into data science metrics. But when they come together along with the supporting cast such as data engineers, IT staff, AI translator, product managers, and other people, you can make sure that all efforts are directed in the right way while everyone gets to work where their skills are most needed.
Automation, amplification, and augmentation
While a lot of the narrative surrounding AI focuses on replacing humans, enterprise AI should be more focused on how humans and machines can better work together. Natarajan highlights three key areas:
– Taking the robot out of the human (automation): AI can automate repetitive tasks taking up the time that humans can spend doing more meaningful work. This is probably the first place you would want to look at when developing an enterprise AI strategy because it improves the efficiency of your workforce and makes their work more enjoyable.
– Doing things that humans can’t do efficiently by themselves (amplification): Some tasks require the perusal of huge amounts of data, making them effectively impossible for humans alone. Creating AI systems that help humans find the needle in the haystack and make quicker decisions should be among the goals of every enterprise AI strategy. One example is anomaly detection, which has use cases across different domains, including cybersecurity and healthcare.
– Creating new knowledge, new questions, and new insights (augmentation): This is closely related to amplification and is the area where humans and AI complement each other and address each other’s limitations. AI can help humans quickly test hypotheses and make new discoveries that can lead to new business processes and outcomes. At the same time, humans can use their intuition and common sense to help direct AI resources in the right way and make sure efforts and outcomes are compliant with ethical guidelines. Insurance, retail, and healthcare are some of the sectors that can make good use of AI for augmentation.
This is in line with the vision for human-centered AI, which many other experts in the field advocate for. Basically, human-centered AI acknowledges that AI systems must be designed to work with humans and have human control and oversight baked into them. This makes sure that organizations can make the best use of both human and artificial intelligence.
In addition to technology and talent, organizations need the right culture to make sure they can take full advantage of enterprise AI.
The culture surrounding enterprise AI—especially as the field continues to go through fast-paced transformation—requires organization leaders to embrace change at different levels, including business processes.
AI culture also requires data-based experimentation and exploration. Trust your intuitions, but test them with quantitative and qualitative data and embrace every failed experiment as an opportunity to learn and move on toward the right solution. AI algorithms are excellent tools for fast experimentation.
And don’t stop at first success. Always try to find new ways to improve. If you look at your organization from the “three As” perspective (automation, amplification, augmentation) and embrace new ways to think about problems, there will always be room for innovation and improvement.
As an organization moves from the nascent to symbiosis stage in enterprise AI, it will have to undergo many changes. Processes, roles, tools, technologies, and products will gradually change as AI permeates every aspect of the business.
For example, in the early stages, you might need a chief AI officer to oversee the AI efforts across the organization. But as AI gradually becomes part of your culture, product teams might do most of the work themselves. Likewise, a chief data officer might help transition the company from a siloed data infrastructure to one where data is more fluid and available to data science and ML teams.
At some point, AI and big data will no longer be separate responsibilities and will become an integral part of the work of every team.
In summary, an enterprise AI strategy is more than just a change of technology. As Natarajan writes, “AI is not a widget, a software tool, an infrastructure deployment, or [an] application-specific black box. In order to get the most out of AI and to drive business outcomes and digital transformation at scale, enterprises must involve their business users and leaders into opportunity identification, product and solution development, outcomes, and insights. Merely treating AI as a technology that is driven by data science, analytics, data, or IT teams is the surest way to underwhelming success and failure.”