Applied Artificial Intelligence: An Introduction For Business Leaders

by Mariya Yao

While engineers and researchers must master the subtle differences between various technical approaches, business and product leaders should focus on the ultimate goal and real-world results of machine learning models.

Inferential statistics is used to draw conclusions that apply to more than just the data being studied. This is necessary when analysis must be conducted on a smaller, representative dataset when the true population is too large or difficult to study. Because the analysis is done on a subset of the total data, the conclusions that can be reached with inferential statistics are never 100 percent accurate and are instead only probabilistic bets. Election polling, for example, relies on surveying a small percentage of citizens to gauge the sentiments of the entire population. As we saw during the 2016 US election cycle, conclusions drawn from samples may not reflect reality!

Machine Learning What happens if you want to teach a computer to do a task, but you’re not entirely sure how to do it yourself? What if the problem is so complex that it’s impossible for you to encode all of the rules and knowledge upfront?

Unsupervised learning occurs when computers are given unstructured rather than labeled data, i.e. no input-output pairs, and asked to discover inherent structures and patterns that lie within the data.

Rather than building custom deep learning solutions, many enterprises opt for Machine Learning as a Service (MLaaS) solutions from Google, Amazon, IBM, Microsoft, or leading AI startups.

To help business executives comprehend the functional differences between different AI approaches, we designed the Machine Intelligence Continuum (MIC) to present the different types of machine intelligence based on the complexity of their capabilities. While we’ve defined the continuum to contain seven levels, keep in mind that the distinction between levels is not a hard line and that many overlaps exist.

Systems That Act The lowest level of the Machine Intelligence Continuum (MIC) contains Systems That Act, which we define as rule-based automata. These are systems that function according to some predefined script, often by following manually programmed if-then type of rules.

Systems That Predict Systems That Predict are systems that are capable of analyzing data and using it to produce probabilistic predictions.

Systems That Learn While Systems That Learn also make predictions like statistical systems do, they require less hand-engineering and can learn to perform tasks without being explicitly programmed to do so.

Systems That Create We humans like to think we’re the only beings capable of creativity, but computers have been used for generative design and art for decades.

Machine learning and AI are not always the right solutions to a problem. Identifying the right problem and its solution requires tight integration and adaptation between your products and your users as well as a collaborative relationship between your team and your users.

Prioritize Domain Expertise and Business Value Over Algorithms When working with Fortune 500 companies looking to reinvent their workflows with automation and AI, we often hear this complaint about promising AI startups: “These guys seem really smart, and their product has a lot of bells and whistles. But they don’t understand my business.” In most cases, having and using a fantastic machine learning algorithm is less important than deploying a well-designed user experience (UX) for your products. Thoughtful UX design that delights users will drive up engagement, which in turn increases the interactions you can capture for future data and analysis.

Thoughtful UX compensates for areas where AI capabilities may be lacking, such as in natural language processing (NLP) for open-domain conversation. In order to develop “thoughtful UX," you’ll need both strong product development and engineering talent as well as partners who have domain expertise and business acumen. A common pattern observed in both academia and industry engineering teams is their propensity to optimize for tactical wins over strategic initiatives. While brilliant minds worry about achieving marginal improvements in competitive benchmarks, the nitty-gritty issues of productizing and operationalizing AI for real-world use cases are often ignored. Who cares if you can solve a problem with 99 percent accuracy if no one needs that problem solved? What’s the utility of a tool whose purpose is so arcane that no one is sure what problem it was trying to solve in the first place?

A key milestone in the corporate digital transformation is the development of a centralized data and technology infrastructure. These two elements connect consumer applications, enterprise systems, and third-party partners and provide access to a single source of truth that contains relevant, up-to-date, and accurate information for all parties. Designing and implementing the infrastructure needed for enterprise-scale AI requires a strong and dedicated technology team that can develop internal application programming interfaces (APIs) to standardize access to both data and your company’s internal business technology. Doing so will enable your company to streamline enterprise-wide data analysis, accelerate product development, and respond more quickly in evolving markets. Internal APIs will also reduce the communication overhead needed to hunt down specific data, negotiate access, and interpret variations. You will also avoid duplicating software development work across different departments that have overlapping needs and goals.

Due to past successes, some leaders prioritize their own beliefs and methods and are openly hostile to analytical approaches and centralized technology. Almost all of us have worked with colleagues with dogmatic qualities in our professional careers. They have a special name: HiPPO, which stands for “highest paid person’s opinion." HiPPOs insist that their strategy is the right direction for the company, based largely on the fact that they came up with the idea.

CEO, CTO, CIO, CDO, or CAO? Finding the right stakeholder to champion a high-risk, high-reward technology initiative is half the battle. In a company that is traditionally conservative towards technology and digital investments, you may have a hard time convincing your CEO to champion AI initiatives. If that’s the case, try to find executive buy-in as high up as possible, ideally within the C-Suite or even at the board level.

Chief Technology Officers create technology for an enterprise’s external business or individual customers. The CTO defines the technology architecture, runs engineering teams, and continuously improves the technology behind the company’s product offerings. Creativity, technical skill, and ability to innovate are essential to a CTO’s success.

Chief Information Officers manage technology and infrastructure that underpin their company’s business operations. The CIO runs an organization’s IT and Operations to streamline and support business processes. Unlike the CTO, the CIO’s customers are internal users, functional departments, and business units. CIOs typically adapt and integrate third-party infrastructure solutions to meet their unique business needs and do less custom development than CTOs do.

Regardless of whether they are your primary AI champion, CIOs will likely play a vital role in implementing AI in an organization due to the need to develop and integrate infrastructure to support AI. ML systems and data mining systems require complex storage, networking, and computing systems that will require the CIO’s input to implement in many enterprises.

Since data touches all aspects of enterprises, Chief Data Officers (CDOs) are becoming increasingly common,(46) but their mandate is more often the security, regulation, and governance of enterprise data.

Along with the CDO, the Chief Analytics Officer (CAO) is a relatively new role that has emerged to manage enterprise investment in big data and analytics. Companies that are early in the maturity cycle for big data may still be working to integrate, clean, organize, prepare, and transform data into an institutional asset. Once a CDO or comparable leader has organized high-quality data, a CAO can then apply meaningful analytics to solve business problems. The roles overlap, and the titles are often interchangeable.

Why You Need a Multi-Disciplinary “AI SWAT Team” Executives alone cannot bring about organizational change, especially of the magnitude that AI can potentially make across an enterprise and industry. Some of your most important stakeholders are your front-line employees and middle managers who will be integrating, using, and overseeing AI tools every day. Many of your employees likely have a strong fear that AI and automation will take away their jobs. Unlike you, they may not initially understand how these powerful technologies can be used to eliminate their lower-value tasks, free them to perform creative and strategic work, and thus augment their output. We have seen firsthand how failure to educate, include, and adapt to these important voices within an organization can lead to resistance, political infighting, and internal sabotage. You will need to put together a multi-disciplinary, cross-departmental “AI SWAT Team” composed of stakeholders in different departments, different functional roles, and different hierarchical levels. The responsibility of this team will be to support your champion and to identify, prioritize, execute, and evangelize your highest-ROI opportunities for automation across the whole company. They will also be critical for identifying potential pitfalls in your organizational design, technical capabilities, and strategic and tactical plans. These diverse views will be critical to help you and your champions form a clearer and fuller picture of t...

Most AI implementations are cross-functional and require input from multiple departments. A retooling of your accounts payable system will need inputs from finance, legal, security, and technology. A project may also pull in human resources, if employees need to be reassigned, and operations, if processes require adjustments. To succeed, you will need support from other executives, their front-line managers, and their staff. Here are some things that you can do to win support from your organization. Focus on Revenue Potential A key strategy is to appeal to your business leaders about the potential of increasing the bottom line. AI can save time and energy, reduce costs, and increase profits, which then provide executives an opportunity to grow their business lines and advance their careers. Stay Ahead of the Competition The fear of missing out (FOMO) is also a very strong motivating force. If business unit leaders fail to take action, emphasize that they are setting themselves up to fall behind competitors who are jumping on new technology. Not investing in the organizational and technical requirements to adopt AI may mean falling so far behind that you’re unable to compete in the future. Start Small and Show Early Wins Pick a smaller, sure-win project to demonstrate possibilities. While returns may be limited, an early success will give you confidence when you request that the project scope be expanded. Aim for something with a short time horizon that can be completed wit...

How to get buy in on AI stratgies from stakeholders and maintain focus on it as a "must-do" implementation.

Research finds that while 45 percent of tasks are automatable, only five percent of overall jobs have been supplanted by automation.(50) AI systems largely handle individual tasks, not whole jobs.

If humans can outsource repetitive and mundane tasks to AI, then they can devote more attention to tasks requiring strategic skills such as judgment, communication, and creative thinking. Eliminating boring jobs that employees dislike can also improve morale and interest as they take on increasingly more meaningful work.

founder of leading AI company Element.AI, calculated that there are fewer than 10,000 people in the world currently qualified to do state-of-the-art AI research and engineering.(53) Most of them are gainfully employed and hard to poach. If you’re looking to recruit fresh graduates, the head of a prominent Silicon Valley AI lab recently confided to us that American universities only graduate about 100 competent researchers and engineers in this field each year!

Based on a study of public job listings among US employers, Forbes found that the top 20 AI recruiters, led by Amazon, Google, and Microsoft, spend more than $650 million annually to woo elusive researchers and engineers.

Due to the limited talent pool, many enterprises may not be able to hire applicants to fill all of these roles in-house. In this case, these companies could potentially fill in the gaps with enterprise solutions like Machine Learning as a Service (MLaaS) or AutoML technologies.

Passion for Your Problem “We get plenty of resumes from people with talented machine learning and data science backgrounds,” says Zhen Jiang, Lead Analytics Supervisor at Ford Motor Company. “What I am much more concerned about is whether they have a passion for cars and mobility.”(56) Talented engineers and researchers can go to any company in any industry that they want. Focus on finding applicants who are particularly excited by the unique problems that you face and the datasets that you own. Check whether they have done past research or projects related to your industry, and seek out talent at topical events that attract both enthusiasts and a more focused audience.

Perfect is the enemy of the good. Look for pragmatic applicants who recognize that a “good enough” model that meets product deadlines is better than a model that sits in development awaiting “just a few more tweaks.” This quality can be hard to find, as most scientists and machine learning experts, especially those with PhDs, are often trained to seek perfection.

You can also host competitions on Kaggle or similar platforms. Provide a problem, a dataset, and a prize purse to attract competitors. This is a good way to get international talent to work on your problem and will also build your reputation as a company that supports AI.

Quality of Data The data that you have for analysis is ideally clean and annotated. In a recent survey of data scientists, 57 percent reported that data cleaning consumed 60 percent of their time and was the least enjoyable part of their job.(59)

8. Plan Your Implementation Once you have assessed that your organization has the requisite culture, leadership, and talent to succeed in AI initiatives, the next step is to identify business opportunities with the highest return on investment (ROI). Rank Business Goals Prior to beginning any technology investment, you and your executive team must be clear on the problems you want to tackle, the reasons why solving these problems is a priority for your organization, and the metrics for success. You should have clear strategic goals at either the company- or department-level. Common goals include increasing revenue, cutting costs, and entering new business lines. Artificial intelligence can advance many of these goals, but implementation difficulty and impact will vary. In the following sections, we give you several frameworks for evaluating current enterprise workflows and technologies, performing opportunity analysis, and clarifying your organizational goals and metrics.

If you want to assess whether automation is worthwhile, you may want to compare your current performance with the performance of organizations that have already implemented automation. If it’s difficult to get department-level metrics for competitors, consider hiring experts who have worked extensively with many companies to help you establish benchmarks. Don’t limit yourself to data from your own industry. Technology companies have disrupted many traditional business models, so look beyond the obvious comparisons.

In a SWOT analysis, first evaluate the internal factors affecting your business. What are your strengths? What makes the department so great? Which projects or teams are finding success? Next consider your business’s weaknesses. Which projects or departments are unprofitable? What resources do you lack? What can be done better?

Next, consider the opportunity size. Is the opportunity big enough to warrant an AI solution, or can your employees or an older technology adequately solve the problem? Conversely, even if this specific opportunity can be solved more cheaply or easily with human power for now, can an AI-based solution be leveraged for similar tasks in the future?

Then, consider the investment level required. How much time and money will you need to allocate towards the problem? Don’t forget to include internal costs. For example, even if an external vendor implements a solution, you will still incur internal management costs.

While the next factor, the return on investment (ROI), is never certain, you should estimate an upper and lower bound and a likelihood of success. Understand your break-even number. Don’t forget to include internal costs for ...

Does this project seem like a sure bet, or is it a moonshot opportunity? Set the performance level that a new technology needs to achieve in order to be deemed successful.

Data: What Happened? At this level, organizations want to use descriptive and exploratory statistics to answer fundamental business questions about what has happened in the past. This requires you to have collected the right kinds of data. In practice, we have found that while many enterprises have collected plenty of data, it is often dispersed across departments and owners, in the wrong format for analysis, or simply insufficient for answering basic business questions. If this applies to the use cases for which you’re exploring AI and automation, you need to more clearly define your needs with requisite stakeholders in order to arrange for the correct data to be collected. Since knowing what has really happened in your business is critical for developing the best strategy for your future, this step is worth the investment of time, manpower, and budget. Information and Knowledge: Why Did It Happen? Once you and your executives are clear on what has happened in your business in the past, the next step is to understand why. During this phase, you move beyond statistical analysis of data into understanding and encoding expert logic for why certain results occurred. For example, your data may show that you had an unusually poor sales quarter last year. When analyzing your sales results along with data from human resources, you may discover that the poor showing is due to your top sales representatives leaving the company around that time. While some knowledge can be encoded, ...

Tighten the loop between data, insights, actions and results.

Artificial intelligence techniques are most economically used to automate problems that are time-consuming, repetitive, and simple in scope. Most machine learning approaches also require large quantities of clean, trainable data.

Just because a problem can be automated does not mean that an AI solution is appropriate. For example, using AI to generate an annual shareholder report may not be a good investment if you can easily hire analysts to do the work more cheaply than you can build or buy a software alternative. The task also requires strategic communication and contextual understanding, which modern AI technologies struggle with. By contrast, automating the sales process provides faster service to your customers and reduces the likelihood of clerical error, which can create value by streamlining paperwork, improving record keeping, and increasing cash flow.

Here are key questions to ask when evaluating whether your problem needs an AI solution: 1.Is this a process that can be solved using machine learning? Break down the process into its components to determine inputs, outputs, and contingencies. How long does it take to perform? How often is each step taken? How many people perform the same task? This gives a sense of the opportunity size for automation. 2.Is it suitable for machine learning? Identify the decision-making process for each task component. Do answers to questions come to you immediately, or do they require longer deliberation? Furthermore, if multiple people were answering the same question, would they all reach the same conclusion? Machine learning is best used to replicate human decisions for tasks where correct answers are clear and measurable. 3.Is data available? Are there sufficient volumes of relevant data associated with the process from which an algorithm can learn? How easily accessi...

If you already store data and build applications on Amazon Web Services (AWS), Microsoft Azure, Google Cloud, or Apple’s iOS platform, using tightly integrated machine learning solutions like AWS Rekognition or Apple’s Core ML can simplify work for your own developers and may be the most economical business decision.

Should you invest in your technical capabilities and turn your organization into a technology company like Google and Amazon? Or should you stick to your core business expertise and find third-party solutions for your AI needs? Here are some criteria that will guide your decision-making process: Building Internally There are several important criteria to consider when thinking about building in-house. First, determine whether your technology is a core functionality of your business. In 2001, Target outsourced its e-commerce website to Amazon because the company executives did not see online shopping as a major revenue channel. Nearly a decade later, Target finalized realized that this was a huge mistake and took back control.(66) As a leader, you will have to judge whether the AI capability that you want to build is central to your long-term business. Second, evaluate the availability of in-house talent. There is a massive skills shortage in the field of artificial intelligence. Even if your organization has a sufficient pool of talent for core products and services, you may hesitate to deploy these scarce resources to departments such as HR or customer service. The need for technical talent extends beyond the initial build, as you’ll need specialists for ongoing maintenance and performance monitoring. Third, set a timeline for deployment. Even if your company has the talent and the data to build a custom solution, you may not have access to them until higher-priority proj...

Given the hype, many third-party vendors claim to “use AI” but are really using the phrase as a marketing tool. When looking at our Machine Intelligence Continuum from Chapter 2, you’ll see that most vendors are actually using technologies that fall under Systems That Act and Systems That Predict, not Systems That Learn.

Data Connection Does the prospective product offer seamless connections with the other enterprise tools on which you depend, such as your data and analytics provider or CRM system? Is the integration built-in, and if so, is it offered via an application programming interface (API) or platform? If not, will it require custom development?

Professional Support Most AI systems will need to be continually trained and updated. How accessible and competent is the vendor’s professional services team to help onboard and maintain your AI system? Particularly for large enterprises, does the vendor have the capability to support the scale of service that you require?

No third-party solution will be perfect for your use case, so it is equally important to understand the features as well as the limitations of the software that you decide to buy. The sales staff at a vendor company may not have enough incentive to be fully honest with you, so the best way to get honest answers is usually to survey their existing customers. Ask them if they have run into problems with scalability, stability, security, compatibility, or ease of use. What frustrates them most about the product? What do they wish it could do that it currently can’t? Keep in mind that technologies and the companies behind them are constantly evolving, so also find out how quickly and robustly a company addresses reported issues and improves the performance of their solutions.

Calculate ROI and Allocate Budget Impressive ROI on successful projects whets our appetites for AI. Through better search algorithms, Netflix reduced cancelled subscriptions that would have dropped revenue by one billion dollars annually.(70) Peloton, a popular indoor cycling company, cut customer support tickets by 25 percent with intelligent self-service.(71) Harley-Davidson used AI to target potential customers and to adjust their sales copy, increasing sales leads in New York by 2,930 percent.