Speaker 1: 6 Secrets to Large-Scale AI Implementation It's one thing to build an AI or machine learning solution in isolation, which itself is not so trivial. But, it's an entirely different set of problems to integrate the AI solution as part of your operating processes. By my estimate, if the former is about 20% of the work, the latter is 80% of the work. But, many business leaders get enarmored by the promise of AI and set unrealistic expectations without understanding the complexities of integrating the solutions into their organization. In this video, I'll share some secrets to succeed. But first, hit that subscribe button and click on the bell icon so you get notified of new videos on architecture and AI that I share every other week. For the purposes of this video, I'll use machine learning, which is a sub-discipline of AI. Machine learning seems to be getting disproportionately more attention among all the other disciplines of AI. It also includes another sub-discipline called deep learning. By the way, sometimes when I say ML, it means machine learning. Here are some business use cases for machine learning. Add ML to your call center to handle most of the routine traffic and leave the really complex queries from customers to be handled by humans. Humans are better at adapting and creative problem solving at this point in time anyway. Add ML to automatically approved loan applications since machines are good at considering multiple dimensions of data and historical data to approve or deny loans. Historical bias, such as approving a higher percentage of loans for men than for women, has to be removed from the data to perform well. Add ML to predict consumer buying patterns based on many factors like location, weather, median income, etc. so that the stores can stock their shelves optimally and costs don't increase, and we don't leave profit on the table either. A large percentage of ML deployments don't live up to expectations. I'll share six secrets for a successful AI integration. The first is data. Data is the most critical aspect of most machine learning algorithms for both training and operations. Data is used to train the machine learning models, of which there can be many types. Models are the core of decision making. Models drive the applications and if these models are wrong because they have been built on poor quality data, for example, then the application will produce poor results. If the application produces poor results, the business processes that use these applications will be fragile, resulting in poor customer experiences. While much of this problem can be tackled in the 20% of ML solution building phase, in the larger context of the organization, we need a data strategy. We have to manage, curate, tag, clean, govern, refresh, and prioritize structured and unstructured data across the enterprise. Just because your organization implemented a great chatbot does not mean that other data is of great quality or other ML projects will be as successful. The second is business domain knowledge and how processes work to perform their activities. A typical organization may have many processes and supporting capabilities. If you automate broken processes, it's still broken, just automated and faster. To prioritize which processes to work on first, you'll need a process architecture, which is a framework that organizes your processes. From this kind of a structured organization, you can pick which processes you're going to automate with machine learning. Assume you have picked the right processes when you want to introduce ML based on some measures such as giving you a competitive advantage or drastically improving customer experience. Then you can look deeper into understanding the steps of the process that have to change. This again is often not easy because each step may have multiple dependencies on multiple systems, data, on people roles, external organizations like partners, government, and even other processes. When you replace this step with ML, you will have to redesign these integrations, which in turn might have other implications causing a domino effect. If you agree so far, type in the word agree in the comment section below. The third are the IT systems. Most businesses use software to drive their business processes. And this is not one piece of software, but a set of them, perhaps running on old systems with different data stores, different permissions and so on. Just think of how many applications you yourself might be using in your daily work. Email, messaging, meeting apps, word processing, spreadsheets, presentation software, maybe employee management, and a whole lot more. The businesses similarly have enterprise resource planning software of different kinds to manage anything from handling customer orders to delivering that on time. Changing such systems is not trivial. Imagine your business just replaced Microsoft Word with Google Docs, and even such a relatively trivial change might cause a revolt. Transformation is often not just a technical challenge, but more of a people, cultural, and governance challenge. You'll have to re-skill people whose roles might change, and you need to have consistent and frequent communication. Which brings us to the fourth Which brings us to the fourth secret, ensuring that the transformation can be done in a systematic and stable way that is adaptable to future changes. You need a discipline, like enterprise architecture, that can handle the complexity of being able to think about and manage multiple moving parts, like the business applications, data, infrastructure, and the external environment. EA looks at the organization and change from a holistic perspective so that the global optimization is preferred instead of local optimization. This often means that multiple departments have to work together to optimize global measures, such as customer satisfaction, instead of each department only optimizing their own measures. What's the point in manufacturing something really fast, for example, when the warehouse can get backlogged and the delivery time to the customer is still not improving? In fact, it'll have the opposite effect of increasing storage costs. So far so good? Then please type in the word great in the comment section below so I know. The fifth secret is a fundamental shift in thinking about ML. While conventional software uses deterministic logic, where business processes operate with a set of clearly defined if-then rules, ML is a probabilistic approach to decision making. Classification problems will provide answers to questions such as whether the transaction is a fraud, or whether a loan is approved or not, or whether a patient has a disease or not. On the other hand, prediction algorithms will provide answers to questions like how much the sales would occur on a holiday weekend, what is estimated delivery time, and so on. Employees have to be trained on such a paradigm shift, the reality and the challenges of the business, conceptual knowledge of AI, and their ability to problem solve and their ability to think creatively. Overseeing all this is the sixth factor of leadership drive. Leaders have to know, craft, and execute the right strategies. And this can only be done by taking a holistic view of the ecosystem of the company that includes customers, competitors, partners, internal and external capabilities, regulators, and technology. When an organization embarks upon a large-scale transformation effort, it is important to specifically consider AI. Tomorrow's companies will compete and thrive on how well they have mastered AI. A top-down approach is necessary to understand the business value that it'll provide, and yet not be oversold on its promise. To sum it all up, the rollout of AI into your organization requires us to consider multiple dimensions carefully and architect the whole system in a consistent manner before starting to implement siloed solutions. Think big, execute small, integrate continuously. If you need a thought partner for your organizational transformation, please reach out to me. If you enjoyed watching this video, please consider subscribing. Thank you.