Machine Learning - eBook (EN)

Keys to Successful Innovation through Artificial Intelligence

Issue link: https://read.uberflip.com/i/1484611

Contents of this Issue

Navigation

Page 6 of 10

Harvard Business Review Analytic Services 5 Briefing Paper | Keys to Successful Innovation through Artificial Intelligence "We discovered that it's really hard to share best practices across many teams," says Viehhauser. "So, [we] build up [our] use cases, and after having executed many use cases, then we take a step back, reflect, and ask [ourselves], 'What can we harmonize across all these use cases? What can I share in terms of know-how?'" According to Viehhauser and Maenner, the BMW Group centralized its resources and created cross-functional analytical teams to better evaluate its needs and to discover new use cases that could be addressed using AI/ML principles. It also made information more accessible and available throughout the company. Even sharing code repositories and data catalogs proved to be a big hit with its employees, all of whom have an opportunity to learn data science through a specified and tailor-made training program. "To make the transformation work, you must take people where they are and train them on the competencies," says Maenner. "They do not need to be data scientists, but they need to understand what data or what AI is. It's important to give tech teams the insight and purpose to help them understand how and when they're making the difference. Having a lot of people trained in the technology and its possibilities is the power in what we have today." ADP's adaptation began with one small team working directly with Berkowitz. Today, the company operates on a hub-and-spoke model, with 11 different teams led by experts in specific areas such as tax, sales, or marketing who are tasked with spreading the capabilities throughout the company. Recognize a failing project. There's a common denominator among companies that fail in their efforts to adopt AI: They lack the vision, discipline, talent, alignment, right information, and right use case. "AI often starts with the sandbox. It's no surprise that moving from a sandbox to production is super hard to do," says Viehhauser. "You have to come up with templates that people can reuse to make the time moving from [the] sandbox to production really short. Then you're able to ultimately create value and leverage benefits." So, what separates successful projects from unsuccessful ones? Many projects fail because they have a data problem. They may have the wrong kind of, not enough, or biased data. Unsurprisingly, AI models need a massive amount of good- quality data. While no company sets out to create a biased AI model, it can happen if diverse perspectives are omitted in the design process. "In the early days of the big data revolution, companies bragged about how much data they had. And I'm thinking to myself, 'Everyone's got a lot of data. That's not a competitive differentiator. Tell me about the productive value!'" says Borne. Viehhauser says that the BMW Group established a centralized, on-premises data lake in 2015 that collects and combines anonymized data from sensors in vehicles, operational systems, and data warehouses to derive historical, real-time, and predictive insights. However, the company needed to more easily scale its platform to support the growing demands of internal and external stakeholders. "To make the transformation work, you must take people where they are and train them on the competencies. They do not need to be data scientists, but they need to understand what data or what AI is," says Mark Maenner, head of data transformation for the BMW Group.

Articles in this issue

view archives of Machine Learning - eBook (EN) - Keys to Successful Innovation through Artificial Intelligence