Class Takeaways — Data Science and AI Strategy

Kaung Xu: Hello, my name is Kuang Xu. I’m a professor of Operation Research at the Stanford GSB. I work on applying AI and data science to solving business problems. Today I’m here to tell you about a class I teach called Data Science and AI Strategy, and here are five key takeaways.

Data science and AI is not a monolith, and we learn in this class that there are at least two pillars to it. The one pillar being analytics and reporting. This is where a team would use data science, machine learning, and statistics to derive insight to drive strategic decision making. They’re typically consumed by an internal audience.

On the other hand, we have AI driven decision making. These are automated tools, typically software, used to generate pricing recommendation often turned into the form of a product.

The understanding of these two pillars, there are pretty significant implications. For one, we want to understand that when hiring a team, analytics and reporting requires huge amount of communication skills and the ability to convince stakeholders. While on the other hand, when you’re building a product using data science and AI, you really want to pay attention to engineering and product design.

One often underestimates the technical risks involved in developing AI and data science solutions. The reason is follows. In traditional engineering or product design such up as software development, it is often easier to measure the incremental progress you’re making as you develop new features and capabilities. This is not true when it comes to AI and data science.

Oftentimes, you have to make a huge upfront investment, collecting data, training model, and then there’s the intrinsic uncertainty of whether the model works or not and how well it works. Therefore, you won’t see the result until the very end having spent a lot of time and resources.

There are ways we can go into to mitigate this risk, but at the high level what we want to think about is to as much as possible, define incremental milestones to de-risk the process, and making sure that you are constantly making progress as you go along.

The third takeaway is that data by itself is oftentimes not a moat. It is very compelling, intuitive sometimes, to try to build a large dataset as your competitive advantage, but here’s a risk. As you collect more and more data, initially, the data you get speaks to the most common consumers. They’re cheap to get, and yet has huge business impact. But as you build into the niche area of the corner cases, the data becomes hard to obtain while at the same time the consumer base you’re speaking to keeps shrinking.

Initially you start with low cost and high value, and you venture into the territory of high cost of acquisition and low consumer value. So if you’re building a company and thinking about using data as your competitive advantage, just be careful. Oftentimes, it is much better to get enough data so your models can be sufficiently trained, while at the same time devoting the remaining resources into designing better products and better services.

Think about what you truly need before you look at what AI can offer. This is actually nothing new. When a new technology came onto the scene, oftentimes business leaders and decision makers can get a little bit afraid. They’re afraid to miss out on the greatest opportunity. Therefore, instead of focusing on their business needs, they look at the capability of the technology and the great showcases they have and build that thing internally when they don’t truly, truly need it.

Instead, here’s a tip. Maybe you can put a blindfold on your eyes, pretend that AI and machine learning never existed. Ask yourself what you truly need to build in your company today and tomorrow to make it grow. Then unfold yourself. Look at all the technologies that’s available, all the advancements, but now combining yourself to only using them to building these capabilities that you committed to. That way you’ll use AI as a great tool and not as some technology that might lead you astray into bad decisions.

Unlike traditional engineering and product design, the output of AI data science can often be an insight that is quite intangible and even looks a bit obvious post the fact. However, these insights often dictate the direction of the product or even the direction of the entire company. If you have a team like that, make sure you set up the evaluation procedure correctly so that your data scientists and AI researchers have incentive to explore and hopefully discover these really, really important insights that will take your business to the next stage.

AI is moving so fast, but how do you know it hasn’t taken my job already? How do you know that it is me speaking to you right now