10 business trends to make or break AI initiatives:

This seems like an obvious one, but with so many potential areas for AI exploration, starting with the right projects—and stakeholders—is crucial for long-term success. First and foremost, the process of identifying and selecting use cases shouldn’t be driven by technology alone. That is, you don’t want to think about AI solely in terms of where you can apply natural language processing, for example, or how you can leverage a labeled data set. Instead, ask where you seek to increase productivity or derive new value. Some of the questions to help leaders focus in this way:

1.     What parts of our business processes are high volume and low margin?

3.     What new markets do we want to enter?

2. Take a portfolio approach that blends practicality and strong potential

Going through the questioning exercise above with the various leaders who may own or touch AI, such as the chief information officer, chief digital officer, chief data scientist, and other specialists (see #3), will enable you to identify where to start.

Ultimately, you want to take a portfolio approach that might include several quick-win projects as well as one or two that are more complex but have significant potential. This emphasis on “practical” AI is one we expect to see a lot of in the coming year.

At the AI Summit, Nvidia’s Brian Catanzaro illustrated this approach. While the company is known as an AI powerhouse thanks to its graphics processing units and AI software tools, Catanzaro, Nvidia’s VP of applied deep learning, talked about how the company is also using AI in less glamorous areas like quality control and resume matching. Likewise, JD.com, the largest retailer in China, not only showcased the more exotic ways it is using AI, drones, and robotics in its warehouses and delivery, but Hui Cheng, head of robotics R&D, also talked about AI’s application in more mundane areas like inventory management, advertising, and pricing.

One thing that came across loud and clear at various workshops was the importance of functional and other business specialists as you develop AI solutions tailored to your organization. These are the people in the trenches who have the knowledge and expertise about your business processes, pain points, and data. They’re also the ones who will help shape solutions and drive adoption.

Fedex Data Scientist Clayton Clouse drove this point home when talking about the key role that operations managers played in the development, testing, and roll-out of a solution that optimized truck inventory. We think of this trend as the third wave of AI in the enterprise. It started with data as a key advantage.

The second wave was all about creating AI platforms using Amazon Web Services, Google Cloud Platform, and Microsoft Azure. And now this new wave will center on vertically focused AI led by domain experts.

Getting a handle on your organization’s data has been front and center for some time now. But AI raises the stakes because it adds another layer of complexity beyond the already formidable challenges of gathering disparate data—including the mushrooming amount of unstructured data that lives outside of the rigid confines of a database—and prepping it for analysis. With supervised machine learning, where humans train and tune the algorithm, you not only need very large data sets, but they also must be labeled so that the model can “learn” to identify the correct outcome.

For even a simple model, labeling might take 30 seconds per label; if you have 10,000 pieces of data, that amounts to about 100 hours of work. Some companies outsource the work of labeling or use open source or crowdsourced data that has already been labeled. Looking outside your enterprise will also become more important as you begin to think about the broader data ecosystem in your industry, such as the partners, customers, regulators, and other entities that are part of your data flows.