Main body of this blog post was originally published on H2O.ai blog. Published here with permission.

Do you want to make AI a part of your company? You can’t just mandate AI. But you can lead by example.

All too often, especially in companies new to AI and machine learning, team leaders may be tasked by their managers to “start using AI” without having clear goals set about how AI may help the business or how to build an effective AI team. It’s understandable that executives want to take advantage of the great potential value they see that AI and machine learning are delivering for other companies, but successfully putting AI to work in your own enterprise requires more than passing down the order to do it. Whether you are the executive choosing to add AI to your business or a team leader responsible for making it so, be aware that there are practical steps to put in place before you plunge into hiring someone to make models. One of the first practical steps to develop and implement AI is to build a data-aware culture within your organization.

Data-Driven Choices

One of the challenges in developing serious data skills is to get buy-in about the importance of data-driven decisions. People with extensive experience in their particular business sector hold valuable insights – knowledge that is important to the success of AI projects built to address essential business goals. But this experience may result in people relying on “gut feeling” about business processes, predicted performance and the potential for new lines of business. A key step in building the data-aware culture needed to support AI systems is to develop an appreciation among stakeholders in your organization for what data can tell you.  Show - rather than tell - how data can augment, reinforce or refute their experienced view or gut feeling.

One way to do this is to make use of data in making your own decisions and to provide transparency to your teams about how data is influencing your decisions. Develop the habit of asking, “What do we know and how do we know it?” This habit may involve collecting more metrics about essential business processes than you already do, but more often it means making use of the data you already have. Help people within your organization to see data as a way to expand understanding, not just as a report card on their own performance.

Data sources have proliferated in the last few years – data is literally everywhere. This variety of data provides a rich resource for AI and machine learning. You also can raise awareness within your organization of the wide range of data sources you have available and how they can inform you not only about your own business processes but also about the behavior of your customers.

Using automated, machine-based learning systems makes it feasible to make decisions based on data at a level and speed that may not be practical for humans to do. This, in turn, lets you take advantage of how data tells you about conditions in the world around you, and how they affect your business.  It also is the best way to inform you about how your business needs to adjust to changes in the world, a situation underlined dramatically by the COVID pandemic.

Why is it important to get buy-in from stakeholders? One of the biggest barriers to building effective AI systems is to set the business context for AI in realistic ways. Getting buy-in from stakeholders is important to ensure cooperation in defining business goals for AI and for ensuring adequate resources are assigned to develop, implement and maintain AI systems.

AI is a Team Sport

Getting up-to-speed with AI may require hiring talented data scientists or selecting the right company to provide data science as a service. But whether you build an in-house data science team of AI and machine learning experts or work with external talent, there’s more to building an AI system than just the specialists who code and train the models.

Who needs to be on your list of AI talent? Data engineers are a critical part of the success of AI and machine learning projects. In fact, it’s essential to budget a major part of the time and resources allotted to a particular AI project to the effort needed to handle the logistics, from data preparation for training sets to deployment and management of AI models. The following figure shows the relative effort of model building to all the other parts of a machine learning project.

This requirement for many steps surrounding the specific model building process creates a need for a range of skills beyond the ability to work with algorithms. It’s important, then, to think of AI as a team sport.

DataOps helps bring AI to life

Making AI practical and profitable for your business requires the cooperation and collaboration of people with different skills. One of the best ways to achieve this is through a DataOps approach in which you assemble a cross-skill team that has members collectively focused on a shared goal. That style of work improves intra-team communication and avoids the sense that asking someone with a particular specialized skill to do their part is “asking a favor” or imposing on their time. Instead, people are more apt to work together efficiently when they share a goal and understand what needs to be done.

Technology can help with this team approach as well. Tools that improve machine learning interpretability make it easier for data scientists to communicate clearly to others how models are making decisions and how data is influencing those decisions. The H2O Driverless AI platform is an example of technology that not only makes it easier to develop AI models but also easier to explain them.

Next Steps

Building effective AI systems also requires efficient access to data. AI is only as good as the data used to train its models, and HPE Ezmeral Data Fabric provides multi-API direct global access to data. H2O Driverless AI, for example, can access data stored in the HPE Ezmeral Data Fabric directly, without having to copy it out to another system. Furthermore, HPE Ezmeral Data Fabric is the core data layer in the HPE Ezmeral Runtime Enterprise for cloud-native and distributed non-cloud native applications. Containerization of AI applications is a big advantage in being able to train, deploy and run AI models in predictable environments that you can easily control. H2O Driverless AI has been tested and validated for compatibility with the HPE Ezmeral Container platform.

 

Another offering in the HPE Ezmeral Software Portfolio, HPE Ezmeral ML Ops, also works together with H2O Driverless AI to make data science more efficient through automatic machine learning.

 

Find out more about these AI-enabling technologies, including how you can try a 21-day free trial of H2O Driverless AI, by visiting the HPE Ezmeral Marketplace.

 

For additional information about building a data-aware culture, as well as other practical steps in bringing AI into your business, read the free ebook Practical Advice for Making AI Part of Your Company’s Future.

 

And please check out the HPE DEV blog for more articles on developer-focused topics.