Image: ShadeON, Getty Images/iStockphoto

A Center of Excellence is “a team, a shared facility or an entity that provides leadership, best practices, research, support, and training for a focus area,” and they are commonly used in healthcare to focus on specific problems or disciplines. I advocate that they can be used in organizations for artificial intelligence (AI) as well.

What makes AI a strong candidate for a dedicated Center of Excellence is its rapidly expanding role as mission-critical technology in enterprises. Companies are finding that people in many different business units—not just data science or IT—want to be or are already involved with AI.

SEE: Natural language processing: A cheat sheet (TechRepublic)

In some cases, people are bringing in their own AI tools and solutions, but there is a need to orchestrate this buying to avoid waste. In other cases, people are independently developing their own AI and AI budgets, so there is no assurance that accountability for total AI spend or deployment exists.

Together, these factors make a strong argument for an AI Center of Excellence. Such a center would include people from multiple business units, as well as from data science and IT. The goal would be to combine efforts, ideas, and budgets for an integrated and well-orchestrated approach to AI.

Here are six tips for building a strong AI Center of Excellence:

1. A multi-disciplinary staff

Many enterprises have “citizen” scientists in end-user departments. They also have a separate data science or IT staff performing AI work. The Center of Excellence would bring all of these people together into a single, cooperative AI unit.

SEE: Hiring Kit: Computer Research Scientist (TechRepublic Premium)

2. Standardized tools and methods

One of the downsides that occurs when individual business units and IT go off on their own to purchased AI solutions and tools is that the solutions and tools don’t interoperate well with each other. This creates AI silos with data that is very difficult to leverage throughout the company. A central mission of the AI Center of Excellence should be standard solutions and tools so that every project uses a uniform methodology.

3. An IT life cycle approach for AI

AI and big data are now mainstream and mission-critical. The era of pure experimentation is over, and it’s time to get AI projects into useful production.

A good way to do this is to borrow a page from traditional IT by using a project life cycle methodology. The methodology could even be the standard: Define, develop, test, stage, and deploy-and-maintain methods for applications.

SEE: Robotic process automation: A cheat sheet (free PDF) (TechRepublic)

One problem with many of today’s AI projects is that they get stuck in a seemingly endless cycle of develop-test-retest, so they never make it into production. There should be more pressure for AI apps to make it into production so the apps can pay off for the company.

4. A user and IT outreach strategy for the CoE

Because of its focus, a Center of Excellence can quickly become cloistered from the rest of the company. This can create a “silo.” To avoid this, analysts from the Center should be assigned as liaisons to end business units and to IT. Continuous communications between the Center and the rest of the company assure that the Center stays in everyone’s minds, which enables it to become integral to the company.

SEE: 85% of organizations are using AI in deployed applications (TechRepublic)

5. An RTB goal

A Center of Excellence should include some experimentation, but the end goal is always getting a return back to the business (RTB). To accomplish this, AI applications must be deployed in production where they deliver measurable value to operations, revenues, product development, and strategy.

SEE: How to govern AI in your organization: 6 tips (TechRepublic)

6. Operational reviews every six months

When you set up a new function, there are bound to be things that go well and others that need to be further tuned. Initially, the Center of AI Excellence should be reviewed every six months in order to effect tuneups based upon what has been learned.