No business problem is so hard that artificial intelligence won’t be suggested as a way to solve it. And it’s the job of chief information officers and information technology leaders to figure out if it can.

That’s what it’s like in the technology world: Every generation of technology results in hype, and AI is no exception. If enterprises fall victim to such hype, they may incorrectly apply AI technologies to existing initiatives, or they may offer AI as a solution to a problem that it cannot — or should not — solve.

AI will not always be the right technology for every digital transformation project. To that end, Gartner has identified seven questions that IT leaders can ask to determine if a project is a fit for AI. The first three questions must be answered “yes,” or there’s no point in trying to use AI. The last four would ideally be answered “yes” in the best-case scenarios.

First, here are the three questions that must be answered “yes” for a project to be a fit for AI:

1. Is authoritative data present? Machine learning, which is at the heart of most AI projects, demands that the data it analyzes be accurate and well-defined. Most business-oriented executives see AI as capable of delivering results that improve as the applications reinterpret the data with refined analyses. To do that, the data must be dimensionally complete, it must be of sufficient volume to discover meaningful patterns even in small distinctions, and it must be substantially correct.
2. Are labeled outcomes in place? Teaching a rat to navigate a maze more swiftly is not possible unless you are running a stopwatch and you know which times are the best. When the rat runs the maze, it needs to know if it did better than last time, so it can decide which tricks to try (or not try) again. Computers are akin to extremely stupid but tireless rats. Organizations often have lots of data, but they don’t always know which data records show a good outcome was achieved in one or more dimensions. Labeled outcomes must be present so the AI system knows if it achieved the desired goal and can adjust its algorithms accordingly.
3. Is the project popular with the sponsor? No project is ever completely successful, but nor is any ever a total failure — as long as trying to achieve its desired outcomes teaches lessons about the subject that are worth knowing. However, no project is worth doing if the person who will evaluate its success doesn’t believe that success is worth anything. It’s essential to look for projects that will thrill sponsors if they succeed and that sponsors will be glad to discuss even in the case of failure. Failing in a way that teaches an organization new skills or new details about data is a success.

If these three questions are a “yes” for the project being evaluated, then CIOs and IT leaders can move on to the following four questions. These are best answered “yes” for a project to be a fit for AI:

1. Will this use case develop key skills that you can transfer to other projects and initiatives in the organization? Skills might be specific, such as developing knowledge about critical machine learning techniques or data preparation disciplines. They might also be more general, such as bringing together data scientists, business analysts and IT specialists. A project that enhances the skills of the people involved is automatically one with a stronger return on investment.
2. Can you build on the data that you prepare for this project through using it in subsequent projects? When you clean a room in the house to prepare for visitors, that room becomes more attractive for other purposes as well. Everyone likes to use the big table when someone clears away the dirty dishes. Focus on projects that lead to the availability of data that many future projects can employ.
3. Is the direction that this use case will take you helpful for future use cases? Gartner recommends that organizations select AI use cases not only for their individual value or return on investment, but also for their relevance to a long-term strategy that will deliver multiple use cases of value. For example, if your long-term goal is to send a customer a recommended set of industrial equipment maintenance tasks, the first use case might be to identify what lubricant to apply following a given set of weather conditions.
4. Will the results of the project be specifically measurable? Among the factors that align to an organization’s greater AI maturity is its decision to track the financial or risk impacts of the projects it pursues. Picking projects that will give the technologists access to the vocabulary that sponsors and executives speak is a sure way to help them understand the merit of the methods used and the lessons that they teach.

Projects that meet all seven of these criteria are the most attractive for AI. Organizations that choose their projects carefully, considering their present value and payback — but not at the expense of their strategic positioning for the long term — will be the most successful in their AI implementations.

Whit Andrews is a distinguished research vice president at Gartner Inc., covering the organizational impacts, use cases and business opportunities for AI. He wrote this article for SiliconANGLE. Gartner analysts will provide additional insights on AI and other data and analytics trends at the Gartner Data & Analytics Summits 2021 taking place virtually May 4-6 in the Americas.

Image: geralt/Pixabay