Considerations for Picking the Right AI Use Cases for Your AI Portfolio

From boardrooms and the C-suite to the shop floor, everyone is looking for opportunities to integrate AI into their operations. However, identifying the "right" AI use cases to work on is not a trivial task. The wrong use cases consume budget, resources, and time, while never delivering on the outcomes expected, and they create risk and liability along the way. But an AI use case that is “wrong” for one company may be “right” for another.

Based on our years of experience helping clients with their innovation and transformation efforts, we’ve developed the AI use case decision framework (Figure One).

Click to expand full resolution

We treat AI as an investment, rather than a technology. We use this perspective to build an AI portfolio of investments (use cases and projects). This involves two steps: identifying a pipeline of the right AI use cases, and deciding which of these to fund. Most companies will have more AI project requests than the budget allows for. Deciding which of the “right” use cases gets the funding is often a process fraught with limited data, subjectivity, organizational politics and inconsistency. Key considerations are described in this article.

Building the AI portfolio starts with identifying the “right” projects by considering them through three decision gates. The gates are not a simple yes or no. In practice they work more like a sliding scale. Some opportunities create more value than others. Some are far easier to execute than others. Part of the job is understanding where an opportunity actually sits on that scale, not just whether it clears some arbitrary bar. The gates help to enforce a consistent and disciplined approach for every AI use case being considered. Once an opportunity clears all three gates, it is placed in the pipeline as a candidate for funding consideration.

Gate one asks a straightforward question: should we do this? Does the opportunity produce value someone can point to, and does it line up with where the business is already trying to go? An AI opportunity is wrong if nobody in the business actually wants the outcome it produces.

Gate two asks a different question: could we actually do this? Do you actually have what you need to make the use case work? Do you have access to the right data, and if you do, is it actually clean and reliable, or does it just look complete from a distance? Do you have the technical means and supporting infrastructure to do this, or is there a quiet list of systems that do not talk to each other standing in the way? Do you have people with the skills to build, run, and maintain it?

Gate three asks: can you actually scale and operationalize this? An opportunity that works cleanly in a controlled pilot does not automatically translate into something that can run day after day, across a full team or a complex plant floor environment. This gate forces the question directly: can this actually be deployed for operational use? Is the team ready to run it once it is live? What systems and operations must be updated? And what risk, planned or unplanned, are you taking on once it is operating at scale?

Once the pipeline of the “right” AI use cases is built, the question becomes: which ones should be funded? This is where the investment perspective becomes critical, because now you are comparing the AI use cases against each other in the pipeline.This is a more complex task than stage one activities.

This second stage starts with examining the diverse range of AI use cases through common perspectives - impact, risk, friction, complexity, synergy and strategic importance. The use cases are then mapped into a common “language” or categories to facilitate analysis, such as low risk, quick-wins, strategic, high impact, high complexity, innovative and other common measures.

Once classified, step six focuses on analyzing the AI use cases to prioritize them for funding and execution. The AI opportunities are analyzed together as a set: which are high impact for low effort, which share infrastructure and get cheaper done together, and which are simply non-negotiable regardless of the rating? Because each use case is different, comparing them against each other highlights where they differ, where they are similar and how they rank against key objective measures.

Step seven builds the AI portfolio. However, this does not mean finding the safest and easiest opportunities to fund and execute.The objectives of the AI project depends on what a given business is trying to accomplish, how much risk leadership is willing to carry this year, and how much internal capacity exists to execute. For example, a company in a highly competitive industry or market may wish to build a portfolio of innovative AI use cases. A company in a highly regulated and stable industry may wish to build a portfolio of lower risk use cases. A company just starting out with AI may wish to focus on quick wins and low risk projects.

A portfolio built entirely of safe, easy wins keeps the lights on but rarely moves a business forward. Companies also need room for the strategic opportunities, the ones that are genuinely hard, may not pay off right away, and carry real risk, but that the business cannot afford to ignore. A well-built AI portfolio ends up with a deliberate mix of these, chosen for complexity, value, risk tolerance, and delivery difficulty in different proportions, and what that mix should look like is not universal.

The result, step eight in the framework, is a list of AI use cases in the portfolio for funding that aligns to those goals of the business.

Getting this right takes discipline at both stages. A strong pipeline of well-vetted AI use cases does not help if there is no clear, consistent way to decide which of them get funded. And a rigorous funding process cannot make up for a pipeline full of opportunities that were never properly qualified in the first place. Each stage depends on the other.

Picking the “right” AI use cases to work on takes discipline at both stages. A strong pipeline of well-vetted AI use cases does not help if there is no clear, consistent way to decide which of them get funded. And a rigorous funding process cannot make up for a pipeline full of opportunities that were never properly qualified in the first place. Each stage depends on the other.