How can you choose the best AI problems to solve quickly?

Can we stop calling it an AI problem! There isn't such a thing! There is a business problem / statement first that needs to be explored. Then one has to have an understanding of what the outcomes/results need to be and once you have enough information/ data about the problem then you can start to think about a solution which may or may not be AI focused! If people talk like this it will give the C-Suite and others with the understanding that AI can solve everything! Stop! You will frustrate those data teams who are attempting to do a good job and deliver value.

Before getting deep into AI solution design, run a rigorous financial analysis. Many business applications sit on old technology stacks. To take a stab at modernizing these old-school technology stacks with AI, you would have to first upgrade the foundational infrastructure - BIG upfront investment. So when you run your business case for the AI application, you need to factor in the cost of platform upgrade. If the business goal is to reduce customer wait time by 20%, your financial analysis might show that you can get the same P&L impact by hiring more staff than by modernizing the workflow with AI. The AI workflow modernization is a strategic choice but short-term it can make a negative impact on the P&L. So financial analysis first!

What? Wrong Question!? While AI has challenges such as transparency, this Article is phrased incorrectly ... by AI itself? ... which reveals another problem of AI relative to its inaccurate text generation. But this Article is really asking what business problems can be solved by AI applications. The question is backwards? Therefore, Professor AI, I respectfully suggest you either rewrite and rephrase this Article ... or scrap it. ????

Artificial intelligence projects typically aim to address three primary objectives: boosting revenues, reducing costs, or introducing novel services or products. To effectively align with these business goals, companies should prioritize one objective and compile a comprehensive list of AI projects that align with it. For successful implementation, it is crucial to obtain buy-in from key stakeholders such as the CEO, department heads, technology, finance, marketing, sales, HR, vendors, and customers, as they may be impacted by the project's outcome. By fostering a collaborative approach, this multidisciplinary team can ensure that the project is closely monitored and followed up for optimal results.

??From my point of view, to effectively choose the best AI problems to solve quickly, I often start with stakeholder interviews to gather diverse perspectives and ground-level insights. Use these insights to inform a SWOT (Strengths, Weaknesses, Opportunities, Threats) analysis. This will help in understanding the current position in the AI landscape, identifying areas of opportunity, and recognizing potential threats or challenges. ??Finally, I will employ a prioritization matrix to rank potential problems, ensuring they align with the goals. As AI and organizational goals evolve, maintain a feedback loop to ensure continuous alignment and relevance.

??After brainstorming sessions, there are some challenges that I believe need to be addressed: 1. Scope: Defining the scope of the problem is challenging. Too broad, and it becomes unmanageable; too narrow, and it may not be impactful. 2. Technical Feasibility: Not all problems are technically solvable with current AI technologies. 3. Resource Allocation: Time, talent, and financial resources are always limited. 4. Alignment with Business Goal: The problem should align with the organization's strategic objectives. 5. Ethical Considerations: Ethical implications, such as data privacy and fairness, can't be ignored.

When creating new algorithms to solve problems, it's not as simple as coming up with a solution immediately. The first step is to carefully frame the problem and determine what the solution should look like: what are the key outputs? What does success look like? Academics often value theoretical solutions, which not only prove the existence of the solution, but demonstrate how to create it, ensuring problem space conditions are met.

Look at your current business processes and identify pain points where AI could provide value. What manual or repetitive tasks could be automated? Where could insights be generated from data? Then, conduct ideation sessions/design sprints with your team to think creatively about AI applications. Encourage wild ideas and build off each other. Another idea is to analyze user data and feedback reported via customer support queries. These may be great ideas with potential for AI solutions.

Brainstorming is your creative sandbox, but it's easy to get lost. To add structure, use the 'How Might We' technique. Frame each problem as a question starting with 'How might we...?' This opens up problem-solving to collaborative, creative thinking. And remember, prioritize problems that align with your OKRs to stay focused. Don't just use "let's solve it with AI" because you can.

What I often see is that non-technical stakeholders sometimes overlook critical aspects when defining a problem to be solved using ML algorithms. They fail to consider foundational questions: Is there sufficient and relevant data available to tackle the problem effectively? Is the proposed solution feasible given the existing resources and constraints? Equally important is understanding the desired outcome. Is it a dashboard that provides visual insights, a structured table summarizing information, or a real-time API delivering continuous data? We need a holistic approach to address these pieces upfront, to ensure the successful application of ML algorithm.

??I believe we should involve a multidisciplinary team, including business leaders, technical experts, and end-users, in the evaluation process. By involving a diverse group of stakeholders, we can ensure that the problems chosen are technically feasible, align with business objectives, and provide tangible value to the organization.

A further way to prioritise AI opportunities is to weight the different factors driving Impact, Confidence, and Ease according to their relevance in your organisation. For example, factors relating to growth may be weighted higher for an early-stage technology company. Alternatively, factors relating to an improved customer experience may be more appropriate for a mature company looking to build loyalty among an existing customer base. In an AI-squared approach, consider adding a layer of machine learning to your prioritisation process.

Analyze each problem against key criteria like business value, user impact, feasibility, and required resources. You can use the DVF framework (desirability, viability, feasibility) to get you started. Rank the problems based on scores in each area. Problems with high business value and user impact that are feasible to implement should rise to the top as high priority.

When evaluating potential projects, consider both value and viability. Assess potential monetary value through cost reduction, revenue generation or other financial metrics. Also factor in less tangible but important benefits like improved customer satisfaction. At the same time, gauge technical viability by reviewing available data, required integration with other systems, and complexity of the problem space.

Thoroughly assessing the value and feasibility of potential AI problems is crucial for prioritization and success. While frameworks provide structure, accurately evaluating challenges requires acknowledging assessment limitations and tradeoffs between ambition and pragmatism. Overall, systematically investigating factors like impact, confidence and ease offers guidelines but must be paired with flexibility to adapt prioritizations as understanding deepens - ensuring a balanced, learning-focused approach.

??Pilot testing is one of the most useful methods to test assumptions. It involves launching a scaled-down version of a project to assess its viability in a real-world setting. By doing so, it offers a practical way to test assumptions, gather feedback, and identify potential pitfalls before a full-scale rollout. ??This method is particularly beneficial because it minimizes risks; if the pilot fails, the losses are contained, and lessons can be learned without significant repercussions. ?? Although It offers real-world validation but is resource-intensive. It also may not be feasible for all types of projects, especially those that require significant upfront investment.

While running a workshop for a global bank, the leadership wanted to tackle the challenge of loan default. The prevailing assumption was that loan defaults were primarily linked to borrowers' income levels. However, with data analytics, we discovered that a combination of spending habits, past financial behaviors, and geographical data provided a more accurate prediction of loan defaults. Testing assumptions is useful before diving deep into a large scale implementation. If your results confirm the hypothesis, that's good validation too.

In the rapidly evolving world of AI, what worked yesterday might not work today. Hence, it's crucial to adopt a mindset of continuous testing. Leveraging A/B testing or multivariate testing can provide deeper insights into which solutions resonate most with users.

Beyond the technical aspects, consider the user experience. Gather qualitative feedback and fine-tune your solution. This iterative approach not only enhances the chances of success but also fosters a culture of continuous improvement, which is a basic in agile project management.

Testing assumptions through lightweight prototyping, experiments and stakeholder inclusion provides valuable early insights into AI problems. But approaches must leave room for flexibility and future learning, as validation at such an early stage has inherent limits. The goal should be refining problems and catching major risks, not constraining innovation or overpromising perfect solutions based on initial testing alone.

There a few ways to get started: 1. Gather user feedback through surveys, interviews, product analytics. 2. Hold retrospective meetings to discuss learnings as a team - what worked, what didn't, and key insights team has identified. Analyze key results to determine if a pivot in problem focus or approach is needed. Brainstorm ideas for quick proof-of-concepts to be able to iterate & learn. In fact, having a backlog of quick proof of concept ideas to continue trying help you be proactive, rather than reactive.

One other methodology is to use "Task Decomposition" and "Value Stream Mapping". 1. Consider the jobs of your employees and contractors, and break them down into tasks. 2. Examine each commonly done task to see if it's amenable to either assistance (augmentation) or automation using AI tools such as supervised learning or generative AI. 3. Assess the value of doing so. Extend it by layering ?? Value Stream Maps (VSM) ?? on top of the task breakdowns. Task decomposition, at times we end up optimizing locally and shifting the bottlenecks upstream in the process. Keeping the end-to-end process and value of each process step in sight can improve the value derived from AI projects.

Don't test forever. Incorporate your minimally viable product(s) involving into your workflows. You will learn how to make the process better while on-the-go. Be sure to set expectations with stakeholders so there are no doubts regarding confidence in what is being developed.

An iterative, learn-as-you-go approach is well-suited for solving uncertain, multifaceted AI problems by enabling solutions to organically evolve based on empirical testing. However, the process requires discipline - systematically advancing work, managing expectations through clear communications, portraying outcomes candidly to outline limitations upfront. This balances innovation with transparency and accountability, cultivating breakthroughs while maintaining stakeholder buy-in essential for responsible AI_development.

Thoughtfully iterating on learnings from assumption testing and collaboratively deriving deeper insights nurturessolutions that progress equitably and for the benefit of all. Flexibility to redefine problems informed by discoveries optimizes impacts. By appreciating contributions and sharing accumlated wisdom transparently, motivationstoward sustained improvements align with stakeholder priorities.

Exactly. I always tell my students that there isn't any such thing like "AI problem". AI is a solution. It's a mechanism to be implemented on a problem of any domain.

I often hear stakeholders talking about the need to “incorporate AI” into products and solutions. However, AI is simply a tool and not a product feature. Instead, start with identifying the business challenges. Then consider if AI could address them.

I think this article should be removed as it leads to confusion or the question should be rewritten to be something like “What frameworks can be used to efficiently evaluate how AI can help solve user problems”

In the realm of business, the primary objective is not merely to implement AI solutions but to effectively address and resolve pertinent issues. It's imperative to recognize that AI may not always be the optimal choice for every problem. Regardless of the approach chosen, the emphasis should remain on expediting development, minimizing obstacles, and reducing the learning curve, ensuring that the selected technique aligns seamlessly with the overarching goal of problem-solving.

I feel like this is really helpful with understanding how I can utilize AI prompts for my work responsibilities and challenges.