How can you prioritize AI solutions for business problems?

Leading with the company’s business strategy in defining the problems to solve with AI is critical. An example formula for top 3 AI big bets, tied to strategy — customer/client-centric problem with innovative experiences, intelligent automation problem to deliver efficiencies, and a new business opportunity to drive growth. Customer/client-facing capabilities, new business growth, intelligent automation of front/middle/back offices and corporate functions, risk mgmt., marketing, leap-frogging tech, venture investments — all present opportunities to elevate with the power of AI. Narrowing that list to 3-5 big bets with clear linkages to the company strategy is vital for stakeholder commitment and execution towards business outcomes.

Articulating a problem in a SMART manner is really important. The direction of problem solving for businesses should be from defining the problem to finding the right solution. It should not be other way round i.e. finding the application of a particular AI solution in your business stream. Also, once the problem has been articulated, it should be translated for various audiences in different manner. For example, reducing cost or turnover might appeal to higher management or board members, but it could be translated to taking increased number of calls for customer service agents whereas it could be reducing the number of critical bugs for a developer; thus a particular problem statement must be translated for various stakeholders.

Take a customer journey or value stream approach. It’s tempting to simply apply AI to an existing piece of tech. While there is nothing inherently wrong with this, full value should be measured in what the customer (external or internal) needs to accomplish and whether a given intervention makes realizing that need easier or otherwise improves the quality of the outcome. That’s where true ROI can be measured.

Artificial intelligence initiatives should be categorized into short-term (six months) and long-term (one to two years) projects. It is essential to involve key stakeholders such as technology and finance departments, department heads, vendors, and clients in the brainstorming and selection process. The focus should be on identifying areas within the company where AI implementation can drive revenue growth, cost reduction, or enhance customer experience. It is crucial not to artificially create problems for AI to solve when there are no existing issues.

This is the most important step in successful AI solution development, deployment and adoption. And like all successful AI solution applications, it starts with human-centered questions about human pain and unmet needs. So many "AI experts" miss or short-change the user research problem because they are focused on the technology or algorithm development. Key human-centered AI problem definition questions include: 1) What keeps you awake at night? 2) What goals is your business not successfully meeting? 3) How happy or unhappy are your customers and why? (An NPS/voice of customer satisfaction program is an important input AI solution development). 4) What does a dream solution to ease pain/delight look like?

User or stakeholder interviews can be another effective method for getting different views onto the same problem. But instead of asking about problem, focus on what the ideal state would look like to get some inspiration and ideas. Keep in mind though that subject-matter experts typically have a deep understanding of the process/tasks they are responsible, but will likely lack the broader visibility into the entire process and the technical possibilities. (That’s where your skillset can shine…)

Make your choice of AI platforms by understanding how those tools or organizations selling the tools look to evolve their offerings over the longer term. Instead of buying a checkpoint tool that quickly delivers an immediate eye-candy benefit, invest the time to ??Build an AI strategy and decide what AI actually means for your enterprise ??Evaluate platforms that can stick with you for the long haul. ??Consider these while evaluating platforms, ?? Are they built with the right Responsible AI and Safety guardrails? ?? Do they provide adequate security and visibility into your data? ?? Does it provide a level of Explainability for predictions? ?? And can sustain with you on a long term strategic journey with a clear and evolving roadmap?

Adding AI to a sensor device can help to augment its performance significantly. Reducing the cross-sensitivity of measurements (improving specificity), and also it can help to improve sensitivity of measurements as well. It can also be architected to compensate for environmental effects such as changes in temperature, pressure, EMI noise, or other sources of noise and interference that may be in your environement that affect measurement. In addition beyond this, using more advanced AI algorithms, you can start to detect emergence properties of materials that are not directly measureable. For example, detecting contaminants that affect human health such as viruses or bacteria, or heavy metals. There are many possibilities.

1) Don't re-invent the wheel, there are numerous methods and standards for documenting products, features and code requirements. It's just another software project. 2) Think about the data, and how do you collect more so that you can perform training, fit analysis, prediction measurements to be sure that the AI is performing as desired 3) Be prepared to be experimental with safe failures to learn from, write down the experiments and lessons, and results 4) Most importantly, resist the urge to implement the prototype experiments, re-visit, industrialise for production scale. Tempting as it is, retro fitting scale and security will always be hard work.

Using the added value metric is very useful. We need to link value to business bottom line either through: margin improvement, costs reduction, risk mitigation, revenue generation ect.

I would also add looking at the potential ROI can give you a view of whether it not you are barking up the right tree. Another aspect maybe not mentioned explicitly is the fact that you need to understand what processes are going to be streamlined and then one needs to think about the change that needs to be delivered. Implementing a potential AI solution and not thinking about how someone's role will change and what the future process looks like and how you manage this change effectively will send you into tailspin. That way your priority order might change based on other complex and human factors.

A simple risk to reward (or effort to upside) 2x2 can work wonders by (1) making it obvious where a high level of effort or risk isn't worth the limited or questionable ROI, (2) helping to prioritize clear win use cases with low risk + high reward, and (3) allowing your team to have a productive discussion and make more rational decisions about the use cases that fall in between these two extremes (e.g., low reward but also low risk may be worth testing).

Some measures to consider are touchpoint reduction (efficiency) and customer effort score. When it comes to productivity, rebalancing the ratio of coordinating to creating will indicate effectiveness of AI in knowledge work environments.

Often overlooked is that the organization should also consider (1) how they are going to assure results from the selected AI solution. Accepting whatever the AI solution spits out is wreckless and (2) evaluate and manage risk if the AI solution is wrong or just doesn't work as the feasibility and impact suggested

There are two tiers of prioritization: 1. Project-level prioritization: This helps determine which AI project(s) to prioritize first based on business impact vs. effort. I like to use a whiteboard for this to visualize and compare the top projects. This makes it easier to select ones to prioritize first and helps build a roadmap of AI projects to implement across a time horizon. Teams should also consider other factors such as time to market, build vs buy, etc. here to help drive implementation decisions. 2. Model-level prioritization: This can be more quantitative in nature. The feasibility exercise above is a key input here, combined with financial metrics such as the total costs (fixed and variable recurring costs), value, etc.

AI provides the power to innovate at lightning speed. However, the single largest speed breaker is the inability to prioritize AI solutions that add short-term and long-term business value. As we all know, a decision delayed is a decision not made. This is where a visualization tool that compares the feasibility and impact scores of each AI solution can be extremely helpful. It is not about having ideas; it is about prioritization. Otherwise, AI will only make us busier without generating any real business.

The "added value" metric could be used to priorities AI projects. These projects need to be part of the AI strategy and in alignment with the overall business strategy. Others goals that should be considered is building AI talents and targeting low hanging fruits first.

Trustworthiness and explainability is very important for certain use cases especially for clients. Another important concern that businesses have is bias-free AI and AI ethics. Even though OpenAI is a really great advancement but until it depicts reliability, bias-free and hallucination-free results, trusting it for scaled business solutions is difficult.

Quantifying benefits, establishing success metrics, fostering a data-driven culture, and ensuring senior leadership is fully vested in your AI initiatives - all these are crucial when prioritizing AI solutions. To avoid pitfalls; evaluate feasibility, ensure data integrity, mitigate bias, and prioritize ethics and explainability. Remember, proactive planning becomes imperative especially when transitioning pilots to full-scale products, preventing budget overruns and ensuring successful rollout of impactful AI solutions.

The true measure of a data product's worth lies in the ability to provide wisdom as an input to the decision intelligence process, it doesn't matter if the product's category or name is "cool" or not. Focusing on the aesthetics without delivering value or wisdom to business decision will lead to unproductive cycle of being a vapourware product, stranded with costly band aids that fail to scale as an "enterprise product", and not a "hackathon" product.

AI solutions in today's landscape have several applications which can directly as well indirectly impact the bottom line - by increasing the efficiency and effectiveness of processes. For example, an AI powered search engine can help advisors search and respond to client queries in a faster and accurate manner, thus increasing the customer retention rate. Another example of AI application in business processes could be a code generator that can actually have the potential to substitute or even overtake stackoverflow for coders, thus saving time and energy for developers to code faster and efficiently.

My experience shows that businesses should prioritize AI solutions that lower the cost directly or via automation, optimization, or outsourcing and AI solutions that increase perceived value in customers' eyes in a way that more customers are ready to buy or existing customers are ready to spend more. All other AI solutions should be treated as future potential, not a priority.

From my experience, prioritizing AI solutions for business involves pinpointing critical issues that AI can address effectively. Consider factors like feasibility, available data, and potential returns. After choosing a solution, assess technical requirements, data quality, and scalability. Develop a detailed implementation plan, including integration, user engagement, and continuous monitoring for sustained value.