The AI solution development life cycle requires very different activities. INFORMS created a Job Task Analysis through surveys of practitioners for its Certified Analytics Professional (CAP) program. The result of this study defined seven performance domains for analytics practitioners, which I view as key components of the AI solution development life cycle. In the diagram below, I group the seven domains into three groups, based on what I believe is the kind of expertise required to perform tasks in that domain.
In this post, I discuss the job tasks and I offer techniques for the first domain—business problem framing—which entails understanding the business problem and the business value in solving that problem, and determining whether the problem is amenable to an analytic solution. This discovery activity relies on the expertise usually demonstrated by management consultants. Getting it right is necessary for end-to-end solution success, which is ultimately attained by only an estimated 15-20 percent of AI projects.
1. The first step in business problem framing is obtaining the problem statement and usability requirements. We define the current situation—how decisions are made today. We describe the vision of the ideal situation and list the actions that will help bridge the gap from the current to the ideal.
It is important to establish agreement on how success of the AI solution will be measured. If possible, we work with or collect existing baseline measurements of cost, profits, service rates, or other types of efficiency metrics for comparison. We seek to determine the cost of doing nothing and remaining with the current situation: is there a penalty to the business or organization for failing to change?
2. The second step is identifying the stakeholders. We specify who will use the solution within and potentially outside the organization, and who will be impacted by it. We work to understand the different organizational roles of employees, who could be as follows: the end users that will be using the graphical user interface; professionals in IT who always want to understand how the solution is going to fit within their infrastructure; professionals who work with databases; and user interface designers. Keep in mind that different stakeholders will all likely want to say, “If you're going to put a new solution in front of us, we want to have a vote at the table.” Additionally, we assess what business processes, even those tangentially related to the process the solution addresses, will be affected by the potential solution and how.
3. The third step is to rigorously evaluate the sets of potential business benefits. At Princeton Consultants, this evaluation may include a cost benefit analysis (CBA), a structured way to assess costs and benefits. We estimate, in the absence of AI, costs that will be incurred and benefits of improvements. We then present the benefits of investing in a solution, which accounts for the costs of buying software, implementing a solution, and other necessary resources.
4. The fourth step is to secure stakeholder agreement to “sign up” for the benefits based on the costs. The parties should agree about the problem to be solved and that the team can solve it.
5. The fifth step relies on principles and practices of change management, an approach to transition individuals, teams, and organizations to a desired future state.
Building and deploying an AI solution often leads to change in the way people work. Humans by nature resist change. The above graphic shows one way to build commitment into a desired change; I sourced it from this blog post.
Our goal is to take people through the journey from preparing for the change, accepting the change, and then committing to it. Failure at one of these points results in unawareness and confusion or, worse, rejection and termination. For example, if all the stakeholders are not engaged (as prescribed in the previous step), some employees may ask, “How come we didn't know about this project?” They are unaware. If the stakeholders’ understanding is not attained, they will hold a negative perception.
My colleague, Jack Levis, who recently retired from UPS, led an extraordinary project called ORION—winner of the INFORMS Franz Edelman Award in 2016 (enjoy this video about the project)--which routes the 55,000 delivery trucks for UPS every day. According to Jack, 70 percent of that project was change management! If you are developing and deploying an AI solution, partnering with change management experts is advisable.
In conclusion, effectively conducting business problem framing for an AI solution will greatly increase the likelihood of success. As you master this domain and subsequent domains of the solution development life cycle, you will likely find yourself working collaboratively, on-site with project sponsors to design and deploy not just a system, but a whole new—and better!—way of working.
To discuss this topic with Irv and colleagues, email us to schedule a call.