Determine dimensions of the problem

Why does this matter?

It is essential to understand a prioritized problem before defining AI product requirements. It can save time and resources. Your organization should consider adopting an AI product as a component of a solution only if it addresses the root cause of the problem.

• Don’t be fooled. Analyzing the dimensions of a problem can avoid falling into AI hype or technological solutionism.¹
• Think in terms of a solution with components. AI products rarely ever solve the root cause of a prioritized problem. An AI model may be a component of the solution, but the model alone is almost never the entire solution. Your organization needs to identify the other non-technical components of the solution.

How to do this?

Step 1: Empower an independent team that dissects the problem

• Prior efforts to address the problem may have failed and the problem may be a source of tension between business units and individuals.
• Assign an independent team without strong alliances to any professional group or business unit to address the problem.
• Ensure that the team reports to an executive leader, while business unit leaders have minimal influence over how the team operates. 
• Proactively work against existing power structures to surface the root cause of the problem. 

Step 2: Form a cross-functional team around the frontline worker who identified the problem

• Ensure that the team includes frontline clinicians from different settings and professional backgrounds. 
• Have senior leaders involved in problem prioritization communicate with frontline workers across business units to join the cross-functional team. 
• Recruit business unit leaders who control resources in the context where the problem is felt most acutely to join the cross-functional team.
• Inform everyone on the team about the problem context and the confidential nature of conversations related to the problem. Employees need to feel safe dissecting the problem. 

“Ideally, the stakeholders that are involved in decision making should be the corporate management or governance body, but also representatives of the clinical team, in particular, those with specific skills in identifying both valuable targets for procurement, and challenges, including bias and unfairness in algorithm deployment.”

AI Fairness and Safety Expert

Step 3: Develop a problem statement

• Discuss the problem with the cross-functional team and make a clear problem statement.
• Seek input from the team members asynchronously and privately to ensure that diverse perspectives contribute to the problem statement.
• The problem statement should not anchor on any particular solution. For example, an ineffective problem statement would be something like: “There is no AI to identify X”.

Step 4: Complete 5 Whys with individual team members

• Five Whys is a method for quickly arriving at the root cause of a problem.²
• Starting with the problem statement, complete the following steps:
  • Reason 1: Answer “Why did the problem occur?”
  • Reason 2: Answer “Why did (reason 1) occur?”
  • Reason 3: Answer “Why did (reason 2) occur?”
  • Reason 4: Answer “Why did (reason 3) occur?”
  • Reason 5: Answer “Why did (reason 4) occur?”
• Conduct this activity with each team member separately to ensure all perspectives are accounted for. If a new affected stakeholder is identified through the Five Whys activity, recruit that perspective to also participate.

“[we have] two technologies focused on the availability of home blood pressure readings, and the correct treatment decision as the countermeasure. So those are the root causes that we were targeting. But, when we did our process map, what we discovered is that those two things weren’t that big of an issue”

Clinical Lead

Step 5: Map all identified problem reasons to the root causes of the problem

• Identify themes that emerged from Five Whys as reasons for the problem. Be aware that different team members will use different language or terminology to describe similar phenomena. 
• Map each theme to a root cause of the problem. 

Step 6: Iterate on the root causes of the problem with the cross-functional team

• Share the identified root causes of the problem with the cross-functional team and gather feedback. 
• Iterate on the root causes until there is a common mental model for why the problem happens.

References

1.  “Behind every seemingly technical problem is actually a human problem waiting to be found.” – Eric Reis
   Full video available on Harvard Business Review: https://hbr.org/2012/02/the-5-whys
2. Five Whys is a root cause analysis tool that is part of the Toyota Production System. It is commonly used in lean management. It assumes that the initially identified problem is not the root cause and allows for your organization to identify multiple sources of the problem. This method is particularly relevant to health care, where processes are complex and involve multiple social and technical interactions.