Making an organization AI-enabled is certainly challenging. As you build the processes and structures outlined in Chapter 5: Cultivating Data and Technology, parallel efforts should focus on applying these structures to new use cases.
A use case is a specific challenge or opportunity that AI may solve. Start small with a single use case limited in scope. Eventually, as business needs and agency infrastructure grow, the number of AI use cases will grow.
Your goal should be to create an evergreen collection of AI use cases where business units organically identify, evaluate, and select the best to proceed to proof of concept, pilot, and production. This ideal state may be far in the future for some agencies, but all agencies can start to explore AI applications to current business challenges right now.
How to select which AI use cases to pursue?
To identify use cases, consider the following:
Focus on agency mission
- Problems directly tied to operational or strategic priorities
- Problems connected to KPIs with significant gaps to their targets
Find the right data
- Areas that are rich with accessible data
- Areas with under-explored data
Identify a champion
- AI needs executive sponsorship to be successful
- Align mission, data, IT, and end-user needs
Framing the problem for an AI project
As with non-AI projects, framing the problem is critical. To ensure success, follow these steps:
Interview users for an AI project. Too often, leadership, program managers, and algorithm developers create solutions for problems that don’t exist, or solutions that are simply not aligned on the right problem. By interviewing an application’s actual users, AI practitioners and implementers can understand the needs and intricacies of the exact audience they’re working to help.
User research ought to be iterative and consistent throughout development. Gathering user feedback is essential not just for the experience the applications would provide, but also how AI ends up being applied.
Whether you develop a potential AI solution internally as a prototype or procure it through the private-sector, you have to know what’s currently available in the market to design and implement an AI solution successfully.
To find the right AI solution or vendor, leaders must know the field from meeting with companies in person, calling people, and doing market research. This vital systematic approach is improved with professional company assessors. Ultimately, the decision to buy or build AI is informed by what is available in the market and the team’s internal capabilities.
Not every identified use case is a great fit to go forward as a project. Technical, data, organizational, and staffing challenges may impede its progress.
Ultimately, prioritization comes down to three factors:
- Problem size
- Impact of model output on the organization’s mission priority and business metrics
- Level of effort required to acquire, ingest, and wrangle data
- Data’s quality and quantity
- Complexity of analytics and model required
- Cost of development and/or implementation
- Alignment with team/unit mission and priorities
- Capacity to build, buy, and/or adopt into existing environment
- Ability to maintain the requisite budget and staff