AI applications should be implemented the same way as other production-ready IT systems, rather than as an open-ended research and development activity. A lesson-learned from unsuccessful AI projects is that they often lacked a focus on their deployment within an operational environment.
AI technologies and products are mature, proven automation solutions to business needs as described earlier in this topic. When an organization begins an implementation project, it should have already completed the decision process discussed in the previous section and should have already evaluated AI products and technology for applicability to their specific business needs, i.e., they have reached a decision to proceed with a solution. This section describes how they should implement that decision.
Implementing an AI application into a production environment requires a combination of both internal and external resources.
The project team must include members with both the organization's business knowledge and the technology skills. This is best accomplished with representatives from the internal stakeholders and external technology and product experts.
Internal Resources include representatives from the organization's internal IT, business operations, and HR teams. These members will provide the experience and information needed to make the implemented AI application relevant and useful.
External Experts include implementation consultants, integrators, and vendor product support engineers. These external members will provide the product knowledge and experience needed to efficiently deploy the required capability.
The project team must be responsible and accountable for executing the AI application implementation.
The project team begins by documenting and coordinating the problem statement to ensure that all the stakeholders agree on what the AI application should do.
Identifying the Problem includes creating a detailed description of the perceived problem that the AI application will address or the capability that it will add to existing business processes, service, or product offerings. Part of the process of defining the problem should be an exhaustive identification of the expected outcome, i.e. what is the goal and will the identified goal solve the problem.
It is important that the documented requirement include all the information and expectations that the organization has for the application implementation. It is often the case that the stakeholders will have different ideas on what the solution will do and how it will be integrated into their business processes. The project team must assemble and socialize these ideas into a consistent, coherent, and complete description that can be agreed to by all the stakeholders and provided to the team members who will be acquiring, configuring, installing, and testing the solution to verify that it meets this requirement.
The project team prepares the project plan documenting the project objectives, schedule, deliverables, resources, responsibilities, and communications.
Objectives includes a clear concise and measurable statement of what the project will deliver, what it will include, and when it will be completed. It should not be open-ended.
Schedule includes the timeline of events and deliverables from the project initiation to the project completion. They must be specific calendar dates and measurable enough to allow progress to be assessed, reported, and managed.
Deliverables should be identified including intermediate artifacts and final products produced during the implementation.
Resources must include a reasonable commitment for the project team to accomplish the objective, including team members, budget, and infrastructure to accomplish their tasking. Members of the team must enjoy the trust of management and their coworkers.
Responsibilities include the team member's assignments including coordination and communications. They also include support committments of the organization to the project team.
Communications are an important part of the project plan and should include outreach engagements with affected staff and provide regular transparent communications to the stakeholders. It is important that everyone understands the business case for the application and that misunderstandings are avoided during implementation.
The specification translates the agreed-upon requirement into a desired solution including a description of the AI application, its functions, inputs, outputs, and business logic. The specification may also identify the intended AI tools and products. It includes performance metrics where these are important to the requirement.
The project team coordinates the specification with the stakeholders to ensure that it accurately and completely describes the desired application capability.
Once agreement has been reached on the system specification, the project team including its external consultants and vendors engage to design and develop the solution including a combination of commercial products, services, and internal data sources and operations. As part of this the project team selects and acquires the products that best fit the specified solution.
The project team prepares a test plan based upon the solution specification, and in coordination with selected stakeholders verify the developed design and validate its functionality. Testing includes component development testing, integration testing, performance testing, operational testing, and others specific to each application.
The project team plans and coordinates the deployment of the AI application solution. They also prepare the deployment documentation including user manuals, support documents, and training materials. During the deployment they provide transition support to the system users, including tiger teams for unanticipated problems.