In recent discussions the importance of AI solution engineering specialists, business teams, and domain experts working together has been a topic of increasing focus. Representatives from organizations involved with applying AI into their operations or applying AI to assist with decision-support often mentioned the difficulties of working with their AI colleagues and/or third-party partners.
The discussions by organization domain experts center around data and the AI Solution Engineering Team fully understanding the use case/business requirements – and the data - for an AI solution. The AI Engineering Team states that they believe a rigorous approach to AI data science and machine learning principles are followed to meet scenario requirements and expectations for accuracy and reliability.
But relevant issues were identified around project execution from the above discussions:
1) While an “agile-like” approach was followed, and user reviews were completed and documented, changes were not made to the AI Solution because the data models and/or algorithms involved were deemed adequate by the AI Solution Engineering Team
2) The resulting AI Solution only “demonstrated” the capabilities of AI to support decision-making or business process automation but were inadequate to be put into a live “production” environment not meeting business sponsor expectations
3) It was not discovered until after the AI project was initiated that data that was needed to satisfy the use case was incomplete, inaccurate, or not available at all
4) The timing and resources for data clean-up efforts (e.g., incomplete but usable data sets) were not added to the AI project plan, and the AI solution was delivered utilizing this incomplete data impacting the reliability of its output
5) Data that was identified as not available did not initiate a change to the scope, expected capabilities, or project schedule to deliver the AI solution, and was simply “created” without project management oversight’s involvement, nor was an approach developed to acquire the needed data
The three themes from these discussions that emerged were: 1) There was a lack of necessary knowledge , perspective and communications on both sides of the project to ensure mutual success; 2) The impact of data on the project’s success, and 3) Differences in the business and AI team’s project experience level led to a general sense of performance dissatisfaction between the two groups for the investments of time and resources committed relative to the expected AI solution capabilities actually delivered.
Conclusions to draw include:
a) Never lessen the importance of data in an AI solution. Before AI solution engineers begin, explore the availability of data needed to assess it for meeting use case requirements
b) Well-defined and scoped, pragmatic use case scenarios and requirements are critical to success when beginning the AI journey
c) Because AI solution engineering and development is new – as is the increased, rapid adoption of AI - there is a need for strong business and AI project management and sponsor oversight. Allow for changes in scope and delivery dates considering the business and AI engineering team’s experience levels