AI is increasingly being adopted to help optimise all kinds of systems, from sophisticated devices with on-board AI capabilities to entire multi-stakeholder processes. Given the relatively nascent (but fast developing) state of the market, in this report we seek to stand back and look at the overall process of defining, deploying, and maintaining enterprise AI solutions.
AI is increasingly being adopted to help optimise all kinds of systems, from sophisticated devices with on-board AI capabilities to entire multi-stakeholder processes. Given the relatively nascent (but fast developing) state of the market, in this report we seek to stand back and look at the overall process of defining, deploying, and maintaining enterprise AI solutions.
Specific focus areas include a range of project stages:
Definition, with significant impact on all subsequent stages, ranging from the information to be collected to the analytics to be performed, to the expected lifecycle of the project, and options that there may be to revise and enhance any approach at a future date.
Information Gathering, including consideration of the required timeliness of the outputs of the AI solution and the sourcing and cleaning of different input data. Often more latitude can be afforded to systems intended to feedback to human operators.
Preparation for Analysis, including the population of information into an appropriate environment for analysis and support processes such as contextualisation and augmentation.
Analysis. Is the aim simply to alert a human manager when certain conditions are identified, or should the AI solution also identify any necessary actions to be taken and initiate those actions autonomously? This has significant implications for security and the governance of the AI solution itself.
Maintenance. AI solutions will inevitably evolve over time, either as human operators refine their understanding and increase the sophistication of their analyses, or as more autonomous solutions self-evolve. In addition, alerts or suggestions generated by an AI-enabled system will often need to be reviewed by humans before adjustments can be made to operational system settings and processes.
Maintenance (continued). There is another aspect of maintenance that applies specifically to AI-enabled assets, which is the ongoing management of on-board AI over time and also across product generations.
Closing the loop. Ultimately, any off-board AI-enabled solution will ideally ‘close the loop’ between the outputs of analyses and optimising inputs to related machinery. We do not discuss this in detail in this report, but it is worth highlighting that this concept introduces many significant challenges, including security, delegated authority and oversight.
