Strategic AI Workload Prioritization

This blog is an excerpt from the co-authored whitepaper, ‘Crafting your Future-Ready Enterprise AI Strategy’ aiming to support organizations on their journey in the Age of AI.

Workloads

Our Workloads pillar gets to what’s on the mind of most folks when they think about artificial intelligence: How will we use AI to solve ​
real-world challenges?

“Workload” is not a throwaway word. It is rather a specific term that we use precisely because I value its imprecision. App-centric people speak in terms of apps, integration-centric people speak in terms of integrations, etc.

Workloads cover it all. They are, simply put, a collection of one or more apps, chatbots, visualizations, integrations, data models, etc. working towards the same end. “Workload” is essentially the combination of the front-end and back-end required to produce an AI-driven response or action.​

This pillar broadly addresses three topics:​

• Identifying and road mapping the best candidate workloads for development via Workload Prioritization;​

• Understanding the spectrum of different workloads through which AI can be used, and why balancing your portfolio across Incremental AI, Extensible AI, and Differential AI is so important;​

• Enabling the organization’s Power Users - also called “Citizen Developers” or “Communities of Practice” - to extend and even develop AI capabilities for themselves.​

Workload Prioritization​

We often say that “We’re not concerned with one app. We’re interested in one thousand workloads.” Maximizing the use of AI throughout an organization, truly weaving it into the culture and ways of doing business. That’s how we achieve real value - how we maximize return on investment - in artificial intelligence.​

Our goal in Workload Prioritization (also known as “workload road mapping” or “app rationalization”, depending on which circle you’re running in) is to create a prioritized roadmap of specific workloads to be modernized with an infusion of AI or built anew to solve an emerging problem or a challenge whose solution may have been out of reach without AI. This prioritization is an indispensable part of an organization's ongoing AI journey. Prioritization results in a workload roadmap, a backlog of workloads that are candidates for development with AI capabilities. It allows the organization to project AI’s business value over time, and is a core driver of return on investment (ROI).​

There are many techniques and patterns through which you might prioritize and re-prioritize workloads, including:​

• Alignment of the workload to the desired guiding principles or outcomes defined in your executive vision;​

• Legacy location in that where the workload lives today, e.g. evacuating the data center on the third floor of your headquarters building may present an excellent opportunity to rebuild previously on-premise workloads to be “AI native”;​

• Legacy technologies, similar to “legacy location” above, but related to technologies you wish to sunset rather than locations you wish to evacuate;​

• Telemetry such as monthly or daily active users, last active use, data volume (remember to consider both structured and unstructured);​

• Security or compliance risks that exist in whatever systems or processes you are currently employing relative to a given workload (i.e. is our current model too risky for the organization?);​

• Target technologies in that you may highly prioritize workloads that are targeted for the same or similar technologies, i.e. if you’ve just made a significant capacity investment in a particular AI development tool on which you’d like to maximize the return;​

• Qualitative Assessment wherein we compare each of a series of workloads’ relative impact or value to the organization with the complexity (technical, organizational, political, budgetary) of implementation to determine which workloads are big wins, quick wins, nice to have, or wastes of time.​

There are likely more considerations, but these are our favorites. “Qualitative Assessment”, in particular, offers a relatively quick and effective way to elicit real business challenges from colleagues and to then prioritize them so that we can make investment decisions as to which AI workloads we will invest in next.​

The qualitative method for workload prioritization asks stakeholders to make qualitative judgements as to the relative impact or business value and complexity of implementation for each workload.​

The roadmap is pivotal to ensuring that there is a flow of workloads inbound to the AI and data technologies in which we’ve invested, and that the organization is focusing its development resources on the workloads that will provide the highest value once deployed. In short, a healthy roadmap allows the organization to:​

• Continuously decrease marginal costs per workload whilst increasing overall return on investment;​

• Provide a trajectory of anticipated usage so that capacity can be tapered up over time;​

• Achieve quick wins and big wins to justify early and long-term investment;​

• Focus development resources on the most valuable workloads;​

• Avoid costly development quagmires by de-prioritizing the least valuable or riskiest workloads.​

We must therefore work directly with business stakeholders, users, and IT to identify, prioritize, and categorize candidate workloads and to create a roadmap for building them. These candidate workloads should be prioritized and re-prioritized on an ongoing basis so that we remain focused on the most important next efforts.​

Finally, it is useful to categorize each AI workload on your roadmap as incremental, extensible, or differential.

There is some risk here that specific AI workloads won’t work as their creators intended. No, we don’t mean that the AI will turn evil and obliterate its makers, rather that because we are (still) in the early days and because the technology’s true capabilities are not entirely known, organizations may find that a particular AI-driven workload just doesn’t produce the hoped-for results either because it lacks the data it needs or because the technology just isn’t ready for what it’s being asked to do. Wise organizations can therefore hedge this risk in several ways:​

• Balance your roadmap of AI workloads in development between incremental, extensible, and differential. Think of this as a spectrum where incremental incurs the least risk to success and differential incurs higher risk;​

• Relatedly, balance your roadmap of AI workloads in development between those that are medium-to-high impact but low complexity, as well as those that are high impact and high complexity. In other words, take a few moonshots, but balance the risk of your moonshots not panning out by also investing in less complex yet still valuable workloads. Distribute risk across your workload roadmap;​

• Make measured, incremental investment, and evaluate progress regularly. This is not the space for lengthy IT projects with big bang go-lives at the end. Instead, develop these workloads such that each sprint makes observable progress, preferably solving specific development challenges such that you either (a) build confidence in the solution as you move towards a minimum viable product, or (b) can quickly recognize when it’s time to pull the plug and invest elsewhere;​

• Augment your AI development with expertise, particularly around specific types of workloads where others have made good progress. Partner with others outside of your own firm, learn from the successes and failures of others, and don’t be afraid to learn and share;​

• Educate stakeholders and investors; set their expectations. AI can seem magical, but it is not magic. “Quick wins” are not particularly attainable if you’ve not done the hard work upfront around your data platform, and it’s still difficult for almost everyone to prognosticate about outcomes in such an unknown space. ​

Finally, many organizations will want to conduct a business value assessment (BVA) to make a business case based on specific goals and expected outcomes of their Power Platform adoption generally, and their development efforts for important and critical workloads specifically. Such assessments allow mature organizations to determine and make sound financial choices around considerations such as return on investment (ROI), net present value (NPV), and other industry-specific factors.​