How to Build an AI Strategy - part 1
Note: this article was originally published in 2020.
2019 may have been the year that Artificial Intelligence (AI) moved into the mainstream consciousness, but 2020 is likely to become the year that AI is introduced into the enterprise portfolio of most organizations in a comprehensive manner. While there are quite a few instances where AI was already creeping in (usually in the context of AI capability being added to tools already present within the enterprise), what's been missing to date in most organizations is a consistent approach for how to 1) exploit these emerging technologies in a coordinated fashion and 2) how to manage introduction of the full spectrum of AI-related capabilities. In other words, what's been missing in most organizations is an Enterprise AI Strategy.
This doesn't mean that AI Strategies haven't already started popping up; they have, for example at the national level or for various branches of the military. Even in those instances, though, the AI Strategies that have been published over the past several years have tended to be extremely high-level, focusing mainly on fairly ambiguous goals and principles but not delving too deep into any specific goals or detailed technology expectations. While the top-level strategic definition is certainly valuable in its own right, it represents more or less the preface for more tangible strategic deliverables.
What is AI, or not?
This may sound like a philosophical question, but it is pertinent to the discussion of strategy given that this is first problem most organizations have been running into when exploring AI adoption (e.g. defining what AI means for them or even what it means for anyone). We can start answering this by first highlighting what it's not:
AI is not one product type or even one product class. AI crosses all potential product domains and all tiers of typical enterprise architecture.
AI is not merely Machine Learning nor a just an ambiguous set of black box algorithms, it is a mix of products, custom solutions, techniques and best practices.
AI is not merely data focused; although data-related capabilities are a particular sweet spot for near-term AI exploitation.
And AI is not yet actually, Artificially Intelligent, at least in the context of Turing Tests or the typical expectations of popular culture and Science Fiction. However, what it can do now is still incredibly valuable and represents a turning point in how we will view all IT capability.
We'll take a look at what it is, or should be, from the perspective of organizations looking to adopt it in the next part of this series.
AI will likely become the most ubiquitous element within every enterprise. At first though, it will become a sort of "integration fabric."
What is Strategy?
Let's review for a minute what tend to be the most common types of strategic deliverables that organizations create in order to help guide their planning efforts. The most traditional strategy document tends to be the high-level declaration of principles with some goals assigned - but typically those aren't not very precise in nature. In large organizations, it's not unusual for various departments to link up to the top-level strategy with their own variation or elaboration of the top-level goals (which tend to become more precise the lower down in the organization they go). In many cases, these top-most strategies tend to resemble Mission Statements, the main difference of course being the Mission Statement typically applies to the organization as a whole and has a very long-term set of business expectations. Strategies tend to have time limits, even if those limits may extend outwards as long as 5 years (anything more than 5 years is probably irrelevant given the current pace of change). No one would expect that a Cyber Security Strategy developed in the year 2000 would be very helpful now (beyond some basic principles which would likely be consistent).
The other type of strategy deliverable is something closer to a detailed plan and typically includes a roadmap and perhaps even some solution design. This second type of strategy is "actionable" in nature and can include the top-level strategy as its introduction. In the military, this second type of document is sometimes referred to as a Concept of Operation or CONOPS as it provides not only the big picture but some concrete guidelines as to how the "transformed" enterprise should look once the concept is deployed (in the context of a unique capability). A detailed Strategy can be referred to as Strategy, a CONOPS or even just a plan; the title doesn't matter so much. The key point is that if an organization wishes to fully exploit something like AI in a comprehensive manner, more thought definitely has to go into the process than what's in the top-level goals and that thought needs to communicated properly.
Actionable Strategy or CONOPS or Plan
Let's look at what goes into the detailed strategy (regardless of what we call it). It's important that we cover a number of generic topics first; these include:
1 - Use Cases: What we wish to accomplish and why (e.g. the key business problems each scenario is addressing).
2 - Assumptions: These must be vetted because sometimes upon reflection, they need to change.
3 - Success Criteria: What we anticipate the benefits will be. This also should include performance criteria.
4 - The Concept: This is a description of how the strategy (and the capability in question), will impact the organization's mission, so it's not meant to be detailed - but rather it is meant to highlight mission or business alignment.
5 - The principles, goals and objectives that comprise the high-level strategy: This is the frame of reference and the basic structure for later planning (as noted previously).
6 - Risk: In this case, both the risk associated with adoption of AI as well as the risk associated with not adopting AI should be included.
This first section might be considered a combination of Business Case and top-level strategy. In the next section of the document, we need to delve deeper into the immediate context; AI, and it ought to cover the following topics:
7 - AI Capability Assessment: What's out there and how it relates other core capabilities.
8 - AI Capability Selection: This involves a targeted description of specific technologies chosen to fulfill capability adoption goals. This does not take the place of specific of product selection activities; those should come later. This activity should however, guide those selection activities by explicitly describing the class of technologies that ought to be included in the roadmap.
9 - Capability Rationale: This highlights how and why specific types of capabilities will be pursued (and can link back to goals and objectives as well as Use Cases).
10 - Capability Roadmap: This is the frame of reference filled in; and ready to elaborate across various project plans. Here we will assign timelines and illustrate inter-dependencies. Timelines aren't exact quite yet, but should be close enough to move forward with detailed planning.
11 - Solution and Operational Vision: This takes the Concept defined earlier a step further and allows it to be illustrated within the context of both the intended technology capability adoption as well as the (existing) target enterprise environment. So, this is in fact the first view of the conceptual architecture with AI included and an explanation of how that will or should impact operational processes.
And perhaps most importantly…
12 - Capability Alignment & Strategic Alignment. This is also important in other similar strategy scenarios (such as Cyber Security or Data Strategy), but AI is particularly ubiquitous in nature. This is where we must necessarily highlight how all of the proposed AI capabilities will or should fit together - both with other key strategies as well as within the to-be architectures. Not all of this can predicted or proscribed in advance; but the more of it that can be captured here, the less the risk will likely be associated with AI adoption.
In the next article of this series, we'll take a closer look at what AI capability represents from an Enterprise IT perspective in 2020 and how a typical organization might go about determining which capabilities ought to go into their first actionable AI strategy.
Copyright 2020, Stephen Lahanas