Everyone Talks About AI: Here’s How to Actually Build a Strategy That Works

Everyone is talking about AI: how it will be a game-changer, how the degree to which it is embraced will define a country’s economic competitiveness, and how it promises to transform our lives in ways that we cannot yet imagine. Yet, notwithstanding its virtues and promises, nobody discusses what AI looks like in practice and how to develop and implement an AI strategy. If you are in a position of responsibility for an enterprise and want to understand what a successful AI strategy looks like, then please read on. This article will cover the essential principles of an AI-driven digital strategy, what you need to consider, and how to know when a strategy is the right one for you and your organization. 

Let us first define a few key terms, the first and most important one being, "Artificial Intelligence (AI)," itself. Since the boom in popularity of ChatGPT and large language models (LLMs), the term, “AI” has been used synonymously with “LLM,” but the two terms should not be confused. The best way to think of AI is as human knowledge that is in usable digital form for the purposes of extending or otherwise enhancing human perception, human thought and reasoning, and/or human action. Hence, language understanding systems, robotics, planning and scheduling systems, and advanced automation systems are all examples of AI systems. 

A second key term is “Digital Strategy.” A digital strategy serves two purposes. The first is to ensure that an enterprise-wide process is in a digital form so that every step of the process can be performed with 100% accuracy, repeatability, reliability, and at arbitrary scale. As an example, think of the degree of automation for chip fabrication, where the best chip foundries are those where no human comes in contact with the silicon wafers. If you are a manufacturer and if your manufacturing processes were 100% digital, it would not matter where they were hosted – they could be on the Moon or onboard a ship in international waters – as long as there is electrical power, a way of supplying raw materials, storing finished products, and then shipping them. The second purpose of a digital strategy is explained by the next key term, “AI-driven Digital Strategy.” 

An AI-driven digital strategy is a digital strategy that cost-effectively facilitates the capture and maintenance of human and/or process-specific knowledge so that specialized AI algorithms can reason about it. 

An example of an AI-driven digital strategy with which I am familiar is in the steel casting and welding inspection domain. Steel casting and welding inspection is a manual process primarily because the reference standards are in non-digital film images. The clients then accept shipments based on non-digital film, inspection archiving in non-digital film, and consequently inspections in non-digital film. A first step in the strategy was to petition the inspection standards body to translate and release their standards in digital format, which they did in experimental form. It then became possible for AI algorithms to automatically perform the inspections that humans do. The industry still needs to converge on standards and supporting software tools, but the AI-driven digital strategy for casting and welding inspection is poised to transform from a costly human-only process that can only randomly sample batches of castings and parts to inspect, to a uniformly consistent, cost-effective and scalable digital process that is made possible by AI vision recognition algorithms. 

With the above three key terms defined, this is when the reader should be advised of a very big CAUTION! The successful deployment of an AI system within an enterprise is neither a given nor inevitable. 

Any AI system is, by its very nature, a software-intensive system, and every software-intensive system has a high risk of failure if its development is not led by people who understand and mitigate against that risk. 

The reasons for this are known, well-studied, and can be summarized as: many software engineers cave to the business pressures of quickly producing the results that the paying stakeholder wants, at the expense of all the other stakeholders, processes, and technologies that must interface with the system. There are many “war stories” of how Organization X commissioned software developers to design the next big enterprise-wide software system. The SW developers did an excellent job, but from the perspective of the stakeholder who commissioned them, and neglected to consider the perspectives of the other types of users of the system. 

No need to worry: if a SW developer understands their software engineering architecture theory – and many successful professionals, indeed, do understand it – then this risk will be small. 

The point is, there are known design flaws which any type of software system can suffer – including an AI system – hence the guidance for “good AI” strongly overlaps with that of “good software architecture design.” 

Now let us discuss the fun part: the characteristics of a robust AI system that will really do something good for the enterprise, be cost-effective, and make the enterprise economically competitive. They are: 

1. Alignment with the enterprise’s business objectives 

2. Consistent alignment with the technical trade-off space 

3. Synergistic integration with the enterprise’s existing processes, technologies, and personnel roles and skills 

4. Cost-effective capture and maintenance of human and/or process-specific knowledge. 

The first robust AI system characteristic should not be a surprise nor require much thought, but accidents happen. This is often because in its haste to acquire new capabilities through an AI system, the enterprise might not realize that they risk sacrificing other capabilities that they presupposed to be unalterable. As an example, think of the ill-fated Titanic: “small” redesigns to accommodate the additional business objective of selling premium tickets through the offer of additional luxury and comfort meant eroding and weakening the technical design elements that offered safety, which was the primary business objective. 

The goal of the second characteristic is to have a consistent AI system architecture with known limits and boundaries. The simpler and more easily understood a system, the easier, more reliable, more cost-effective, and the more scalable it is to use. Rather than extend one system beyond its current limits and boundaries, it is significantly more cost-effective and significantly less failure-prone for an enterprise to just create a new one with the desired functionalities and performance characteristics that are complementary to the first system. Doing so usually helps the enterprise to more deeply understand and refine their business case that motivated it, in the first place. It also predisposes the enterprise’s AI strategy to “Agentic AI,” which future newsletter articles will discuss. 

The third successful characteristic of a good AI system is that it is designed to integrate synergistically with the enterprise’s existing processes, technologies, and is intuitive for the roles and skills of the existing workforce. This is because the enterprise’s knowledge is already tacitly captured and maintained by those three entities. If it is designed well, the AI system will not only capture process-specific enterprise knowledge, but also enable it to be reviewed for quality and consistency, improved where there may be small gaps, and reliably recalled and repeatedly executed at arbitrary scale to enable the enterprise to be maximally competitive at maximum efficiency and at an affordable cost. 

Here is an example that can help ground the above abstract benefits in reality. A common challenge faced by U.S. defense contractors is that selling to the DoD is often an on-again-off-again process. Private companies are frequently commissioned to produce a DoD-specific product in very small quantities and in a one-off batch so that their product can be evaluated. Many years pass without any more orders or engagement. Then, seemingly out of the blue sky, they receive an order for a very large quantity of their specialized product. The barrier to entry into business with the DoD is that a company must be big-enough to have a diversified product base and profit margin to absorb the costs of waiting and idling production capabilities. With an implemented AI-driven digital strategy, a much smaller company with orders of magnitude smaller profits can still be a viable competitor in the DoD market. DoD procurement costs would be lower, too, to the benefit of both the DoD and to the tax payer. 

The fourth characteristic is unique to AI systems and usually derives from the third, namely, that if you are able to synergistically integrate your AI system with existing processes, technologies, and personnel, then you have implemented a “small data” AI system. Small data AI systems are usually more cost-effective than big data systems at capturing and maintaining human and/or process-specific knowledge. Small data AI systems do not usually appear in headlines because they are best applied to small and specific problems that excite the imaginations of a few persons, but they are still incredibly useful and very practical. 

Planning and implementing an enterprise-wide AI-driven digital strategy is within every enterprise’s reach. There are ways of doing so that are commensurate with almost any size budget, business objective, market opportunity, timeline, organization, etc. In order to remain relevant and competitive in your enterprise’s business sector, having an AI-driven digital strategy is the best mitigation to almost any type of uncertainty. 

Learn more about Linden AI here.