AI’s challenge starts with definition. We hear all the time about how AI raises productivity, and many have experienced that themselves. But what, exactly, does “productivity” mean? To the average person, it means they can do things with less effort, which they like, so it generates a lot of favorable AI stories. To a business, though, it means more revenue-generating output per unit of input, such as labor, capital, and materials. It means a business case, which any tech project requires working with the CFO to prove.
AI projects have to prove that they can transform personal productivity feelings into business value. Enterprises tell me that isn’t being done very often with AI because there’s no IT project, so no testing of a business case. Right now, most of AI is driven by the same force that drives “citizens development” things like low- or no-code. Line departments just expense access to a cloud-hosted AI service or tool, and the whole process of justifying costs and proving benefits is skipped.
With AI, you have to use the data that’s available. Citizen AI is almost always divorced from the core business data that defines how a business operates. Without that, how valuable can AI be in making business decisions? Data governance policies rarely allow this data to be hosted in the cloud, much less used to populate a giant AI model, so the cloud AI tools are left with the task of helping with emails and reports, and that’s not likely to transform business operations or make a business case. Enterprises are almost unanimous in saying that creating a real impact on the bottom line means gaining new insights from the core business data.
That’s how it has always been. Core software, the applications that run the business today, have access to all the good-stuff data, which is one good reason why the cloud chatbots that line organizations rely on for productivity can’t replace enterprise software. What about self-hosted AI? Could enterprises run their own AI instead of these core applications?
CIOs think that’s laughable, and the reason goes back to the concept of an IT project and its business case. AI requires different computing, different networking, right? Then all the old stuff has to be tossed. Capital investments are depreciated over a period of three to five years, and so roughly half remains at any given time. Toss it, and you have to write the remaining depreciation off, which means the project’s benefits have to cover not only the cost of the new AI models and hosting but also the write-off of the old stuff. CIOs say that they’d never even propose such a thing because they know it could not possibly get approved.
Suppose that AI can do everything that the current IT tools can. Same output, then, but with a new AI cost plus a write-down of the current infrastructure. There’s no business case there. But suppose AI makes everyone more productive, so the company can get rid of a quarter, or a third, or more, of workers. Now can you get approval? Enterprises tell me it would never happen. Every line manager who was either displaced by AI or whose workforce was eliminated would fight your project. And who are the senior executives who’d decide? Not former IT people. Nike and Reebok make sneakers, not software, so sneaker people will lead them. Anyway, what company would even consider something that radical, and what CIOs would put their careers on the line to suggest it? Show of hands?
Sunk IT costs aren’t the only stake that past projects tie you to. There’s also the issue of practices. How could something cut a big chunk of the workforce without radical changes to how the other workers do their jobs? There’s also the question of risk in this sort of situation. Companies were always uncomfortable with the notion that a single key person had too much knowledge, too big an influence on the business, to be lost. Suppose the key person isn’t a person but an AI entity that no company on earth has long experience with? Who will insure that?
Enterprise IT organizations have their own view of AI agents; they see them as software components. That fits them into project practices, and business-case principles, honed through decades of experience. Lack of those is a big barrier to AI…because it should be. Where AI can be added to a workflow, where it can augment business intelligence, enterprises are happy to host an AI model to fit the new mission and work out a business case like they would for any other software component. Augment, not replace. That’s the reality that AI needs to work with, to be successful.
