As everyone who reads my blogs surely knows, I’m trying to get a line on a vision of AI that enterprises believe could really transform their operation. One major challenge is that enterprises are themselves unsure of what the optimal AI solution would look like, and thus often can’t offer me much in the way of suggestions. I now have 52 solid exchanges on AI with enterprise strategists in 48 companies, and I want to summarize what I consider expert views.
A little about the 52 is in order here. Of that group, 41 offered comments on their role or background, and it was surprisingly diverse. The largest group were software architects, then developers with a DevOps role, then team leaders. There were no CIOs in the group, or development managers. Whatever the role, all had experience with AI tools, and 37 had some formal AI training.
Let me start by saying that none of the 52 said that AI had already transformed their business, but 47 said it had a transformational impact on some part of their operation or another. All 52 experts say that AI is potentially a business-transformational technology, meaning that it has the potential, but only 13 of 52 (both 13 experts and 13 enterprises represented) think that the public-LLM-generative-AI stuff that dominates the news is transformational. For that type of AI, there is in fact only one application that the 52 agree is valuable, and that’s the “support chatbot”. Almost every enterprise has pre- and post-sale support missions that have traditionally required human call centers, and about three-quarters have already (pre-AI) gone to some form of automated support. Most have also tried offshore call centers, but as of now that strategy is being questioned because of pushback from callers. Even with a human support agent, though, there’s still the challenge of getting the right answer delivered. The 13 enterprises who believe that generative AI chatbots are transformational are all companies that have large interactive support needs.
The other 39 experts from 35 enterprises tell me that to transform their business, AI would have to integrate into their business processes in multiple places. Just as no single traditional software application alone could transform a business, these 39 say no single AI application could do so. Getting AI into any single point in a business process, it turns out, is more complicated than it seems.
One obvious way to achieve AI introduction is to link AI to individual workers already involved in the target business process. The 39 experts agree that there are places where an AI “assistant” or “copilot” could have a positive impact. They believe this would not be the result of broad empowerment (integrating AI with document development or email for a large body of employees), but by introducing a more specialized AI model to a small number of high-value employees whose work product and effectiveness is highly valuable. In my terms, assistant technology’s value depends on the unit value of labor of the workers being assisted. They believe that some of these assistant missions might benefit from training with public data, but none think that it would require training on data with the breadth of the Internet. A high-value worker typically has a specialist job, and requires specialist assistance.
If you’re not going to rely totally on AI integration via an assistant-to-worker bond, then you have to introduce AI as a component of a business process flow in its own right, which would mean as a part of an application workflow. Enterprises have long recognized the need to pipeline applications (or their components) together in a workflow, and things like the enterprise service bus (ESB) and business process execution language (BPEL) were foundations of the original IBM Service-Oriented Architecture (SOA) designed to do this.
ESB/BPEL is only one example of the broader need to integrate business process elements together in a way that reflects their real-world context in the business. You can also bind them explicitly (each calls its successor and passes the needed data), you can use a publish-and-subscribe event processing approach, you can use a digital twin model, a service mesh…you get the picture.
So, apparently, do IBM and Oracle, the two vendors that enterprise AI experts recognize as playing a positive strategic role in getting AI organized into a business process. Both companies have told their strategic accounts that AI is essentially an application component to be integrated, and that the same rules and policies that guide the orchestration of application workflows have to work for AI. Thus, there may be different approaches taken depending on the specific nature of the business process, the way workflows are currently steered, and the role AI is to play.
Where AI is used for forecasting, modeling, planning, or other analytic missions, enterprises tend to think naturally about having AI integrated into existing analytics tools, which would likely mean one of the explicit or static mechanisms of binding it in. Where AI is to be used in processing business data, commercial transactions, it’s very likely it would simply look like an application component linked in via whatever mechanism was employed (an ESB, a broker, etc.).
Using AI in event-driven applications, which would include both control of process systems and IT or network operations, is an application that all 52 of our experts think would be useful and 50 think could be transformational. It’s also the area where there’s the most variability of viewpoint on the nature of integrating AI into the picture. Industrial process control, transportation systems, and similar real-world control missions seem to 39 of our experts as being specific applications where digital twins should provide for AI integration and context control. For AI and network operations support, the optimum solution (according to 46 of the experts) is more similar to the technique used to integrate AI into business planning. This likely reflects that IT and network operations is visualized as a planning and supervisory task, where industrial process control involves actually involving AI in the work itself. So far, none of the experts is suggesting that AI would actually take control of every aspect of network routing, for example.
While most (46 of 52) AI experts say that their organizations have a handle on the integration of AI with their business processes to the point where significant business improvements could be generated, they all agree that the issue needs to be addressed in a more effective way to impact the broad market. All 46 said that their CIO was integration-literate, but none believed that any other C-suite people in their company were even modestly AI literate, and 49 thought that was a problem because it made framing a project and getting approval more difficult.
Chatting with this group has demonstrated, to me at least, that there’s a big difference between the way that transformational AI is visualized. The majority of enterprise management/executive personnel think of it purely in chat-generative terms, but the people responsible for transformational projects think of it almost as a programming language. This difference, I think, isn’t just a matter of exposure, but surely it’s a result of a longer and more formal assessment of AI value. Are we all headed in that direction? Maybe, but one somewhat demoralizing point is that 51 of the 52 said they believed that vendors were more focused on the hosted-chat model of AI. Said one, “The amateurs are winning.” That would be bad for those who hope for optimal AI adoption any time soon.
