To make cloud or AI projects successful and complete them on time, you need a clear understanding of business goals and technology capabilities, and that understanding needs to be kept current and shared by IT and line-of-business stakeholders.
For the last two years, enterprises have been telling me that the failure rate for cloud and AI projects is roughly three times as high as the failure rate for traditional IT projects. And cloud and AI projects take longer than traditional projects. The higher failure rate likely isn’t a surprise to most of you. But weren’t both cloud computing and AI supposed to accelerate IT?
Enterprises have described three problems that they believe cause cloud and AI projects to take longer, and they also shared suggestions for how that might be fixed. Let’s do a countdown to the No. 1 cause, just like game shows do (multiple responses were accepted).
But first, let’s do a level-set. Cloud has been around a long time, and AI is just getting started. We should expect to see project practices evolving further with the cloud. And we should expect to see enterprises learning project lessons from cloud adoption that could be applied to AI. So far, we’re seeing some of the former but less of the latter.
So, onward to the countdown.
No. 3 problem: Missing essential skills
More than half (56%) of enterprises say they don’t have the essential skills needed on staff, and it takes longer to acquire them than expected. Interestingly, enterprises who report this problem also suggest that somehow it was unexpected, meaning that they thought they were properly staffed already (the most common situation), or they thought the needed skills could be readily acquired.
In both cases, the issue was that they had cloud and AI skills, but not the ones they needed. It turns out that what enterprises need for both cloud and AI is architects, and they had developers or operations specialists. Their projects stalled in the planning phase because they couldn’t plan effectively. They couldn’t plan effectively because they couldn’t frame new technologies into a business context.
No. 2 problem: Unrealistic expectations lead to problematic requirements
Early planning and business case validation show that the requirements set for the project can’t be met, which then requires a period of redefinition before real work can start. This situation – reported by 69% of enterprises – leads to an obvious question: Is it the requirements or the project that’s the problem? Enterprises who cite this issue say it’s the former, and that it’s how the requirements are set that’s usually the cause.
In the case of the cloud, the problem is that senior management thinks that the cloud is always cheaper, that you can always cut costs by moving to the cloud. This is despite the recent stories on “repatriation,” or moving cloud applications back into the data center. In the case of cloud projects, most enterprise IT organizations now understand how to assess a cloud project for cost/benefit, so most of the cases where impossible cost savings are promised are caught in the planning phase.
For AI, both senior management and line department management have high expectations with respect to the technology, and in the latter case may also have some experience with AI in the form of as-a-service generative AI models available online. About a quarter of these proposals quickly run afoul of governance policies because of problems with data security, and half of this group dies at this point. For the remaining proposals, there is a whole set of problems that emerge.
Most enterprises admit that they really don’t understand what AI can do, which obviously makes it hard to frame a realistic AI project. The biggest gap identified is between an AI business goal and a specific path leading to it. One CIO calls the projects offered by user organizations as “invitations to AI fishing trips” because the goal is usually set in business terms (“improve sales/competitive position” or “reduce inventory cost”), and these would actually require a project simply to identify how the stated goal could be achieved. From that, it would be possible to frame an actual project to implement a strategy.
Why doesn’t this happen with traditional technology? According to enterprises, the big reason is that line organizations can experiment with AI, and draw conclusions about its benefit to them, without any IT involvement at all. In the past, with non-AI technologies, line departments tended to work with IT just to learn what could be done. “Early partnership with IT makes a big difference,” one IT professional with AI skills noted.
This particular problem, though, happens a lot less often for enterprise IT leaders who have a strategic vendor partner who has practical AI experience. A “strategic vendor” is usually one that has broad enterprise engagement and credibility. Combine that with AI skills, and you have a combination that can meld business and technology, which overcomes the problem of translating business goals to steps that can be implemented.
No. 1 problem: Questioning the approach while execution is underway
The most common problem, reported by 74% of enterprises, is that during execution, experience with what’s being done causes stakeholders to question the whole approach. What makes this problem different is that the project is well underway before it’s even recognized. Why does it take so long to figure this out? Most interesting is the fact that this group connects the implementation problem with how new technologies like the cloud and AI are seen by IT personnel and business management. This is most obvious with the newest technology revolution, AI.
Enterprise IT naturally thinks of things in the terms of the current IT infrastructure. Companies have spent a boatload of money buying servers, networks and applications and applying them all to the flow of business activity. IT does stuff, in the minds of IT professionals. AI, as line organizations and company executives see it, answers questions. The value of AI is in the answers not in the actions. The difference in perspective is so profound that it takes some direct experience with the project’s results to make it clear. One IT/AI developer, showing a stakeholder an application of AI to a current application that significantly improved the manufacturing cycle’s parts and goods efficiency, asked in a bewildered tone, “Where do I talk to it?”
This sounds a lot like the “autonomous agent” story, but it’s not. Humans are still making the decisions and subordinating AI under the IT view, but more at a policy level. AI’s role is the same as the role of any application component. Not so for line managers. For them, since your application of AI depends on generating an AI response to a question, AI works through you, through workers. This is a logical way for a line manager to think of AI; it’s something that makes workers work better. Both views of AI are logical, but they’re different.
It’s obvious how this difference impacts a project. It seems like the terms we use in AI, like “generative” or “agent” or even “autonomy,” are interpreted by each group of stakeholders in totally different ways, and that this conflict can’t be seen until there’s something concrete to demonstrate. Resolving it, at this point, often involves compromise. About a third of what IT organizations see as workflow-coupled AI projects end up having an interactive component, even if it’s actually just a way of offering management oversight on the state of the work AI is doing.
Since line organizations don’t usually set the goals of cloud projects, the differences between line organizations’ and IT planners’ views on the project aren’t much of a factor. Here, the big problem is simply a lack of proper evaluation of the claimed benefits. Cloud repatriation, for example, seems driven by failing to assess cloud costs, which means a failure to validate the benefits. It’s not caught until actual testing slaps reality in everyone’s face.
Fixing the problems
OK, if all this is bad, how does it get fixed? It seems to come down to a point enterprises made about “strategic vendors.” Do we ask enterprises to pick a strategic vendor for cloud and AI projects? That wouldn’t be practical, given that enterprises who have such a relationship say it can easily take a decade or more to develop. The good news is that what the strategic vendor does could be done in house. To make cloud or AI projects successful and complete them on time, you need an ongoing understanding of business goals and technology capabilities shared by the stakeholders. Strategic vendors can facilitate that, but there are other ways to achieve the goal too.
The enterprises with the greatest number of cloud project successes said they had established a “cloud team.” It’s too early to be certain, but it appears that the enterprises who get the most from AI without the influence of a strategic vendor are establishing an “AI team.” This team is composed of both line management and IT AI experts, but both groups are trained in all the ways AI can work. Both groups have a full picture and can select an AI approach that makes sense. Most importantly, then can assess its business case and drive its adoption cooperatively.
We always realized that AI was forcing us to learn a new language. We need to realize that we have to work hard to ensure that it introduces only one, common, language for business and technology. And, realize we need to learn the language effectively, even within IT.
