Over the last month, I’ve had (and blogged about) many conversations with enterprises on their AI transformation, covering goals, approaches, and project outcomes. One thread that’s kept coming up is the role of vendors in the AI project process, and in particular the concept I’ve called “strategic influence”. This is a measure of the extent to which a vendor can promote a project, set its goals, and guide its progress. Vendors with a high level of strategic influence are not only more likely to win a deal, they’re more likely to be able to convince a buyer to actually do a deal in the first place. With an emerging technology like AI, where buyer literacy is low, this is critical.
To give you an idea of how critical, of 274 AI projects on which I’ve gotten detailed information, 199 were launched by enterprises where the incentive to go forward came in part from a vendor who had significant strategic influence, and 75 where the project were not so motivated. Of the 199, 145 were either completed successfully or considered to be progressing to a successful outcome, 28 were still too early to assess, and 26 were considered not to have met goals. Of the 75, 54 were considered not to have met goals, 11 were too early to assess, and 10 had or were progressing toward a successful outcome. What this shows is that vendor-driven enterprise AI has a way better chance of success.
It’s also significant to note that these numbers refer to AI projects that were controlled by IT, not projects that applied AI-as-a-service and were controlled by and through line organizations. I have no reliable information on this class of project because almost all my contacts are on the IT and network side, but from 32 CFO-organization contacts I did get a little data, which suggests that only about 12% of these “citizen AI” projects succeed.
Of the 199 vendor-influenced projects on the list, 173 were what would be called “AI agent” projects, 7 were self-hosted generative AI projects, and the remaining 19 were IT-coordinated AI service-based. Of the 75 without vendor influence, 18 were “packaged AI” agents in the embedded AI model I’ve discussed, and 10 of these were considered a success, accounting for all the successful projects. Another 12 were agent projects, four of which had failed and eight of which were still in-process. The remainder couldn’t be reliably classified in terms of the AI approach, or for which the approach was in a state of flux.
What seems to be true here is that where AI is influenced by a vendor who has strategic credibility with an enterprise, the project is more likely go get approved, to get off to a strong start, progress quickly, and succeed. Where no such influence is present, the project seems to stall in the early stages, often stumbles in picking an approach, and is more likely not to meet its goals. This seems almost entirely due to lack of a real grasp of AI within the enterprise.
It’s interesting that over half of the enterprises who reported AI project failure believed they had proper AI credentials in-house for the project. However, all the failed-project examples reported that they had spent more time than usual/expected in the planning phase, had faced unexpected complications in execution, and had struggled with dealing with these complications. The ones who did not think they had reasonable in-house expertise had come to that assessment during the project; they were confident at the start. However, about a fifth of this group said their assessment of their AI skills at the onset of the project had included an expectation they’d have to rely on some outside help. That could mean that they found their knowledge of AI insufficient to muster external AI expertise, or enable them to pick good partners.
Does this mean that all you need to do to make AI transformational for your company is pick a “strategic” vendor? I don’t think so, though I can’t validate that view expressly from comments. I think that the real issue is the relationship between an AI-aware vendor and enterprise IT, and that’s complicated.
Generally, enterprises report vendors have strategic influence when the vendor has a dedicated account team who is either on-site regularly or meets regularly with enterprise IT and CxO types. That vendor is then likely aware of the way the company uses IT and can then fit AI into business processes by leveraging the way applications are already fit. This encourages the “workflow” model of AI, where an AI agent is like a software component, integrated into current operations at the IT level in a familiar way, and working to support line departments in a way similar to how IT has supported those departments in the past. It’s a familiar model at both levels, and thus more likely to be understood by all the stakeholders.
When this happy combination doesn’t exist because no vendor has real strategic involvement, it’s usually either that enterprise IT doesn’t really exist (SMBs may not have such a thing), or that it’s relatively small and perhaps really doing almost nothing but sustaining current applications. In this situation, even if a vendor with AI expertise is brought in, the vendor may not have, or take, the time needed to build a real understanding of the business, and thus can’t effectively help with mapping AI capabilities to business problems or opportunities. Often specialized AI providers simply don’t want to get into what they see as the dreaded “educational sale”, which takes a lot time, a lot of their resources, and often then results in the more-aware buyer now looking for competitive bids that their teacher-vendor may lose.
I think there’s a point to be made here that goes way beyond AI, several in fact. First, are real advances in applying technology to business applications getting too complicated for companies to assimilate themselves? If so, there may be a growing gap between what we could do to advance IT spending, and what we will do. Second, has this been a partial reason why new projects have, over the last two-and-a-half decades, contributed a shrinking amount to IT budgets? If so, then it may be that this has reduced the number of “architects” in enterprises, people capable of designing new things. Finally, is this not a vicious circle; because that decline in architect skills would then make it harder to advance new projects? Scary thought.
