peshkova - Fotolia
For years, promoters have made AI technologies sound like the all-encompassing technology answer for enterprises, a ready-to-use piece of software that could solve all of an organization's data and workflow problems with minimal effort.
While AI can certainly automate parts of a workflow and save employees and organizations time and money, it rarely requires no work or no integration to set up, something organizations are still struggling to understand.
In this Q&A ahead of the publication of his new book, Alan Pelz-Sharpe, founder of market advisory and research firm Deep Analysis, describes enterprises' AI expectations and helps distinguish between realistic AI goals and expectations and the AI hype.
An AI project is different from a traditional IT project, Pelz-Sharpe said, and organizations should treat it as such.
"One of the reasons so many projects fail is people do not know that," he said.
In this Q&A, Pelz-Sharpe, who is also a part-time voice and film actor, talks about AI expectations, deploying AI, and the realities of the technology.
Have you found that business users and enterprises have an accurate description of AI?
Alan Pelz-Sharpe: No. It's a straightforward no. I'll give you real world examples.
A very large, very well-known financial services company brought in the biggest vendor. They spent six and a half million dollars. Four months later, they fired the vendor, because they had nothing to show for it. They talked to me and it was heartbreaking, because I wanted to say to them, 'Why? Why did you ever engage with them?'
It wasn't because they were bad people engaged in this, but because they had very specific sets of circumstances and really, really specific requirements. I said to them, 'You're never going to buy this off the shelf, it doesn't exist. You're going to have to develop this yourself.' Now, that's what they're doing, and they're spending a lot less money and having a lot more success.
AI is being so overhyped; your heart goes out to buyers, because they don't know who to believe. In some cases, they could save a fortune, go to some small startup [that] could, frankly, give them the product and get the job done. They don't know that.
Are these cases of enterprises' having the wrong AI expectations and not knowing what they want, or they are cases of a vendor misleading a buyer?
It's absolutely both. Vendors have overhyped and oversold. Then the perception is I buy this tool, I plug it in, and it does its magic. It just isn't like that ever. It's never like that. That's nonsense. So, the vendors are definitely guilty of that, but when haven't they been?
From the buyer's perspective, I think there are two things really. One, they don't know. They don't understand, they haven't been told that they have to run this project very differently from an IT project. It's not a traditional IT project in which you scope it out, decide what you're going to use, test it and then it goes live. AI isn't like that. It's a lifetime investment. You're never going to completely leave it alone, you're always going to be involved with it.
Technically, there's a perception that bigger and more powerful is better. Well, is it? If you're trying to automatically classify statements versus purchases versus invoices, the usual back office paper stuff, why do you need the most powerful product? Why not, instead, just buy something simple, something that's designed to do exactly that?
Often, buyers get themselves into deep waters. They buy a big Mack Truck to do what a small tricycle could do. That's actually really common. Most real-world business use cases are surprisingly narrow and targeted.
Editor's note: This interview has been edited for clarity and conciseness.